Mathematical Optimization of Food Sorting and Grading Systems for Improved Efficiency - Madhavan Kadiramangalam, Shamita Sherry, Rohith S., Sutharsan S., Karan Sinkha Sonaji ABSTRACT The following paper discusses and details techniques and methods in the food sector for grading and sorting. The quality attributes of food have been an important aspect in the production of value-added products. The raw material must be consistent, uniform in nature in terms of colour, figurative shape and general appeal of the shape, large and equivalent sizes with maximum ‘flesh’ content value as well as a defect-less appearance. Most grading systems today use mathematical models that depend on artificial intelligence, neural network systems, colorimetric reading algorithms and mathematical models that take into study the hundreds of variable parameters that define our food at a basic level which create ‘hype’ and demand in the consumer market. The objective of this paper is to highlight these mathematical techniques and their applications in the food industry. CHAPTER 1 Madhavan Kadiramangalam I. Introduction Food grading and food sorting are age old practices used for scanning, identifying and changing/replacing samples that do not match a fixed set of parameters which attest to high visual and organoleptic quality. These qualities vary and depend on multiple independent variables like colour, shape, size, mutative deformities and possibly flavour, sweetness and other bio-parametric indices. e.g., Figure 1: Mango considered unfit for processing, ‘Over-ripened/Decayed’ Illustration 1: Mango considered fit for processing, 'ripe' Illustration 1; depicts a mango considered ‘ripe’ and fit for processing or for commercial purposes which could be identified and categorised by creating a cluster chart for chemical, physical and visual characteristics of the mango like TSS, colorimetric analysis, peel firmness, stiffness, pulp firmness, acidity etc. (1) This is an example of mathematical application for standardising independent variables into a graphical medium, creating a broad spectrum of analysis to identify the highest quality of a food sample in a continuous system for sorting and commercial purposes. The grading of food quality can be easily categorised on a high-speed conveyor line for quick sorting by identifying physical and visual defects rather than chemical defects. On a larger scale, the probabilistic chances of a given food sample to be simultaneously visually pleasing and chemically, pathogenically defective is extremely low, this extrapolation can be considered naturally and logically. The parameters to discuss are, colour (converted to chromaticity), size, shape, defect detection (2). II. Colour as a parameter Colour in food products, vegetables and other consumptive media is possibly the most important indicator of quality, commercial consumer appeal, defects, extent of ripening. Despite the available sorting and grading techniques, colour is the most sensitive and vast subject. A simple method for using colour in computer vision grading and sorting is to convert the pictures taken into raw chromaticity pictures using the formulae; Equation 1: Raw colorimetric RGB data conversions to Chromaticity and Colour Values (2) π ∑π πΊ πΊπ = ∑π π π = π΅π = π΅/(∑ π ) πΆ = (π − πΊ)/(π + πΊ) We convert our raw data from RGB (Red, Green, Blue) values to chromaticity values. This disregards any form of luminescence and gives us raw output data which does not depend on the L value of the given sample space. This conversion allows us to isolate a given region and train an algorithm to consistently recognise a colorimetric value. The average colour value of an orange in hex is given by #FFA500 but R G B typically holds around the values of 100%, 64.7%, 0% respectively. This means we can remove our image background and create a general colouration index for further grading and sorting of the oranges. This method of colour isolation is also used to identify the differences between healthy parts of a given food sample by comparing it to the algorithmic standardised index of colour spectra we have used for identifying the biological variation in natural and healthy colours/ mix of colours. The defects usually come in the form of a discolouration in different bio-variations. The defects could be black/brown, lack of natural and healthy colour like greenish unripe like tinges and hues in oranges. Extremely deep colouration beyond our fixed spectra identifies over-ripening and potential internal spoilage(Kondo, 2010). i) Defect identification by colorimetric identification Optimisation of defect detection in food grading is done using the concept and mathematical application of CNN’s (Convolutional Neural Networks) which depend on training data and thousands of iterations to detect variation in our samples. A keen example for the use of neural networks in defect classification is the categorisation and identification of defects in potato sorting and grading. Colour pixel classification is the technique used here to classify change and variation in our variable parameters. Illustration 2: Sorting of potatoes into two grade categories(3) Illustration 2 represents the core idea behind classification of defects based on pixel analysis. The method to apply this uses a variety of neural network concepts, mathematical vector support machines. Vector support machines are mathematical functions that maximise a specific mathematical function based on collected data to give a relevant output to the trained model. CNNs use collected and trained data further supported by SVM maximisation functions to give an optimised output. The idea here is to use an SVM to maximise and optimise the classification into two categories we use to decide quality. Either defected or healthy. The decision equation for an SVM can represented best by; Equation 2: SVM optimisation function(3) π π¦ = π ππ(∑ π¦π ∝π πΎ(π₯, π₯π ) + π) π=1 x is a dimensional vector sample, xi is the sample space for our ith samples, N is total spaces, K is our kernel function while α and b are our training parameters in which this case would be our colorimetric defects that govern the maximality optimisation of our function set. SVM uses statistical learning as well as optimisation techniques and practices to improve the quality of output in its general working sample to give a suitable final result. Pixel quality detection can be further improved by using SMO’s (Sequential Minimal Optimization) to iterate and parse pixel defect values which then monochromatize our product giving a surface match to defects (4) Illustration 3: Surface mapping of pixel defect variation(4) The issues with this technique are that it limits our classification to primarily binary classification which parses as either a 0, defect, 1, healthy. It may also classify very minute quality control issues as a defective quality form. It can easily identify genuine rot or quality defects but false positives are common. The pixel density storage of the given technique limits classification to 448x336 pixels of study (3). Colorimetric defects can also give clues to physical defects and lack of quality in skin of many fruits and vegetables. An example of this is studying the textural effects of ‘gloss’ in monochromatic sorting systems like in the sorting of brinjal Solanum melongena L. The colorimetric technique in the sorting system for brinjal involves removal of image noise using gaussian noise filtering function by parsing pixel data through the function to identify strange and unexpected variations that come as a result of poor image handling and formation of changes in brightness, RGB values and other colour related parameters. The filter identifies the mean deviation in a pixel from the surrounding pixels and iterates the value to match the pixels, smoothing out a uniform cluster of pixels rather than creating any changes in entirely different pixel colour values. The value is then adjusted using kernels with respect on the aggregate relation. The form of gauss functional filtering used is; Equation 3: Gaussian noise filtration function(5) πΊ(π₯) = 1 √2ππ 2 π₯2 ( 2) π 2π Where, π = √∑ (ππ − πΜ )2 /(π − 1) π G(x) is our gaussian filtering function which takes the standard deviation of the input pixel values to parse through and create final distinction pixel colour values to smooth out and fix the noise in an image. Xi is the specific ith value in a set of data values {a1, a2, …. An} πΜ represents our mean data set value. n is simply our aggregate set of values to have parsed for our final function. Finally, the images are processed in a black and white monochromatic light wave(5). The brinjal is then pixel scanned from 4 different mutually perpendicular angles under fixed lighting conditions to maintain uniform colouration and to create consistent concentration spots for reflective surfaces. Illustration 5: Spectral variation in gloss characteristics of brinjal(2) From Illustration 5 we can easily visualise the stark difference in gloss in both the brinjals. Figure A displays the rotational imaging of a brinjal with consistent and noticeably sharp visual gloss which is further sharpened with application of gaussian noise filtering. In Figure B, Illustration 4, we notice the brinjal creates a dull after shine of the gloss we expect from a lower surface texture quality. We can easily extrapolate differences in textural surface characteristics of both brinjals where brinjal A has a much smoother texture with a more perfect condition of ripening and setting while brinjal B can be assumed to have a