Research On Classification,Identification And Detection Technology Of Impurities In Machine-picked Seed Cotton

China is a major cotton producing country,and Xinjiang is the most important cotton producing area in China.As the level of mechanization in Xinjiang increases year by year,the processing technology of machine-picking cotton has also developed.In the process of cotton processing and production,a variety of cotton processing machines are involved,such as cotton dust separator,seed cotton control box,seed cotton cleaning machine,flower tying machine,lint cleaning machine,etc.The speed adjustment of these equipment needs to be based on the cleaning effect to produce higher quality cotton,and the processing technology needs to be reasonably arranged according to the cleaning status of each equipment.Seed cotton cleaning is the first part of the process,and it is important to monitor the condition of seed cotton cleaning in real time.In view of the above problems,this paper studies the technology of classification,identification and detection of impurities in machine-picked seed cotton based on image processing and deep learning technology.The main contents of the research are as follows:(1)In this paper,the impurities in machine-picked seed cotton were analyzed,and the density of various impurities was measured.According to the testing requirements,the structure of the test bench is designed,the light source and image acquisition equipment are selected,and the test bench for machine-picked seed cotton is assembled.(2)In this paper,an algorithm for impurity segmentation of machine-picked seed cotton is designed.Aiming at the problem that machine-picked seed cotton and impurities are difficult to distinguish,the distinguishing effect in different color spaces is analyzed.For the problem of noise interference of machine-picked seed cotton images,the edge detection of traditional Canny operator is improved by multiple noise reduction and increasing gradient calculation direction.Finally,based on the test bench for machine-picked seed cotton,the test is carried out,and the algorithm with better segmentation effect in the test process is selected for comparison,and the stability and accuracy of the algorithm are verified.(3)This paper improves the YOLOv4 neural network,and designs a loss function for the classification and identification of impurities in machine-picked seed cotton to increase the training of difficult samples,and compare and select the activation function.Design experiments to compare the recognition effect and speed of the improved algorithm and the traditional algorithm.(4)In this paper,a V-W mathematical model is established,and the impurity content obtained based on the area information of the seed cotton image is extended to threedimensional space to obtain the volume-based impurity content,and then the mass-based impurity content is obtained through the density of substances in the seed cotton.Taking the impurity content obtained by the mass method as the standard,the design experiment compared the impurity content obtained by the V-W model,the image area and the mass method.It is concluded that the impurity ratio error based on the V-W model is effectively reduced.(5)This paper develops a real-time detection system for machine-picked seed cotton based on the Pycharm Community 2021.1 environment.Use the deep learning framework Pytorch1.9.0 to build a network model,use the open source library Open CV 4.1.2 for image processing,and use the Py QT5 open source library to write interface programs.In order to improve the frame rate of real-time detection,the multi-thread technology is used to divide the working threads into image processing threads and data processing threads,and an experiment is designed to compare the processing time of multi-thread and single-thread,and realize the realtime detection of impurity rate.