طبقه بندی سه رقم بذر ذرت با استفاده از تکنیک پردازش تصویر

Introduction
Maize (Zea mays. L) is one of the most important crops acrossthe world that ranks third in terms of acreage behind wheat and rice. As this crop can adapt to different climatic conditions, it is of great importance and has a large area under cultivation.Therefore, maize is one of the major products of temperate, warm-temperate, subtropical, and humid regions. After wheat, rice, and barley, this plant is the main crop in Iran with the largest cultivated area.
There are different types of maizeseeds, so their classification is essential to ensure quality. A key component of sustainable agriculture is quality assurance. On the one hand, techniques such as drying, cooling, and edible coating must be used to maintain the quality of agricultural products. On the other hand, effective and efficient methods should be developed to evaluate and classify their quality, which is used in seed and seedling processing centers, silos, and mechanized warehouses.
The detection of various varieties of crop seeds using instrumental methods has been the subject of extensive research. As a non-destructive and rapid inspection method for the recognition and classification of cereal seed varieties, the visual machine is available. Machine vision-based automated methods can have a positive impact on food processing. In other words, this tool is the process of preparing and analyzing images of a real scene using a computer to obtain information or control a process. The features of images can be extracted using this machine to recognize and identify the quality of different types of products. To identify the types of plants, their growth patterns, and the effects of the environment on them to obtain more and superior products, machine vision occupies a special place and is one of the most important research areas. Inspection and quality control of factory output products is an important application of machine vision.Advances in image processing technology have opened up a wide range of machine vision applications in agriculture. The development of powerful microcomputers and specialized software has led to the use of image processing for the inspection of fruits and agricultural products, especially for quality control and sorting. Many agricultural products sorting systems used to separate fruits or crops based on color, shape, size, the extent of damage, crushing, bursting, spotting, etc., now rely on visual machines and image processing functions.Images of products moving on the conveyor system are taken by a CCD camera, transmitted to a computer for processing, and in these systems, the necessary data are extracted from them. Depending on the information obtained, commands are then issued to activate or deactivate a mechanical part so that the product can be removed from or allowed to cross the main path. Sorting is a common practice in many industries. Compared to mechanical systems, machine vision technology offers the highest accuracy and quality at the lowest cost and with the lowest error rate, so it can be considered the most effective solution to this problem. The agricultural industry is one of the areas where sorting and grading systems based on machine vision are urgently needed.The core elements of machine vision are image processing and analysis used together with new methods and classifiers such as neural networks, backup vector machines, fuzzy logic, etc. to perform classifications and required measurements. This study aimed to identify seeds of three maize varieties using macroscopic imaging techniques, evaluate the morphological and chromatic features in maize grains, and discriminate varieties using a stepwise method and remove some variables using LDA and ANN.
Methodology
Three seed varieties of single cross 703,single cross 704, and single cross 705 were provided by the Agricultural and Natural Resources Research Centre of Ardabil Province in Pars Abad Moghan. The seeds were then taken to the Biophysical Properties Laboratory of the Department of Biosystems and Mechanical Engineering, MohagheghArdabili University.Three samples (20 g) of each variety were stored in a laboratory oven at 105 °C for 24 h to determine the initial moisture content of maize grains. According to the dry weight of grains, the initial moisture content of them was calculated by 10.50%. To distinguish 3 maize varieties, 180 samples were analyzed as single seeds (30 replicates in the anterior direction and 30 replicates in the posterior direction) for each variety with 60 replicates. In addition, 9 more samples were used in bulk with 3 replicates for each variety.Thus, we imaged a total of 189 samples. In addition, a digital scale with an accuracy of 0.001 g was used to measure the weight of the grains. Computer vision systems consist of five main components: lighting chamber, camera, analogue-digital card (for digitization), computer, and computer software. Images were taken using a Canon IXY DIGITAL 510 IS digital camera. A dome-shaped chamber was used to reduce noise and control ambient light. The system was illuminated with four fluorescent lamps and two rows of LED lamps, one white and one yellow. While the camera was pointed perpendicular to the imaging surface, it provided images with a resolution of 12.1 megapixels.In this case, the images were processed using MATLAB software. First, 10 maize seeds were randomly sampled from the first variety (single cross 703) and weighed using a digital balance. Then, parameters such as the large and small diameters and thickness of each grain were measured using a caliper of 0.02 mm. Then, these grains were placed at appropriate distances from each other on a plate of red cardboard in the opposite direction to be imaged. Finally, 30 maizeseeds were imaged in both directions and 60 images were taken as single seed. In total, we obtained 180 images of all three varieties as single seeds. To prepare the mass, first, some seeds of the first variety were placed in a cylindrical container (1.5 cm high, 4.2 cm in basal diameter, and 70.62174 cm2 in volume) so that the container was filled. After weighing, the mass of grains with a certain volume was poured onto the red plate in a circular pattern. In the end, the camera was placed on the bulk sample and the image was taken, just like the single grain image.The same procedure was repeated twice more on two more bulk samples of the first variety. Similarly, three bulk samples of two more varieties were imaged. In this way, a total of nine images were obtained. After each imaging, we measured and recorded the dimensions and weight of 10 randomly selected seeds from the imaged bulk. In the end, 189 images were obtained, including 180 single-grain and 9 bulk images.In the single sample feature extraction step using the bwlabel function, all samples were labeled and the grain morphological features were extracted. Then, the set of Regionprop functions was used to determine eight parameters, including area, perimeter, major principal axis, minor principal axis, integrity, irregularity, convex area, and equivalent diameter. An artificial neural network (NAA) and a statistical linear discriminant analysis (LDA) method were used to identify maize varieties based on their morphological and color characteristics. The data were normalized before analysis. LDA is a statistical method for classifying objects based on independent variables. The analysis was carried out using SPSS software. The diagnostic analysis includes stepwise analysis, principal component analysis, and elimination of recursive features. In this study, the stepwise method was used. In the usual method, all variables are included in the analysis. However, in the stepwise method, some variables were removed and only the variables with the greatest influence on the model were included. To classify the maize varieties, a network consisting of three layers: input, output, and hidden layers was used.
Conclusion
We performed image processing to classify three maize varieties based on the results obtained. A linear diagnostic analysis method was used in this study. A total of 16 predictor variables were used with an accuracy of 70.6%. Some variables were eliminated by a stepwise method. In addition, eight other predictor variables were analyzed with the same accuracy of 70.6%. Thus, although the number of predictor variables was reduced, the detection accuracy remained constant. Moreover, the highest accuracy of diagnosis (80%) was associated with the first variety (single cross 703). Additionally, the accuracy of the methods of ANN with 16 and 8 predictor variables was 75.6% and 72.2%, respectively. These values were higher than that of LDA.Predictive variables included areas, perimeter, major principal diameters, minor principal diameters, irregularities, concave areas, equivalent diameters, color indices (red, green, and blue) resulting from maize grain sample processing, weight, and grain size. The following factors were the most important predictors of varietal discrimination: thickness, major principal axis, minor principal axis, irregularity, equivalent diameter, integrity, red color index, and green color index. According to the results, the length and width of individual grains had no significant effect on variety classification.Our finding demonstrated thatmachine vision technology can be used in seed and seedling processing centers, silos, mechanized warehouses, and other places where maize seed crops need to be identified and separated in a non-destructive manner.