slightly soggier textural structure that deduces either over-ripened surface textural characteristics or possibly even decay and internal damage (2). Another major technique utilised in defect detection is the usage of various wavelengths of electromagnetic radiation (EMWs). EMWs have different frequencies, energy levels and consequently, varying absorption frames. Most biological defects like the affectation of moulds, medfly egg implantations, citrus surface scarring are easily given a sharp contrast which can then run through a pixel defining to properly define and categorise levels of damage based on a simple fuzzy logic model like serious (mould, medfly eggs), slight (scarring, discolouration). The various EMWs used in (6) as reference are; 450750nm (visible light), 700-1000nm (Infrared), 380-710nm (Fluorescent) and finally 100-400nm (Ultra violet radiation). The extent of absorption of these forms of radiation varies across biological surfaces hence when we compare the same mould across different light spectra, we conclude the pixel density variation per surface area of a two-dimensional image can be given a huge contrast relative to the change of radiation used. Illustration 6: colorimetric readings of orange in (a) Visible spectrum (b) infrared (c) fluorescence (d) ultraviolet (6) Illustration 6 depicts a ‘green’ mould typically found on citrus fruits seen under a variation of radiations. We can clearly see and extrapolate that infrared and fluorescence give a dead obvious contrast compared to visible light and ultraviolet and hence will be used to categorise and classify the quality. We can utilise a simple algorithm that takes weights of each defect we wish to categorise and create a simple function that may categorise by giving weight to each defect noticed by computer vision analysis. We can deepen the contrast with simple image contrasting algorithms Illustration 9: Green mould under contrasted fluorescence (6) Illustration 8: Ultraviolet reading of stem-injury(6) Illustration 7: Infrared reading of anthracnose(6) We can create a simplistic classification score to go along with our support vector machines in our computer vision classification as; Equation 4: Colorimetric weighted pixel defect scoring algorithm(6) ππ = π€ππ» × π» + π€ππ × π + π€ππΌ × πΌ + ππ Here, Zj represents the score for any class of defect we wish to study for j = {1,2 …… n} here the values for j are all the defects we wish to grade and categorise. In Blasco. J et. Al 2007, 12 defect readings were used so his integral variable limited itself from 1 to 12. H, S and I are coordinates of any mean colour sample we study in our readings. w is the weight of each sample relative to its spectral intensity and finally Cj which is simply our sample constant which varies on chromatic parameters like noise, angular variation etc. III. Size and shape as a core parameter Most veggies in the contemporary sale are sold due to their cost is to size ratio as well as their relative structural appeal. Here we note, shape and size seem to be the parameters to maximise in grading and sorting. A uniformly ‘large’ size generates an idea uniformity in quality as well as flavour in an FMCG market. Ziaratban et al. 2017 talks about various techniques utilised to model and measure quantities like volume, surface area, and shape. Golden apples from a local market were studied by simply using RGB input values from a camera from where the background was isolated and removed, the image was binarized and finally the major and minor diameters of the apples were generated. The volumes of the apples were studied by using a very primitive technique of where volume displacement in water is used to find the absolute volume. The fact is that apples have a density lower than water at 0.75 kg/m3 which creates unwanted floatation behaviour in the apples. This meant that an extremely thin and light weight wire had to be utilised for forced submergence and displacement. The technique used here to study the mathematical relation between size, shape, volume and surface area of our apple is done using an Artificial Neural Network (ANN) (7). Oke. M. O. 2017 talks about the application of ANN mathematical systems to read, grade and products of agricultural produce. There is an extreme variation in biological products due to unexpected mutative qualities and variation in nutritive quality of produce. The reason trained artificial neural network models are so help is because the training creates a method to ignore extreme bio-variation like how the human brain does, so naturally, creating a more general perception of quality which can be continuously trained and improve to create a near perfect grader-cum-sorter. Illustration 10: A multi-layer perceptron(8) Illustration 10 depicts the mathematical format of a many layers’ perception megatron-based artificial neutral idea network model which has multiple inputs, hide and seek layers based on fuzzy logic and relevant outputs to the model we are attempting to train. Parameters are fixed and are used to train around the system around the model to give us a satisfactory output. Another model we can use for automatic retraining in our ANNs are as follows; Illustration 11: A recurrent network structure (8) These neural networks are key technological advances in our ability to categorise complex forms of food products and agricultural produce into a large variety of quality categories. We can use convolutional layers for our grading requirements and size and shape isolation(8). We can make use of our length based diametric measured quantities to create a simple projection of size. Every product has some sense of two-dimensional variation in terms of ‘Degree of curvature’. This can easily be exploited in programs like MATLAB. An example is given below; Illustration 12: Elliptical classification of a potato into major and minor diameters(3) The reason this is so important is because a simple detection algorithm can be used to mention something like, ‘a potato may only have uniform and singularly concentrated rounded shapes. What this essentially means is that running the shapes through a neural network can create an idea of what a general product must look like in general for it to look appealing. We can simply monochromatize, binarize and finally run ratios of a large detected uniform shape relative to any protrusions, angular deepening etc. and create a ‘defect ratio’ as shown below; Illustration 13: (a) Binarized image (b) segmented and measured image of potato(3) Note how the protrusion in the potato was segmented as a defect because the majority body did not move smoothly with it, but rather as an unusual protrusion due to our simple sorting algorithms that we used earlier. Illustration 14: Elliptical size detection in two different defective potatoes(3) Notice how we have to deal with different shapes which creates the need for image resizing and pre-processing. Pixel density is greatly affected when resizing images for colorimetric processing as well as size processing which may result in a different expected surface area when compared to literally measured surface areas(3). Zheng et al. 2021 discusses the use of convolutional neural networks in dealing with surface feature extraction while simultaneously grading mangos on size and shape. He uses the technique of pre-processing as mentioned. Illustration 15: Pre-processing of mangos with resizing and colorimetric isolation (9) Notice the feature extraction only occurs after zoom processing of the image taken of the mango. He grades his mangos into 3 primary categories. A) First Grade: - Mango shape isn’t deformed and size is there. The surface is nice and smooth, no defective seems at appearance, with at max, only two spots it may seem, and the diametric measure the black is at max 2 millimetre. B) Duodonous Grade: - shape does not have deformation. Colour is normal and 75% or greater of the surface is fine. The surface characteristic is smooth, with 4 spots at max, and the diametric measure is onto three mils. C) Tert Grade: - bad qualities of mango does not attribute to mango product grading. Colour of the fruit is good and 40% surface is slick. The form is mostly trim, with 6 spots on each fruit at max, and the big shape of the spot is three point four five millimetres.(9) IV. Conclusion The FMCG market demands quick, easy solutions to optimise and create high quality, import level products at any consumer level with the efficiency of a robot. We find that sorting and grading of agricultural produce is a major branch that, with automation, could possibly entirely eliminate heterogeneity in product development and manufacturing. A consistent quality of say mangoes, potatoes, dragon fruit, strawberries, citrus fruit contribute to high-quality value-added products. Mathematical models, neural networks, Convolution neural networks, sorting and grading algorithms, technology like cameras, electromagnetic spectrum based radiative machines to create specific wavelengths of radiation for quality evaluation are some of the techniques used to sort and grade food. The mathematical optimisation of these techniques has been discussed in this paper as well as the parameters themselves that create our problems as well as being the reason for ability to sort with high efficiency. The application of mathematics has been very obviously highlighted in this paper as a means of generating improvements in the existing technologies. REFERENCES AND CITATIONS 1. Nambi VE, Thangavel K, Jesudas DM. Scientific classification of ripening period and development of colour grade chart for Indian mangoes (Mangifera indica L.) using multivariate cluster analysis. Sci Hortic. 