Study On Classification Model Of Wheat Seed Varieties Based On Machine Vision

Wheat is the second largest grain crop in China,and it is also an extremely important grain reserve resource.The production of wheat is inextricably linked with our lives and even the development of society.The detection of wheat seed varieties is directly related to the subsequent development of wheat in terms of growth,yield,and processing.In recent years,with the rapid development of machine vision technology,it has been applied to wheat seed quality detection and variety classification.It not only has the characteristics of fast,nondestructive,high accuracy,but also helps the liberation of labor force and the improvement of working efficiency.In this thesis,the classification methods of wheat varieties are complex,subjective and low efficiency.The wheat varieties of Guohe experimental base in Lujiang County,Anhui Province,including Huafeng 6,Ningmai 15,Shengxuan.6,Wannong 55,Xinong 979,Yang fumai 5,Yangmai 23,Zhengmai 9023 were selected as the research object.The main research contents and results are as follows:(1)The de-noising method of wheat seed images was studied.By comparing the three methods of image enhancement methods of neighborhood average method,median filter method and self-applicable median filter method,the adaptive median filter method can not only weaken the image noise,but also protect the edge information of the image well.It was able to enhance useful information and reduce interference information.(2)In this thesis,the segmentation method of gray image and color image was studied.The image segmentation methods,such as threshold segmentation and region growing were compared.The results showed that these algorithms can segment the grayscale image of wheat,but can't directly segment color images.Aiming at the problem that the commonly used segmentation methods can't segment color images.The color image segmentation method based on SVM was proposed,which can be used to segment the color of the color wheat directly,and the processing speed is fast.However,it is easy to produce the wrong segmentation phenomenon.Segmentation using the SLIC algorithm requires manually setting the number of super pixels.This process is tedious and wasteful of time and space.Therefore,an improved segmentation algorithm based on SLIC and DBSCAN clustering algorithm was proposed,which can determine the number of super-pixels according to the color information of the image.The image was segmented into several pixels with uniform size by improved SLIC algorithm,and the feature regions of the image were effectively distinguished.The improved super pixel segmentation method effectively reduced the over segmentation phenomenon and obtained a nice segmentation result.(3)A wheat seed variety identification method based on PSO-PNN neural network was proposed.The recognition efficiency was improved by particle swarm optimization(PSO)algorithm.which was used to optimize the smoothing factor of probabilistic neural network.The extracted wheat seed characteristics are taken as input to construct a classification model.The classification results from PNN network,BP network and PSO-BP network were compared with the results of PSO-PNN network.The recognition accuracy of classification model based on BP neural network was 80.21%.The root mean square error(RMSE)was 1.4562.The classification accuracy of BP network model optimized by particle swarm optimization was 87.26%.The root mean square error was 0.9521.The accuracy of classification model based on PNN neural network was 90.1%.The root mean square error was 0.6892.The classification accuracy of PNN network model optimized by particle swarm optimization was 96.32%,and The root mean square error was 0.433.The experimental results showed that the particle swarm optimization algorithm is effective in optimizing the PNN neural network and can be used to classify wheat seed varieties.(4)A wheat seed classification system based on machine vision is designed and developed based on MATLAB language.The functions of image de-noising,image segmentation,feature extraction and variety identification of wheat seeds have been realized in this system.