Wheat Varieties Identification System Based On Machine Vision

Wheat is one of the main food crops in China,and about 40% of the population in China live on wheat.Extreme weather,pests and diseases all contribute to wheat loss,so breeding fine varieties plays an important role in increasing wheat yield.The quality of grain is the basis of breeding,and the traditional seed identification methods can not meet the needs of modern breeding,so improving the accuracy of grain recognition is significant to cultivate excellent wheat varieties.Based on feature selection and modeling technology of machine vision,this paper proposes a wheat grain variety recognition method,which identifies 8 varieties of wheat.The main work is as follows:（1）Wheat grain image dataset was constructed.The grain images of 22 varieties with wide planting area in the market,such as Bainong 207,Bainong 307,Bainong 419,Xinmai 26,Xunong 14084 and Yuyuan 916 were collected.Three angles（abdominal groove upward,abdominal groove forward 45 degrees and abdominal groove downward）were selected to display the main characteristics of wheat through study and comparison.1000 wheat kernels were collected for each variety.In order to distinguish different varieties,grains and angles,wheat grain images were renamed.And the background of wheat image was removed to construct wheat grain image data sets.（2）The wheat grain image features were extracted and analysed.Firstly,the image with background removed was preprocessed by graying,filtering,noise reduction and binarization.The color features（R,G,B,H,S,V and the mean,standard deviation and slope of HSV）,morphological features（area,perimeter,long axis length,short axis length,eccentricity,circumscribed rectangular area,moment of inertia,roundness and rectangularity）and texture features（inverse variance,energy,entropy and contrast）of wheat grain images,28 characteristic parameters in total,were extracted,and the correlation analysis of the extracted features was carried out.（3）Based on feature image classification of wheat grains,8 varieties of wheat were classified by bayesian optimized BP neural network combined with different feature processing methods.The recognition models of single angle and angle fusion were constructed respectively.There were 7 feature fusion models: morphology,texture,color,morphology + texture,morphology + color,texture + color and morphology + texture + color.Principal component analysis,linear discriminant analysis and important feature selection of Sklearn-Select KBest function were constructed.And the data enhancement models of wheat grain image brightness,chroma,sharpness and contrast were established.By comparing KNN and BP neural network recognition models and combining with different feature processing methods,wheat grain images of 8 varieties were recognized and analyzed.The recognition accuracy was greatly improved after the bayesian optimized BP neural network.Based on the bayesian optimized BP neural network + data enhancement model,the highest accuracy reached 95.58%,which improved the recognition accuracy of wheat grain varieties and enhanced the generalization ability of the model.（4）Wheat grain variety recognition system was designed and realized.By integrating the work in this paper,a wheat kernel recognition system based on feature was implemented.After logging into the system,users could upload wheat images,and realize the automatic recognition of wheat grain varieties through image preprocessing,data enhancement,automatic extraction and processing of image features and variety recognition.