نوع مقاله : مقاله پژوهشی
نویسندگان
1 گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی و منابع طبیعی، دانشگاه محقق اردبیلی، اردبیل، ایران.
2 دانشیار گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی و منابع طبیعی، دانشگاه محقق اردبیلی، اردبیل، ایران
3 استاد گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی و منابع طبیعی، دانشگاه محقق اردبیلی، اردبیل، ایران.
4 دانشیار گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی و منابع طبیعی، دانشگاه محقق اردبیلی، اردبیل، ایران.
5 استاد گروه مهندسی مکانیک بیوسیستم، دانشکده کشاورزی دانشگاه تهران، پردیس ابوریحان، پاکدشت، ایران.
چکیده
کلیدواژه‌ها
عنوان مقاله [English]
نویسندگان [English]
Abstract
Introduction
Wheat is one of the most important grains in the human food basket, which is known as a major source of energy, protein and fiber due to its valuable nutrients. The post-harvest stage of the wheat crop is explained in two ways: either it is sent to food processing factories or it is stored in silos for sale at regular intervals. Various parameters represent the quality of wheat grain that the percentage of purity of the mass is one of the main factors affecting the purchase price of the product. Several types of non-wheat grains, including germinated grains, broken grains, legumes, weed seeds, insect-damaged grains, foreign matter (pebbles, straw), etc. are the main sources of impurities in wheat. Researchers have always tried to develop computer-based solutions for impurities in wheat grain to be able to develop automated wheat grain separators. Image processing based on morphology, color and texture characteristics of grains has been used for various applications in the grain industry, including grain quality assessment and wheat classification. Various grading systems based on image processing have been studied. The presence of various impurities at the time of wheat harvest is one of the important factors in reducing the quality of wheat, so it seems necessary to detect impurities in wheat grain. The quality of wheat has a significant effect on its marketability. In addition, if wheat is used as a crop seed, the impurities in the mass will be a determining factor in the yield of the future crop.
Methodology
In statistical analysis of data, situations are sometimes encountered in which the relationship between problem variables is very complex. This makes it difficult to analyze and process the data, so that sometimes no definite relationship can be found between the variables. In these cases, instead of purely theoretical research, applied research is done. Artificial neural networks are one of the solutions that, by processing experimental data, discover the knowledge or law behind the data, and transfer it to the network structure. In this study, the possibility of identifying wheat in wheat grain mass and estimating the amount of impurities in the mass, based on video processing using two types of artificial neural network (ANN) algorithms and hybrid genetic algorithm (GA) has been investigated. For this study, the code related to the artificial neural network with two hidden layers and the number of different neurons in each layer was written in MATLAB software. This code was used to identify and classify each component in the wheat grain mass. The main task of ANN is to learn the structure of the model data set. To achieve this, the network is trained with examples of related outcomes to generalize the capability. Multilayer artificial neural networks (MLPs) are the most common ANN models. In the present study, to reduce the computational load and increase the accuracy of the results, as well as to save time, some parameters that can be changed in the genetic algorithm were extracted as a fixed number using trial and error method. Among these parameters is the number of layers in the main structure of the neural network. Results: A hidden layer with a number of neurons 2 to 12 was used as an even number. It should be noted that the number of neurons above this amount of computational time increased dramatically and did not have much effect on classification accuracy. Another parameter in this field is the Max Reproduction factor (Max Generation) which according to the results of trial and error for this factor, the results showed that increasing this value more than 30 has little effect on classification accuracy and decreases the mean squared error. And only increases the computation time, so a constant value of 30 was considered for this parameter. 4 values of 50, 100, 150 and 200 were used for the Pop Size parameter. Values above 200 dramatically increased computational volume and processing time, so values over 200 were omitted. Values less than 50 also reduced classification accuracy, and values less than 50 were excluded from the analysis process. After preparing the video of mass movement on the conveyor belt, using MATLAB software and image processing toolbox, 17 shape features, 12 color features and 6 texture features were extracted from each grain sample in the image. The data obtained from the image processing section were classified into five categories: wheat, barley, oats, straw and weed seeds. Two types of artificial neural network (ANN) algorithms, feeder (newff) and feeder (newcf), and hybrid genetic algorithm (GA) were used to achieve the highest classification accuracy and minimum error.
Conclusion
Techniques related to image segmentation were used to separate objects within the image. In this stage of image processing, an attempt is made to separate interconnected objects using a variety of morphological and color methods in the image. In fact, the purpose of separating interconnected objects in the image is to make it possible to examine the individual objects in the image separately and extract the different characteristics of each of them. The results showed that from 36 different artificial neural network (ANN) structures, the 5-4-10-35 structure for the newff algorithm with 100 and 89.74% accuracy for training and testing conditions, respectively, with a processing time of 10.39 seconds and the structure 5-8-10-35 for newcf algorithm was obtained with 100% accuracy for training conditions and 87.17% for test conditions with a processing time of 44.94 seconds. On the other hand, the results of the hybrid GA algorithm showed the highest classification accuracy with 95.55% and 86.66% for training and testing, respectively, in a structure in which 8 neurons in the hidden layer with a population size of 200 were used. Was obtained. According to the obtained results, the use of video processing using ANN artificial neural network and newff algorithm due to high accuracy and lower computation time is a powerful tool for detecting impurities in wheat grain mass. Therefore, the use of artificial neural network with the help of video processing has the ability to classify wheat grains and can be used in a practical way. Given the importance of grain mass velocity in the discussion of industrial application, it is suggested that higher grain mass velocities be investigated in a similar way.
کلیدواژه‌ها [English]
)