Research Of Intelligent Recognition For Maize Seed Purity

Maize seed purity is important factors to affect maize yield and quality and its recognition has always been a thorny problem in seed circulation field. A method of intelligent recognition for maize seed purity was studied based on color extracted from the images of both the maize crown and the maize side was proposed for solving the problems in recognition of maize seed purity based on digital image processing technology, which has important significance for promoting maize detection technology progress,maize quality improvement,market norms and special skilled works. The mainly following work has been completed:(1) The image acquisition hardware system of maize seeds was designed based on the color features,recognition speed and identification accuracy and the image of seeds which is up to the mustard can be got by the hardware system.(2) The image preprocessing plan of maize seeds was designed. First the original maize seed image was carried on gradation processing, median filters was used to complete the image enhancement; then the iterative method was used to segment background through starting experiment by a few partition methods and comparing the experiment results. Finally, clearly binary image was obtained. Separate seeds were distilled mainly by contour labeling method and showed with color images.(3) This paper finds that the crown center colors were useful for identification of maize seed purity in addition to the lateral colors. Lateral color areas of maize seeds were extracted by the B plane, crown center region extraction was completed based on gray distribution striving corner detection. RGB model and HSV model were combined together during picking up the color characteristics. Results show that both the areas extracted through the two methods and the color characteristics picked up from the areas can meet the identification requirements.(4) The color features extracted from crown center region and lateral color region were studied. The results show that both comprehensive color features are the best. In order to get one-dimensional function, multidimensional eigenvectors were projected into one-dimensional space through applying Fisher discriminant analysis; then curve fitting was carried on the basis of one-dimensional function; finally, identification for maize seed purity was completed based on the point to the curve distance. (5) This paper chooses 9 kinds of maize seeds, such as the jundan20, nongda108, zhengdan958, ludan981, etc. as object of study, the above methods and models were applied into the actual algorithm, and were carried on experimental methods based on colors feature. It's proved by experiments that the method proposed in this paper is feasible, the lowest recognition rate is 93.4% and the average recognition rate is 96.87%.