Study On Wheat Appearance Quality Inspection Based On Machine Vision

Wheat is one of China's three major grain production,the output of the world first.In the process of wheat storage,there may be insect erosion,germination,damage and other problems,so it is particularly important to detect the appearance quality of wheat in the circulation link.At present,there are still some problems in using image processing technology to identify wheat appearance quality,for example,the image algorithm of wheat adhesion is complex and the speed of wheat detection is slow.In view of these problems,the main content of the paper is as follows:(1)Construction of wheat appearance quality inspection system based on machine vision.The selection of industrial camera,lens and light source was completed according to the detection requirements,and the image acquisition equipment was set up on the conveyor belt frame to realize the real-time acquisition function of wheat on the conveyor belt.A single grain guide groove with three transverse dispositions was designed and installed on the frame of the conveyor belt.The wheat was separated by the relative friction force between the single grain guide groove and the wheat,which provided hardware support for the use of a simpler image segmentation algorithm in this study.(2)Preprocessing and feature extraction of wheat image.In this study,median filter was used to filter wheat image.After the wheat image is converted to H channel,the wheat region is extracted by threshold-based segmentation method.Morphological processing of wheat was done with open and closed operations.After the pretreatment,the color feature,shape feature and texture feature parameters of wheat were extracted,and the extracted 22 feature parameters were analyzed by principal component analysis.The results showed that the cumulative variance contribution rate of the first 9 principal components could reach more than 95%,and the first 9 principal components were used as input layer data in mathematical modeling.(3)Pattern recognition.The BP neural network model and SVM model were constructed to classify wheat appearance quality.The BP neural network model used Sigmoid function as activation function,the number of hidden layers was 3,and the number of nodes of hidden layers was set as 38.The recognition accuracy of wheat could reach 97%.Using radial basis kernel function and "One-Against-One" method in SVM model,the recognition accuracy of wheat was 91%.Therefore,this paper uses BP neural network model.(4)Software design of wheat appearance quality recognition.Using Halcon and C# programming,design a simple operation of wheat appearance quality recognition software.The camera SDK is used in the software development,and the queuing and multi-threading are introduced to improve the software running speed.The detection speed and recognition accuracy of the software were verified.The detection speed was7 grains/second,and the recognition accuracy was 92.28%.