2015 Sep 2;193:90–8. 2. Kondo N. Automation on fruit and vegetable grading system and food traceability. Trends Food Sci Technol. 2010 Mar;21(3):145–52. 3. Razmjooy N, Mousavi BS, Soleymani F. A real-time mathematical computer method for potato inspection using machine vision. Computers and Mathematics with Applications. 2012 Jan;63(1):268–79. 4. Ali SSE, Dildar SA. An Efficient Quality Inspection of Food Products Using Neural Network Classification. Journal of Intelligent Systems. 2020 Jan 1;29(1):1425–40. 5. Hemamalini V, Rajarajeswari S, Nachiyappan S, Sambath M, Devi T, Singh BK, et al. Food Quality Inspection and Grading Using Efficient Image Segmentation and Machine Learning-Based System. J Food Qual. 2022;2022. 6. Blasco J, Aleixos N, Gómez J, Moltó E. Citrus sorting by identification of the most common defects using multispectral computer vision. J Food Eng. 2007 Dec;83(3):384–93. 7. Ziaratban A, Azadbakht M, Ghasemnezhad A. Modeling of volume and surface area of apple from their geometric characteristics and artificial neural network. Int J Food Prop. 2017 Apr 3;20(4):762–8. 8. B HJ. APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS AND GENETIC ALGORITHMS IN DRYING OF FRUITS AND VEGETABLES : A REVIEW. Vol. 11, LAUTECH Journal of Engineering and Technology. 9. Zheng B, Huang T. Mango Grading System Based on Optimized Convolutional Neural Network. Math Probl Eng. 2021;2021. 10. Bhargava A, Bansal A. Fruits and vegetables quality evaluation using computer vision: A review. Vol. 33, Journal of King Saud University - Computer and Information Sciences. King Saud bin Abdulaziz University; 2021. p. 243–57. 11. Patil PU, Lande SB, Nagalkar VJ, Nikam SB, Wakchaure GC. Grading and sorting technique of dragon fruits using machine learning algorithms. Vol. 4, Journal of Agriculture and Food Research. Elsevier B.V.; 2021. CHAPTER 2 Rohith S. VEGETABLE QUALITATIVE SCANNING THROUGH MACHINE VISION INTRODUCTION Sensory, visual descriptors in the literature, as well as a quality assessment of veggies based on their characteristics and as well as any disease kind that may be present. Additionally, the fundamental concepts, theories, and analysis and processing techniques that match and arbitrary outcomes. These checks and the expiry help to set market values. The quality check was carried out by skilled human investigators by touching. viewing, too. This approach is very erratic, capricious, and seldom yields consistent results among investigators. Cam notice checks are ideally suited with older type checks and good for checking in this sort of scenario since it is a continuous activity to analyse fruits and vegetables for various aspect criteria. Computer vision systems and image processing are a rapidly expanding field of study in agriculture that are important analysing methods for crops before and after harvest. To the best of our knowledge, pix display the numerical of searching publishing houses publishes on the year by year. photographic pictures However, it is difficult theoretically to quantify or analyse photographic data. As a result, digital image processing technology aids in the processing of pictures and tries to extend analysis of them. The area of agriculture uses image processing for a variety of purposes, such as land identification. the assessment of N2 seeing plants Tewari et al. in 2013. the checking of bug-bed regions Krishna and Habert in 2013 again, the automated categorization and checking of green boy diseases using characteristics by Patel and Kumar in 2011, As computer vision rapidly advances, proven methods for evaluating the safety and quality of many agricultural applications include image processing and pattern recognition. When evaluating the quality of fruits and vegetables electronically, computer vision technology replicates the effects of human vision. Information is provided to the quality grading and sorting mechanism. There have been a number of research articles published on a range of subjects Naik and Patel in 2017 Dubey and Jalal, 2015 and Zhang et al., 2014, some of which focus on certain fruits in quality assessments and others on particular approaches The quality assessment of fruits and vegetables is not fully summarised. The objective of this review paper is to give a thorough overview of computer vision and image processing methods applied in the food industry. The different segmentation, picture features, and image descriptors discussed in the literature will also be examined. and there are numerous techniques available to assess the grade quality of veggies and grains. INSPECTION OF THE QUALITY OF VEGGIES getting of images, preparation of images, segmentation of images, 2.1. Image capture Cameras, ultrasounds, MRIs, electrical tomography, and computed tomato graphy (CT) are just a few examples of the image capture technologies used in food uses. Lighting, an image capturing board (digitizer or frame grabber), big noggin computer parts, and charged coupled device (CCD) are the five fundamental components of a conventional computer vision system (Fig. 3). For the purpose of studying fruits and vegetables, the front and back lighting systems have been set up. Front lighting is intended to examine the colour, texture, and skin defects that are surface quality attributes. However, back illumination is used to assess border quality elements including size and form. Back lighting for shapes is specified. Computer vision system 2.1.1 The first conventional computer vision systems were created in the late 1960s, and today the industrial and aerospace industries heavily rely on them. autoamtic, saftey checks, intelligent transportation systems, medical picturing applications in the military, bort guiding, self-driving cars, and food quality and safety inspections are a few examples. The primary colours in traditional computer vision systems are red, green, and blue (RGB), hence images taken with RGB colour cameras are centred at the Red Green Blue spectrum. Using computer vision systems, a variety of features, including texture, form, colour, size, and flaws, can be automatically assessed and rated. However, because of how similar in texture and colour they are to skin, some flaws can be challenging to spot. Spectroscopic and imaging techniques are combined by the hyperspectral computer vision system (Lorente et al., 2012) to provide spectrum information. In a two-dimensional image, each pixel is grouped together. The most frequent methods for creating a hyperspectral picture cube are point scanning, line scanning, and area scanning. which are shown in Figure 4. The monochromatic images of two or more wavebands are used in multispectral computer vision systems. 2.1.2. Infrared and ultrasound Initially, ultrasonography was used to gauge the amount of moisture in food products; draw conclusions about the inherent fat of cattle; and assessing and researching the orange peel's moisture and turgidity. Longer wavelengths are utilised for image capture when ultrasound and computer vision systems fail to provide the required image. The term "thermographic photography" refers to a method of creating infrared light, which is in the range of 700 to 1000 nm. The idea behind thermographic imaging is that all things release a certain amount of thermal radiation in proportion to their temperature. Tomographic imaging, 2.1.3 Tomography is utilised when researchers need to examine a sample's inside. It has significant nuclear energy. attribute for depicting the two-dimensional distribution of an item is generated from a series of one-dimensional projections by scanning various data sources. This procedure is applied across a wide range of viewing angles until the necessary big of all big data is realised. The applicano have progressed to monitoring food preparation, storage, packing, and distribution in real time. The following characteristics are present in images obtained by various researchers utilising computer vision systems, as indicated in Table 1. Processing 2.2. Images captured using various technologies contain a variety of noises that degrade image quality. As a result, it cannot provide relevant big for the processing. Before process improves the picture knowledge and eliminates reluctance. Preprocessing improves the picture data by overcoming unwanted distortions, enlarging image characteristics required to improve the (poor condition) source image through processing. original for a certain objective. Image preparation techniques for assessing food quality include pixel pretreatment and local preprocessing. Pixel preprocessing converts images from input sources into output images by "correlating each iterative pixels to the out-integer input pixel having the very same coordinating index." The term "colour big boy transformers” refers to the most used pixel pre-processing method for assessment food qualities beautifully conveys texture of the flesh in a monochrome picture. Local pre-processing (Filtration) creates a new brightness value by using a tiny area around a pixel in an input picture. the final picture. It employs a Modified unsharp filters, median filters, basic noise-reduction filters, and detect egg breaks Segmentation (2.3) picture segmentation, which divides Preprocessing is followed by segmentation of a digital picture. The primary goal divides the background. so that the significant region may be processed during the object assessment. For image analysis to advance, segmentation must be done correctly; otherwise, the classifier's performance