Soybean Quality Verification Method Based On The Computer Vision Technology

With the development of computer vision and pattern recognition, the quality verification of the crops has become a very active research field gradually, and further quality verification for the crops especial for the seeds has been needed. More and more experts and scholars have been devoting themselves to this domain, and they have made great achievements on it. The computer vision-based quality verification involves two issues of pattern recognition: the extraction of relevant characteristics of the seeds, and the right selection of the classifier.In this paper, after the pretreatment of the sample images, including the noise removal and image enhancement, the effective features were made more obvious.In the period of feature extracting, the features of color, shape and texture were extracted, the color feature through the mean pixels got from the samples was acquired; the shape feature through the circular metric which calculated by perimeter and area of the regional was obtained; the texture feature through the energy, the entropy and the moment of inertia was procured. The algorithms for each feature were elaborated, and the experiments for the algorithms of the color-based feature, the shape-based feature and the texture-based feature were carried on.A recognition method was proposed, which based on rules of the color feature, shape feature and texture feature, each of the features had two corresponding rules respectively. The recognition method based on rules had some advantages, such as quick recognition and the clear relationship of rules.With the color feature, shape feature and texture feature, the recognition method using rules to verify the quality of the soybeans were adopted. And 50 sample images had been adopted, 32 soybeans of each image, amounted to 1600 soybeans. The recognition rate reached 93.2%, which closed to the artificial recognition rate. From the result of the experiment, both the algorithm for feature extracting and the methods of pattern recognition are effective.