Abstract:

The aim of this study is to improve the accuracy and efficiency of the on-line classification of peanut appearance defects. Preprocessing of RGB images of intact, damaged and moldy peanut kernels was conducted using mean shift method, grayscale processing and threshold segmentation. The research extracted 6 color eigenvalues of the first and second moments of the H, S and V components in the HSV color space of the peanut image, and four texture feature values, including energy, entropy, contrast and deficit moment, based on gray level co-occurrence matrix method. Then, we constructed a feature model combined of color and texture vectors. Finally, peanuts were detected and classified by BP neural network and SVM classifier, respectively. The results showed that the overall recognition accuracy of peanut kernels was 96.67% by BP neural network and 97.22% by SVM classifier, and the latter is better than the former. The recognition time of BP and SVM was 2.5 s and 1.1 s, respectively, and SVM classifier also had better recognition efficiency. Considering the accuracy and efficiency of recognition, the SVM classifier model is preferred in distinguish peanut kernels.

Key words: peanut kernels, computer vision, color features, texture features, BP neural network, SVM