Applied Research Of Machine Vision And Pattern Recognition In Soy Color Separator

Machine vision and pattern recognition technology which is an important research area in artificial intelligence has wide application in the industry. Soybean separator is a dedicated device used to eliminate impurities from raw soybean on production line. The automated detection system integrated by modern photoelectric technology and computer science, greatly improving the production efficiency, is widely used in agricultural products industry at home and abroad. The separation algorithm research was carried out by combining the specific needs of the soybean separation industry and in-deep studying of theories and techniques of machine vision and pattern recognition. The main contents of this dissertation are as follows:The overall structure design of soybean separator system was demonstrated. The feeding mechanism, photoelectric detection as well as separation mechanism design were presented.The camera image processing platform was introduced. Because soybean separator has a higher real-time and accuracy requirements, the digital image processing platform must be able to support the complex algorithm and have the ability to handle huge amounts of high-speed data. By comparison with separator machines hardware platform at home and abroad, the embedded system based on DSP and FPGA was adopted as the core of image processing system in the paper. Part of the embedded system hardware design was introduced along with data communication module between FPGA and DSP.Soybean separation algorithms based on machine vision and pattern recognition technology were proposed in this paper, including the image data read preview, pattern recognition, image preprocessing, data distribution model establishment and color identification lookup table establishment. The naive Bayes classifications as well as multi-layer perceptron method, RBF neural network, k nearest neighbor method and decision tree algorithm were studied in the paper. Considering the color accuracy, color chosen speed and calculation difficulty, the decision tree algorithm was finally chosen as color sorting algorithm. The accuracy of the classification algorithm is as high as 99.998%, the area under the ROC value is 1.