would suffer. It has several uses, including in the agricultural and medicinal fields (Christ and Parvathi, 2012; Payne and al. (2013); Deepa al Galakshmi (2012)). (Brower et al., 2010; Mahub, 2011; Grge and associates, 2013) Clustering and thresholding are two common segmentation approaches. The simplicity and consistency of the thresholding approach are what make it popular. The thresholding process divides the digital picture into several parts depending on the image's grey levels. The Otsu approach (Otsu, 1979) gives a grey level histogram from the grayscale image in order to choose the best threshold image. Otsu technique, which has several applications, offers certain benefits including optimising It is not known what the original image was before threshold value and grey level image processing was applied. The disadvantage of this method is that it takes longer to establish the appropriate threshold value as the number of clusters increases. To save processing time, the clustering methodology is employed rather than thresholding-based methods. A clustering methodology, which may be divided Partition-based and hierarchical strategies are used to form the cluster. comparable pixel properties. The previous approach is based on tree topologies, where the clusters and database as a whole are represented by the root. However, the latter method benefits from optimisation. The two basic types of clustering are hard clustering and soft clustering. A straightforward method known as "hard clustering" separates a picture into portions based on the pixels that are a part of linked clusters. Hard clustering is illustrated with one k-means. Each pixel is given the nearest centre once the distance from the centre is measured. The intra- and inter-clustering are maximised and minimised, respectively, through hard clustering. Although means is simple and distinctive, the initial cluster centre has a significant impact on the algorithm's success. Feature extraction (2.4) Features are calculated for further analysis after picture segmentation. These characteristics are the fundamental components of mostly due to the fact that they provide beneficial information for object categorization, picture interpretation, and perception. Features are extracted throughout this method. create classifiable feature vectors to identify an aspect. These feature vectors accurately and uniquely define the object's shape. The goal of feature extraction is to increase recognition rates by feature extraction. These characteristics provide the explicit data needed for quality assessment and analysis in the food sector. To assess fruits and veggies' flaws and ripeness, colour, texture, and morphological aspects are frequently utilised. 2.4.1. Features of colour Colour is one of the variables that customers use to decide whether to accept or reject fruits and vegetables. It is a method of measuring indirectly According to Pathare et al. (2013), quality qualities including freshness, attractiveness and variety, maturity, and safety are determined During the stages of development, maturation, and postharvest handling and processing, numerous internal biochemical changes, microbial alterations, as well as physical and chemical changes all take place. The first and most common visual characteristic utilised in photo retrieval and index is the colour feet. The colour feature offers several advantages, such as high efficiency, ease of removing colour information from photographs, independence from gang angular, effectiveness in displaying vision stuff of pix, resistance to backdrop issues, also effectiveness at separating pix to one another. The three different colour models are HSI, CIELab, and RGB. Morphological characteristics 2.4.2 Size and shape-related morphological characteristics are widely employed to categorise fruits and vegetables. In the agriculture sector, Grades for fruits and vegetables are established by to distinct biggie groups throughout processing stages since fruit and vegetable size affects pricing. Comparing the While the diameters of vegena are spherical or almost spherical, the proportions of complicated cuisines are by their very nature unequal, examination is relatively simple. Using the predicted Dimensions, Principal and Minor Axes, Area, Perimeter, and Dimensions one may calculate the size of the feature. These attributes are frequently utilised for automated sorting objectives. The area (scalar amount) determines how many real pixels are present in the area. The region's pixels capture the projected area. The extraction of features is based on the separation between two adjacent pixels. It is crucial to rapidly memorise the angle at which long and width are computed since the forms of foody frequently vary as they are being made. The item's longest line, which depicts the distance between each border pixel pair, is the first six. The minor axis is the longest line drawn across the item that is also perpendicular to the main axis. Since it is challenging to quantify how similar different shapes are to one another, shape is a crucial yet inaccurate visual feature for representing the content of a picture. Shape descriptors come in two varieties: regional descriptors, which are based on an object's integration region, and control-based shape descriptors, which use copied features to separate boundaries. As measures for form qualities, take into account round, tiny, and asphyxiation ratio. Maximum length, breath, and diameter are used by Kondo (2009). REFERNCE 1. Arakeria, M.P., Lakshmana, 2016, Computer Vision Based Fruit Grading System for Quality Evaluation of Tomato in Agriculture industry. In: International Conference on Communication, Computing and Virtualization, 426-433. 2. Ashok, V., Vinod, D.S., 2014, Automatic Quality Evaluation of Fruits Using Probabilistic Neural Network Approach. In: International Conference on Contemporary Computing and Informatics (IC3I) IEEE, 308–311. 3. Pereira, L.F.S, Jr, S., B., Valous, N.A., Barbin, D.F., 2018, Predicting the ripening of papaya fruit with digital imaging and random forests. Computers and Electronics in Agriculture, 145, 76–82. 4. Riyadi, S., Rahni, A.A.A., Mustafa, M.M., Hussain, A., 2007, Shape characteristics analysis for papaya size classification in Research and development, 1–5. 5. Naik, S., Patel, B., 2017b, Thermal imaging with fuzzy classifier for maturity and size based non-destructive Mango (Mangifera Indica L.) grading, International Conference on Emerging Trends & Innovation in ICT, 15–20. Nagata, M., Tallada, J.G., Kobayashi, T., 2006. Bruise detection using NIR CHAPTER 3 Shamita Sherry APPLICATION 1: 1.1 INTRODUCTION Based on the study and observation being conducted on the paper “Evaluation Different Sorting Criteria and Strategies Using Mathematical Programming” Kjærsgaard, Niels Christian Publication date: 2008 Document Version Publisher's PDF, also known as Version of record the mathematical application has been discussed here: According to the Danish Meat Association (2007), the pig industry is a vital and significant sector of the Danish economy, with exports totalling more than DKK 28 billion in 2006. The market for conventional Danish exports is becoming more competitive, and prospective new competitors from low-cost nations should be anticipated. The production, meat, and delivery of Danish pigs must therefore be optimised in every way possible. The pigs that are utilised as the meats houses' raw material vary greatly in size, weight, layer of fat, and other qualitative attributes. The dead areas check this diversity when they do be grouping sewer into various sorting gangs, greatly reducing variation within each group. selecting the sorting boundaries and criteria. 1.2 THE MODEL The Mixed Integer Programming (MIP) model utilised here is the same as the MIP model presented in the writing Kjaersgaard , regarding limits in the production and stock, which is basic on the moo produced on the study with respecting of enhanced scaling Kjrsgaard. Two optimizations—one under the existing conditions and the other under improved conditions—are performed to determine the costs. The difference between the two optimisations' earnings can then be used to calculate the improvement. 