Research On The Overall Design Of Pepper Impurity Removal System And Impurity Recognition Algorithm

Chili pepper impurity removal is a necessary process in the process of picking pepper.The raw material of pepper will be heavily mixed with various impurities,such as peanut,stone,plastic bag,chili stem,etc.Traditionally,the manual removal efficiency is low,the mechanical impurity removal device can't fully meet the requirements.Chili pepper impurity removal urgently needs mechanization and automation.In view of the above problems,the main work of the design of pepper impurity removal system was as follows.(1)According to the requirements of pepper impurity removal,a set of production line was designed.The system mainly includes the heavy impurity removal unit of pepper,which mainly removes the heavy impurities(such as gravel);the light impurity removal unit of pepper,which mainly removes the lighter substances(such as leaves);the objection detection module and color recognition module,and removes the remaining impurities.(2)The improved SSD objection detection algorithm was analyzed and improved.Peanut was taken as the test object.In the experiment,the recognition speed of the improved SSD(Single Shot MultiBox Detector)algorithm was 54 FPS,and the recognition accuracy was 88.21%.The results showed that the improved SSD algorithm improved the detection accuracy and speed greatly compared with the original SSD algorithm.The improved SSD algorithm and STC(Spatio-Temporal Context)algorithm were fused to track the target object,and the tracking speed reached 150 FPS while ensuring the accuracy.The tracking results and camera calibration were fused to calculate the position information of impurities,and the impurities were removed by the manipulator.(3)Researched on impurity color recognition and location algorithm,peanut ws taken as the experimental object,and the processing speed of 640×480 image was about 10 FPS by using color space conversion,mathematical morphology and clustering algorithm.Combined the clustering results and camera calibration,the impurities were removed by the manipulator.At the same time,the clustering results are saved,and the target detection algorithm is trained to improve the accuracy of the target detection algorithm.The system requires an overall recognition rate of 96%.The experimental results show that the target detection and recognition rate is 88.21%,the color recognition rate is about 77.43%,and the comprehensive recognition accuracy is 97.33%,which meets the requirements of the system.