Study On Feature Extraction Of Maize Seeds And Its Application Based On Computer Vision

The study that is feature extraction of maize seeds and its application based on computer vision is one part of the task--"Study on automatic detection technology of agricultural products based on computer vision" which is imbursed by science and technology program of ShanDong Province. The feature extraction system of maize seeds based on computer vision was established. First of all, the algorithms of image preprocessing were chosen. Then morphologic features of maize seeds were extracted, and lastly the identification of maize seeds using MATLAB and Neural Network with these morphologic features was studied primarily.1. The morphology and optical features of maize seeds were analyzed. Computer vision hardware system used to extract the features of maize seeds were designed. The system has simple configuration, easy processing, and wide application. It can be not only applied in the feature extraction of maize seeds, but also provided certain reference for the detection system of other crops such as wheat, soybean, peanut and so on.2. Series of image preprocessing, such as geometric transformations, color transformation, image enhancement, morphological processing, image segmentation and so on were analyzed and compared. The scheme of image preprocessing used to extract the features of maize seeds were designed. Experiment shows that the image preprocessing scheme can effectively satisfy the request of image which called for image feature extraction latterly.3. Multi-contour tracing algorithm based on contour mark was proposed. The algorithm can be fit well for the shape and contour of maize seeds, and it can trace all the contours maize seeds with one scan, So it has high execution efficiency.4. A new tip location algorithm of maize seeds based on multi-target feature measurement was put forward. Location parameters were optimized using areas and perimeters of maize seeds. The algorithm has strong adaptability because it wasn't affected by images size and orientation of maize seeds. Tip locations were computed with 93% success ratio.5. 22 morphology features of maize seeds were defined and extracted. The algorithm can extract features of multiple maize seeds at the same time, the efficiency of maize seeds feature extraction has been greatly improved.6. Structural characteristics and design parameter of BP artificial neural network were analyzed. BP artificial neural network facing MATLAB was designed. Parameters of the neural network such as training aim, network structure and transfer function were optimized.7. The feature selection was optimized by test. 14 morphology feature parameters of maize seeds were chosen as input vectors. They were Perimeter, Long axis, Short axis, Equivalent diameter, Figure parameter, Roundness, Elongation, Compactness, Green mean values, Bed mean values, Lightness mean values, Red mean square deviation, Green mean square deviation and Lightness mean square deviation. The recognition rate of LuDan 981,NongDa 108 and ShanNong 96 can be over 92% under our experiment conditions.