1.3 THE DESCRIPTION OF THESE MODEL The meat houses have a lot of logistical issues [Kjaersgaard, N. (2008c)]. The equalising room, where the ambient temperature is balanced throughout the entire carcass, is the most crucial. The carcasses are arranged in the equalisation room of the abattoir that served as the foundation for the modelling on bars that hold 80 carcasses each. Every bar must be emptied from the side that faces the filling side. The same order of production (i.e., the same package of goods) is theoretically utilised for carcasses that are set on the same bar. 43k+ sewer was slain out of the dead areas, and those data were used in the tests. The recorded thick layer and true dead weight are used for sewer. The recorded thick is taken into account in the calculations as the actual value. A generated scaling error is compiled to the recorded thick and used as the measured thick when doing calculations for enhanced measurements and sorting. The model is based on many additional applications for pigs. Each alternate usage includes a "package" of goods tailored to that part. There are two additional uses for the back, two for the eat and one for the poo, there are four additional uses for each in total. Depending on the thickness of the thick, this is assumed into the wen sun when big or small the cost for various items. This is accomplished in the model by dividing the pricing into two distinct contributions: A predetermined price per kilogramme of material for the product in question and a cost xyz that specifies how much the costs will fall if the thick thickens on one millimetre are both present. There are a few special requirements that must be met for agricultural produce to be used for two of the food: • The ham item Only when the thick layer is less than 14 mils and of breast product may be produced if it does not exceed 4 kg. 1.4 THE MATHEMATICAL FORMATION We have a collection of carcasses represented by the set I={1,...,I}. Each carcass can be used to create various product options, denoted by the set N={1,...,N}. Each product alternative for a carcass consists of a range of different products, represented by the set J={1,...J}. The carcasses are then hung on bars in the equalization room, which are represented by the set K={1,...,K}. To determine the optimal utilization of the carcasses on each bar, we introduce the decision variable y,k,n. This binary variable takes the value of 1 if the pigs placed on bar k are being used to produce product alternative n, and 0 otherwise. The main objective is to find the best solution that maximizes the utilization of carcasses and identifies the product alternatives for each bar. Additionally, we aim to calculate the total profit achieved by this optimal solution. The goal function (1) determines the optimal alternative use for each bar when all pigs placed on the same bar are used for the same product alternative in order to maximise the sum of the worth of carcasses at each bar. The restriction (2) governs where the carcasses are put. Every bar is only ever used once. The model makes use of a variety of parameters. The two most significant ones, which are either directly or indirectly used in the goal function, are illustrated in numbers (4) and (5). The ValuePig(i,n) (4) parameter determines the value of each carcass i used to create product alternative n. Based on a price per kg for each prospective product, a price coefficient based on the thick, and a price reduction, the value is calculated. 1.4.1 THE SORTING WHEN CONSIDERING DEAD PIGS WEIGHT The four assorting clusters encompass approximately an equal quantity of swine and are being employed to position the remains on bars. The dissemination of pigs based on their deads weight and the assorting clusters can be observed in the graphical representation depicted in Figure 1 beneath. (kjaersgaard N. c., 2008) Figure 1:The distribution of pigs on dead’s weight and sorting groups. The moneys of the 43k+ pigs used in the check is calced as to: (Table 2) MARGIN DDK 37 Million (kjaersgaard N. C., profit computed, 2008) The final of trial with gang of 4 sorting groups with respect on deads weight 1.4.2 THE SORTING BASED ON LIPIDS When mounting the carcasses on bars, all four sorting groups—which are around the same size—are used. At the present level of thick measurement, the distribution of pigs We can see an instance of assessing accuracy and sorting groups seen in Figure 2 below. (kjaersgaard N. C., 2008) Figure 2: The distribution of At the present degree of measurement precision, pigs on the thick measurements and category designations. The simplest method to determine the profit using the bars is as follows: (Table 2) PROFIT DDK 37,827,885 (kjaersgaard N. C., 2008) The profit with sorting based on thick at the mode extent of scaling correct. 1.4.3 THE GRADING WHEN CONSIDERING CADAVER MASS AND LIPID CONTENT In this particular scenario, the sorting process has been enhanced to account for both the deads weight and the thickness of the animals, introducing a higher level of precision and detail. Instead of completely overhauling the existing sorting methodology, the previous sorting limits have been retained. However, to accommodate the increased complexity and achieve more accurate categorization, a significant expansion has taken place. The sorting system now encompasses a total of 16 distinct sorting groups, a substantial increase from the original four. This expanded framework allows for a more nuanced and refined sorting approach, ensuring that the carcasses are precisely placed on the bars based on their unique weight and thickness characteristics. By incorporating a greater number of sorting groups, the overall efficiency and effectiveness of the sorting process are significantly improved, leading to enhanced productivity and streamlined operations in the overall workflow. (kjaersgaard, 2008) The sorting groups and sorting limits. The money big for the dan sewer dead’s to better scaling with sorting based on both dead weight and thick can seen this graph: (Kjaersgaard, 2008) It demonstrates the value of enhanced measures for Danish deadshouses. The improved revenues for Danish deadshouses could be more than DKK 120 million annually if the measures were exact (present measuring error decreased by 100%). Even if the 16 sorting groups shown significantly increase profit, there is still room for profit growth by defining the sorting limits more intelligently rather than just mandating that they all roughly fall within the same size range. Perhaps even with considerably fewer sorting groups, this is still feasible. CHAPTER 4 Sutharsan S. INTRODUCTION Over 75% of the population of India works in agriculture either directly or indirectly, making it the foundation of the country's economy. Beyond the customary The status of rural communities has changed as a result of altered planting patterns in agriculture.th e horticulture industry produces around 12% of agriculture's value added, its importance cannot be understated. Grading of these small fruits is highly significant because it might increase the grower's profit. Grading enhances post-harvest processes like handling, packaging, and other. Grading is the process of distinguishing various homogeneous groups according to characteristics including dimensions, form, and sturdiness. Time and resources may be saved in a variety of ways. reducing handling losses during transit and processing processes. Fruits are often rated manually in the nation. When labour is scarce during busy times, Human marking turns into a costly, laborious procedure which even affects the company's bottom line. (Narvankar and Jha, 2005). Graders were first created more than 50 years ago, and they were basically a slat with an incomplete hag attached at the end. Before getting actually placed into the carrier, the things were examined on the monitor. These devices, known as "slat graders," were the forerunners to automated assessors. Grading hasn't altered all that much in the past 50 years. However, the grading procedure is now entirely automated. A chain made up a mechanical grader. a bag is at the end of a conveyor line. Produce that was smaller fell through the cracks, simplifying the grading procedure. Grading is still done by hand in India. There was a hunt for automated solutions due to a less on of uniforms in the process and a labour scarcity. Vegetable grading provides a larger emphasis on quality assessment in order to respond to market demand, which increases the Malcolm. The lights, the scanners and the running locale. Which set rat a The number of commercial pathways which can be applied to reach the optimum level throughput and quality need for better, more precise grading and sorting procedures. Onions and potatoes, which vary in size Many types of grader have been invented by a number of researchers. On the basis of the market demand and processing aspects. In general, size, shape and type are used as a basis for grading. MATERIAL AND METHODS According to market demand and processing considerations, numerous researchers created a wide variety of graders. Typically, grading is done according to size, form,colour, weight, etc. Here, a variety of studies show how various sorter and grader types are utilised in the processes of separating and categorising different food kinds, such as veggies, fruit, and other crop-related goods. FORM CATEGORISATION Employing spinning boards that might alter the illumination, detectors, working environment, rate of motion, and rotational rotating velocity, Malcom and Del. Garmo assessed apples and potato in 1953. They offered a number of advantageous suggestions. which can be applied to reach the optimum level throughput and quality. The translation rate shall be approximately 9-6 al in every minute and rotation. The translation speed is 6 to 12 r.min.1. Three grades with sizes between 50 mm and 57 mm, 63 mm to 64 mm were defined by Mack, Larson and White in 1956. seventy-one milis. They made a measuring belt-style assessor from of robust metal that was connected in rectangular connections for grading the fruit. The fruit were separated into three groups after washing. To ensure the protection of apples against bruising, it recommended an additional foam cushion thickness of 25.4 mm. Van (1965) developed a process of grading tomatoes spirally, machine. The vegetable descended an opening and landed in an upward path. with this machine, then dropped straight into the proper opening of the rotating disc. The unit It includes an elevator to lift the via the cleaning roller to the rise container with fruit. The conveyor's channel precursors are used to place fruit accurately on the coordinated cleaning blades. In nineteen sixty seven, But created an instrument for sizing and classifying apples. With the use of soft polyurethane rollers and lifting fruits from a float tank, as well as loading them into two sides on an optional rotating brush belt, very smooth handling