The Research Of Color Sorting Algorithm Based On Support Vector Machine

With the continuous development of machine vision, the color selectionalgorithm also take a great deal of progress has been extended to many fields, andmade a lot of research., because many theories continue to mature and perfect. In thispaper, it is the rice particles automatic detection of damage using the digital imageprocessing technology. This method can be a rapid and effective identification of riceappearance and improves the quality of the detection of a grain of rice. Color sortingof other granular materials can also be the application of the proposed algorithm.This paper first introduces the concept of the structural elements ofmathematical morphology, describes the expansion and corrosion, opening andclosing operation, the top-hat transform is defined. Through the introduction of thebasic wavelet theory, it leads to the wavelet packet theory and its decompositionprocess to do a simple derivation on the basis of the understanding of wavelet theory.After study of the morphology and wavelet theory, we have a detailed andcomprehensive understanding of the characteristics of them.The image acquisition device is designed for image processing on a PC in orderto obtain the ideal image of the rice. This device selects the area CCD sensor imageand the use of LED lamps as a light source. The rice on the CCD sensor images weremade by morphological edge detection, wavelet packet edge detection, the edgedetection of morphology and wavelet packet. It can be seen by the results obtainedcompare three different. Morphological edge detection produces distortion, waveletpacket inspection is a marked improvement for more details, the result of thecombination of morphology and wavelet packet is very well.The support vector machine is used as a classification algorithm to achieve theeffective recognition of the rice particles. Firstly it introduces the basic theory ofsupport vector machines and the LIBSVM package, and then extract thecharacteristic parameters of the rice shape and rice chalkiness. The obtained data are converted to the required data format, are trained to find the optimal parameters ofthe SVM model. Finally, the model is builded according to the the optimalparameters. The model was tested using test data. It can be seen from theclassification results that the classification of the classifier is97.2%.