Study On Apple External Character Detection And Grading Based On Computer Vision Theory

In order to improve fruits' quality and production efficiency, reduce the labor intensity , it is necessary to research on nondestructive and prompt automatic detection technology. Aiming at fruit sorting and taking apple as study object, the research investigated and developed some fundamental theories and general methods of computer vision, in addition, it put forward some new algorithms in fruit sorting. The studies are concentrated at following aspects:1. In the lower-layer image processing, super median filter method is proposed. To detect the edge of the image, a polar searching method was used. In this way, it accelerated the preprocessing and economized time for the subsequent works.2. In order to measure the size of fruit, the algorithm used in this paper oriented the stem-calyx axes firstly , then crosscut the apple in the direction normal to the axes and drew some line sections, finally confirm the size of the apple through measuring the maximal line section . In this research, a simple threshold was used to realize the size classification.3. Presented the method that it use radius sequential Fourier coefficient as shape descriptor. Put the descriptor into neural networks ,and train the networks using Levenberg-Marquardt algorithm ,then the shape of apple was graded. According to the result, the correctness rate was 90%.4. a novel fast intelligent grading method was presented, which was based on the particle swarm optimization (PSO) algorithm reappearing swarm intelligence and the neural computation technology. The main process is to acquire the hue histogram of apple surface by the computer vision technology and extract its features, then train the neural network architectures by improved PSO algorithm, finally grade the apple color with the trained network. The actual application showed that the method can achieve high precision, and get very high grading speed. In apple color grading, the application effect was very notable.5. The defects segmentation was realized by comparison between a normal hue model or a reference image and the defected image.6. A software system for apple quality detecting and grading was developed.The research brought forward new theories for further development of practical fruit sorting system based on computer vision.