has been achieved. The first three metres (ten feet) of the route was occupied by organisers who divided second-grade produce into two paths and discarded culls. The location of the strokes then revealed. would rotate to an appropriate speed so that they were polished and brushed. Goodman, In 1968, Hamann created, tested, and built a sweet potato collecting apparatus. They were driven so that it could assess the depth of the roots. There is a sloping design on the uncovered bottom of this V. The vehicle has gotten wider as it moves towards the front end to the back. To either side of the V, circular transparent vinyl plastic was positioned. The belt from step one was sheaves keyed into an axle operated by wheels and put in a cushion wedge bore from the power motor's link.. This differential speed caused any object falling onto a sizing belt, such as roots, until the longitudinal axis became aligned rotated with that of the belts. Roles to the bottom of the pyramid would be the root at this point. V. The unit had a capacity of 160 bushels an hour with the belt speed of 27.5 m per minute. In passing over the size estimation mechanism, damages were minor and in accordance with accepted limits. The potatoes are checked at 2.5,6.0 cm, 5.0, 15.0 cm of varying sizes. Hunter and Yaeger (1970) completed experiments using the rolling tabletop, examining elements such feed rate, percentage of defective tubers, the velocity of translating, spin of the rolling elements, angle of a rotation, manner in which the tubers were taken off, and the strength of the light. It was Lastly, the feeding rate may be changed to achieve this goal. removing the harmed tubers. There are currently within 6.2k kilogrammes of faulty tubers, each weighing 150+ g on median. Each hour, the roots are being taken out.. Once the error degree hit 20%, the wheel's velocity was reduced. In an ordinary assessment, over ninety percent of the problematic potatoes have been eliminated at an average rate of flow of approximately 4x100+5x10 tubers each min. In order to remove the most defective tubers, they correlated feed rates, crop translation and rotation of rollers in a test. The type of potato sizer that Grover and Pathak built in 1972 could be equipped with a wire belt for the measurement of four tuber sizes. The prototype has been driven with 2 horsepower electric motors. Numerous variable belt is where Vsize assessors for permanent use as well as packing machines for cucumber and sweet potato were developed by Brantley, Hamann, as well as Whitefield in the year 1975. The dependability as well as the longevity of these assessors in actual usage were evaluated. The outcomes demonstrated that the good had little harm and a reasonable dimension correctness. wobbles, For sweet potatoes, 95% to 98% and for cucumbers, 94%. With the belt speed of the sizing belt, the best results were obtained. 1 + 0.1*7 velocity units, angling the roller at spheed ratio 2 m per g. In 1980, Davison created a peanut sizer that drove the product from the apex of an incline pair of wheels down using the departure wheel method. the course and speed of a blad had an influence on how fast each roller moved through the various clearance ranges between them. The size classes have been identified in the catch pan. Developing a pair of parallel blade rapid adjustable hole grading in the year 1980. was Hutchison and McRae. That became the simplest tool ever made for gauging crops and crops. For crops like carrots and parsnips, which have a tubular or curved shape and are vulnerable to fracture and poor size, a rectangular grid filter was utilised, they had been popular for some time. Compared to a hand operating parallel bar puzzle, it has been efficient in grading Mark Piper variety. 96.8% . The main features of the study paper should be summarised in a conclusion that will allow readers to read information more clearly. Though conclusions normally do not present new information that is not referred to in the article, they often recast issues or offer a different view on this subject.(10) FOOD GRADING METHOD The health of humans greatly depends on the calibre of meals. due to an immense population and the increased demand for food products, the preferred quality has decreased. In contrast to When employed in a wide range of diverse industries, artificial intelligence is rapid, efficient, and dependable; yet, manually arranging vast quantities of output is time-consuming, expensive, and inconsistent. The expansion of fully computerised processes has made it possible for speed and accuracy to meet manufacturing and quality standards. Automation is a technique used to control a progression as effectively as possible. Additionally, it controls the system using a built-in instruction. The automatic grading system shortens the time it takes to advance while also minimising its flaws. The following are the steps included in the suggested method: (i) Preprocessing - Equalising the histogram MRG segmentation (ii) A grey-degree The combination Matrix's traits and characteristics colour histograms (iii) ANN The preprocessing procedure is first applied to remove the extra noise components that impede computing. The preprocessing approach uses the histogram equalisation method along with an RGB (red, green, blue) to grey level picture remodelling technique to enhance the image's contrast. HISTOGRAM EQUILIZATION The histogram equalisation process in this section boosts the image contrast. To achieve enhanced contrast, the histogram equalisation method spreads The degree of detail levels throughout the entirety of the set of information. beliefs that are radically at odds, such as when the backdrop and foreground neither completely brilliant nor entirely gloomy, are an effective way to convey an image. The transformation of greyish degree s to grey degree occurs when the movement of grey degree t is the same, is known as histogram equalisation. The procedure in this conversion expands the range of grey levels for the histogram maxima. As a result of the conversion, many image properties are easier to spot because the contrast is increased across more image pixels. Contrast has the effect of changing each pixel's strength according to its local area. The greatest and weakest luminance levels that comprise the image are in sharp contrast with one another. As a result, the histogram balance picture's regional difference zone gains a greater brightness that is suitable with the general difference. photos are then made available with characteristic elimination improvement to improve the contrasting characteristics of the images. DEFECT DETECTIONS OF FOOD Commercial sorting equipment uses traditional computer vision to analyse images of fruits and vegetables for colour, texture, size, and form. Due to the wide variety of fault forms, defect Identification remains to be an obstacle Uay & Gosseln, the year 2006. The automatic detection of fruit disease is crucial since it manifests on fruit. Fruit disease reduces yield and quality and manifests during harvest. Apple fruit diseases include scab, rot, and blotch. The demand for real, unbiased quality assessment of food goods is constantly growing in line with rising standards for food products. Computer vision systems offer automated, economical, and nondestructive methods to handle these requirements. Hyperspectral image analysis using principle component analysis. The existence of bruising can easily be determined by looking at the second and third main component pictures. The classification shows that 93.00% of the apples are unblemished. Citrus fruit detection algorithm proposed by Blasco et al. (2007) uses region-oriented segmentation. Interest is focused on the noise, the base, and faults. In ninety-five of situations, using the suggested approach will effectively pinpoint the issue if the item's greatest skin area is still preserved. Zhang and colleagues in 2010 differentiated skin from fruit and vegetable textures utilising a variety of rgb sensors. A given apple image's region of interest is found by it. GRADING OF DRAGON FRUIT In deteriorated and rainfall places across the globe, fruits such as fruit such as dragon fruit, pitahaya, and cherry pears etc.) are becoming mega commodities. It mimics a growing cacti plant and is strong enough to endure either natural and artificial conditions. Once established, the fruit blooms very quickly. After the plant has been left in place about an entire year, it can start to produce fruit. Every annual, this tree produces fruits for a duration of 5 months; following flowering and pollination, the fruits mature in around 50 days.(11) MATERIAL AND METHOD The suggested method identifies and divides a dragon fruit according to their specific characteristics using 3 artificial intelligence methods: ANN, Support Vector Machine (SVM), and CNN. In this approach, a collection of photographs of the dragon fruit and the camera-captured pictures of dragon fruit are utilised to carry out image processing tasks as collecting images, picture preparation, image division, and extracted features. Multiple kinds of illnesses, as well as sizes, forms, weights, and colours that indicate the fruit's quality, are among the characteristics of the dragon fruit block. The first one transmits the feature data together with the training and test photographs of the dragon fruit into the computation loop of the technique. A raspberry 3,14 function is used to count all the different fruits inside the fruits box. The picture comparison mechanism of the programme is then used to compare the camera's pictures. Before categorising the berries using an assessment or arranging block, the system assesses the kind of disease contained in the fruit along with its dimensions, pounds, form, and pigment. 3 preareas derive out of an A N N: a parameter layer, a layer that is concealed, and an additional layer at the end of the process. Weights, prejudice, and additional characteristics are all influenced by synapses. The type ANN utilised in the algorithm will determine The total number many skew factors & concealed layers. Every neuron performs a summation process and stores information about input data in weights. When an error is made in the output, Through the resultant system, the source network gets return or an additional dispersion connection.(4) SORTING OF CITRUS Since the majority of consumers identify quality with a pleasing look and the complete absence of visible faults, The frequency of skin flaws is one of among the most important factors affecting the price and calibre of newly picked produce. Because of this, packinghouses demand more sophisticated technologies that can identify skin flaws without mistaking them for fruit.(6) MATERIAL AND METHOD The Sony XC-003P, a trio CDC (in conjunction charging gadget) cameras were used throughout experimentation to snap shading illustrations. It can capture images with a resolution of 768 576 pixels. Eight 25 w daylight fluorescent tubes made up the lighting system, which also included polarised filters to prevent cross-polarization from producing bright spots in the scene. Off-line image acquisition was used to capture images of a single produce at an aspect ratio of 0. seventeen mm per pixel. The fruit's flaw-containing side is being manually shown to the camera. For each fruit, if necessary, multiple photos were captured. CONCLUSION In order to determine which colour space would be best for categorising the various faults under investigation and the stems, five distinct colour schemes were compared. The same results were obtained using discriminant assessment for the remaining hue environments, nevertheless altogether, HSI exhibited the greatest rate of achievement eighty-seven. REFERENCES 1. Nambi VE, Thangavel K, Jesudas DM. Scientific classification of ripening period and development of colour grade chart for Indian mangoes (Mangifera indica L.) using multivariate cluster analysis. Sci Hortic. 2015 Sep 2;193:90–8. 2. Kondo N. Automation on fruit and vegetable grading system and food traceability. Trends Food Sci Technol. 2010 Mar;21(3):145–52. 3. Razmjooy N, Mousavi BS, Soleymani F. A real-time mathematical computer method for potato inspection using machine vision. Computers and Mathematics with Applications. 2012 Jan;63(1):268–79. 4. Ali SSE, Dildar SA. An Efficient Quality Inspection of Food Products Using Neural Network Classification. Journal of Intelligent Systems. 2020 Jan 1;29(1):1425–40. 5. Hemamalini V, Rajarajeswari S, Nachiyappan S, Sambath M, Devi T, Singh BK, et al. Food Quality Inspection and Grading Using Efficient Image Segmentation and Machine Learning-Based System. J Food Qual. 2022;2022. 6. Blasco J, Aleixos N, Gómez J, Moltó E. Citrus sorting by identification of the most common defects using multispectral computer vision. J Food Eng. 2007 Dec;83(3):384–93. 7. Ziaratban A, Azadbakht M, Ghasemnezhad A. Modeling of volume and surface area of apple from their geometric characteristics and artificial neural network. Int J Food Prop. 2017 Apr 3;20(4):762–8. 8. B HJ. APPLICATIONS OF ARTIFICIAL NEURAL NETWORKS AND GENETIC ALGORITHMS IN DRYING OF FRUITS AND VEGETABLES : A REVIEW. Vol. 11, LAUTECH Journal of Engineering and Technology. 9. Zheng B, Huang T. Mango Grading System Based on Optimized Convolutional Neural Network. Math Probl Eng. 2021;2021. 10. Bhargava A, Bansal A. Fruits and vegetables quality evaluation using computer vision: A review. Vol. 33, Journal of King Saud University - Computer and Information Sciences. King Saud bin Abdulaziz University; 2021. p. 243–57. 11. Patil PU, Lande SB, Nagalkar VJ, Nikam SB, Wakchaure GC. Grading and sorting technique of dragon fruits using machine learning algorithms. Vol. 4, Journal of Agriculture and Food Research. Elsevier B.V.; 2021. CHAPTER 5 Karan S. Sonaji INDRODUCTION Maintaining agricultural products' safety at the end of harvest In order to make sustainable agriculture possible, the stage has become very important. The security of food and agricultural products is at stake when it comes to reducing waste. It is thus currently being considered as a matter to be urgently addressed by policymakers at the macro level in numerous countries. The after-harvest procedure in Tehran results in the loss of twenty million metric tonnes of the sixty-seven million tonnes of crop products generated there annually, or over thirty per cent of total output. That much could supply food for twenty million individuals for an entire year. The loss or destruction of agricultural commodities exceeds thirty percent. Also, water used for agricultural purposes is wasted. Therefore, the high level of waste in agriculture and water may be mainly controlled by means of proper harvesting, transportation, storage and processing. A reduction in farming waste may contribute to the growth of gross domestic product. Gross Income & the added value of crop products. One of the agricultural items that is most commonly consumed worldwide is carrots. They contain a lot of nutrients that are healthy. for maintaining your health. Precursors of the carotenoid pigments in vitamin A, such as those found in carrots, support better vision. In the past, specialists and human inspectors' eyes and hands were the process of assessing the superiority of crop products. Evidently, the effectiveness of the approach is slow, and conventional methods have proven to be expensive and ineffective in meeting rising customer demands, which have led to the demand for higher-quality products and quicker sorting processes. Sorting agricultural products according to their quality and shape is one of the most fundamental and crucial processes that must be done after harvest. Sorting activities make it easier for customers to recognise a product's quality and promote a better organised supply and distribution of agricultural goods. Consequently, the form and look of agricultural products serve as a market standard. Carrot fruits with irregular forms are a result of developmental genetic abnormalities. Due to their poor marketability, these uneven shapes make the product difficult to sell, which prolongs its time on the market where it will eventually rot. However, implementing appropriate techniques for quality grading and packing this product properly can reduce waste and improve its marketability. In order to accurately categorise carrot, produce in order that you can use it in a computerised vision device that is independent, this study attempts to propose a workable solution based on improved deep CNN. The accuracy and applicability of such a system in grading the product are two key evaluation factors. MATERIALS AND METHOD Fresh carrot fruits for this investigation were bought from an agriculturalist in Kermanshah, Iraq. eighty-seven instances of carrot in multiple forms four fifty instances of carrots in their traditional form and 428 Samples of an erratic shape were chosen. The picture of the samples was then captured utilise the photographic equipment. The graphic system included a camera made by Canon in Japan as well as a lighting box with two LED bulbs. In this work, background picture removal is used to do preprocessing and classification. Additionally, the size of carrot photos (4128 3096 pixels) slows down image loading processing and analysis. The size of the picture was subsequently decreased to the following dimensions: 16 16+8 16x3 with the goal to contrast and achieve the best degree of precision in classification. Convolutional neural networks are known to be gregarious approaches; They develop a knack to categorise things in a systematic manner require a significant volume of labelled data. A fundamental method for obtaining a significant amount of training data andenhancing deep learning model generalisation in the process. Fast AutoAugment is a newly proposed technique (Shorten and Khoshgoftaar, 2019) for automatically searching augmentation methods. This approach significantly improves the outcomes of picture categorization while reducing the computational complexity of deep learning algorithms (Lim et al., 2019). Parts of Deep Augment include the Controller, Augmenter, and Child model. A search software acting as the control system selects an information improvement technique from the available options. The convolution layer is the foundation of the CNNThese layers of data are used to identify and extract the irregular aspects of each object in the pictures. One might imagine the CNN output as a group of neurons in three dimensions. The parameters for the convolution layer are filters with learning capabilities. These filters Each of the layers of inversion must create a multidimensional activation matrix while retrieving the visual attributes map. The early stages of the cortex have minor features like routes, curves, and corners; later layers reveal higher-level elements like portions or even actual objects in the images. In the proposed model's Configuration 1 there are two convolution components, which include a completely connected layer, two regularisation batch sections, and an aggregate tier. Variant #2 within the suggested architecture contains 6 layers of convolution, a trio of batch standardisation, a pair of accumulating, and a layer that is completely linked. The physique of the model third configuration has four batch normalisation layers, three pooling layers, eight an entirely linked layer, the network layer as well as a convolutional layer. Relearning acted in the activation function in the suggested architecture. This method scales and modifies activations to normalise the input layer. The initial proposal was to address internal covariate shift. The allocation During the training phase of networks, The data provided in each layer beneath it alter in accordance with the characteristics of the levels above it. As a result, which current layer constantly needs to adapt to new distributions. With regards to advanced systems, the issue is made considerably worse since the propagation of tiny changes in the network's shallower hidden layers amplifies those changes, leading to a significant change within the deeper-buried layers. Therefore, normalisation by batch is used to reduce undesired changes, quicken instruction, thus provide a more precise representation classifying carrots. In addition to lessening internal covariate shift, batch normalisation provides a number of other benefits. With this extra layer, gradients don't vanish or explode, allowing the network to learn more quickly. Batch normalisation also regularises the network such that generalisation is facilitated, which eliminates the need for dropout to reduce overfitting. The network becomes more reliable and applicable with various initialization strategies and learning rates, which is yet another advantage. A batch normalisation layer normalises each input channel across a minibatch. Convolutional neural networks be trained more quickly and with less initialization sensitivity by using batch normalisation between convolutional layers and nonlinear layers, such ReLU layers. Efficiency is a key concern in various industries, including the food sector. As food sorting processes play a crucial role in ensuring product quality and safety, optimizing these processes can significantly enhance overall efficiency. Mathematical optimization techniques offer valuable tools to address the challenges associated with food sorting and improve its effectiveness. This essay aims to explore the application of mathematical optimization in food sorting for improved efficiency. I. Background Information Food sorting involves categorizing and organizing food products based on various attributes such as size, shape, color, and quality. This process is essential for streamlining operations, reducing waste, and maintaining product integrity. However, achieving high levels of efficiency in food sorting can be challenging due to several factors, including variations in product characteristics, time constraints, and resource limitations. II. Problem Statement The specific problem that needs to be optimized in food sorting for improved efficiency is the allocation of resources such as manpower, equipment capacity utilization, and time management during the sorting process. Optimizing these aspects becomes crucial because they directly impact overall productivity and cost-effectiveness. III. Mathematical Efficiency Techniques Mathematical Efficiency techniques provide systematic approaches to finding optimal solutions by maximizing or minimizing objective functions while satisfying specified constraints (Li et al., 2021). Various mathematical optimization techniques can be applied to solve problems related to food sorting efficiency. Linear programming (LP) is one widely used technique that represents problems through linear functions with linear constraints (Hovorukha et al., 2019). It has been successfully applied to optimize transportation logistics in distribution centers (Correa Issi et al., 2020). Integer programming (IP), an extension of LP where decision variables must take integer values,is also applicable when certain parameters are discrete or non-linear relationships exist(Ogurtsov et al., 2021). Other optimization methods include genetic algorithms(GAs), which mimic natural selection processes to find optimal solutions, and simulated annealing (SA), which simulates the cooling process of metals to find global optima. These techniques offer advantages such as flexibility in handling complex problems and the ability to search for global optima (Li et al., 2021). IV. Formulation of Mathematical Model To formulate a mathematical model for food sorting optimization, relevant variables, constraints, and objective functions must be identified. Variables may include workforce allocation, equipment utilization rates, product flow rates,and scheduling parameters. Constraints can represent limitations on resources such as time availability or capacity restrictions of equipment. For example, the number of workers assigned to each sorting task should not exceed their maximum capacity or conflicts between multiple product lines. Objective functions are designed to maximize productivity while minimizing costs or other performance indicators. This may involve maximizing throughput while minimizing labor costs or optimizing resource allocation based on time efficiency. V. Solution Methods Several solution methods can be employed to solve the formulated mathematical models. - Simplex method: Used for solving linear programming problems with continuous variable domains. - Branch and bound: An algorithm used for solving integer programming problems by systematically exploring possible solutions through branching and bounding techniques. - Genetic algorithms: Iterative optimization algorithms that simulate the evolution process to identify optimal solutions through selection, crossover, mutation operations. - Simulated annealing: A probabilistic optimization technique that searches for global optima by accepting suboptimal moves initially but gradually moving toward better solutions over iterations. The choice of solution method depends on factors such as problem complexity, computational resources available,and desired level of accuracy in results. VI. Case Studies/Examples Real-world case studies demonstrate how mathematical optimization has been successfully applied in food sorting processes: - In one study (Hovorukha et al., 2019), hydrogen production from food waste fermentation was optimized using mathematical modeling techniques. The study aimed at finding optimal parameters such as temperature, pH levels, and substrate concentration to improve efficiency. - Another study (Correa Issi et al., 2020) utilized optimization models to schedule truck arrivals and departures at a distribution center. This enhanced overall efficiency by reducing wait times and improving resource utilization. These examples highlight the potential benefits of applying mathematical optimization techniques in food sorting processes to achieve higher productivity, reduce waste, and optimize resource allocation. VII. Challenges and Future Directions Despite the potential benefits of mathematical optimization in food sorting for improved efficiency, certain challenges exist. These include the complexity of real-world scenarios with multiple variables, uncertainties associated with product variations,and computational constraints. Future advancements can be made by incorporating big data technology into mathematical models (Li et al., 2021). By leveraging large datasets on product characteristics, production rates, equipment capabilities,and historical performance data, more accurate optimization models can be developed. Additionally,collaborative research efforts between mathematicians,modelers,and industry experts can further enhance the application of mathematical optimization in food sorting. Such partnerships can ensure that models capture essential industry-specific factors and are practical for implementation in real-world environments. VIII. Conclusion Mathematical optimization holds great promise for improving efficiency in food sorting processes.Given the challenges faced during these processes,optimizing resource allocation,time management,and equipment utilization is crucial.Mathematical modeling enables precise representation of problems while various solution methods provide avenues to find optimal solutions.Realworld case studies have demonstrated the effectiveness of these techniques.Future advancements should focus on utilizing big data technology as well as fostering collaborations between academia and industry.This will drive innovation towards enhancing efficiency,reducing waste,and ultimately benefiting various aspects such as cost reduction,responsible resource utilization,and overall productivity REFERENCES Li, Z., Huang, Y., & Li, Z. (2021). Stability Evaluation and Optimization of Food System Using PEFS Mathematical Model and Big Data Technology. Journal of Physics: Conference Series, 1952(4), 042113. https://doi.org/10.1088/1742-6596/1952/4/042113 Hovorukha, V., Tashyrev, O., Tashyreva, H., Havryliuk, O., Bielikova, O., & Iastremska, L. (2019). Increase in efficiency of hydrogen production by optimization of food waste fermentation parameters. Energetika, 65(1). https://doi.org/10.6001/energetika.v65i1.3977 Ogurtsov, V. V., Kargina, E. V., & Matveeva, I. S. (2021). Optimization of Log Sorting by Diameter. Lesnoy Zhurnal (Forestry Journal), 1, 150–158. https://doi.org/10.37482/0536-1036-2021-1-150-158 Correa Issi, G., Linfati, R., & Escobar, J. W. (2020). Mathematical Optimization Model for Truck Scheduling in a Distribution Center with a Mixed Service-Mode Dock Area. Journal of Advanced Transportation, 2020, 1– 13. https://doi.org/10.1155/2020/8813372