In this paper, perimeter, roundness, boundary roughness, rectangularity, and center of mass displacement are selected as candidate features, the electric coal image dataset is collected, the set of feature vectors is selected based on the importance, relevance, and distinguishability of the features, and the KPCA algorithm is used to reduce the unnecessary image feature data and reduce the vector dimensions to obtain the optimal subset of the features, and the color features, such as color moments of the image, and the grayscale covariance are extracted matrix, Tamura texture features and filter these features. Then an image segmentation network CLUNS based on convolutionalized recurrent neural network is proposed. The classification and recognition results show that compared with other segmentation algorithms, the segmentation accuracy of the segmentation network proposed in this paper is 97.82%, and compared with the original CLUNS network algorithm the proposed algorithm improves the segmentation accuracy by 4.30 percentage points, and significantly reduces the loss rate in the validation set, respectively, by 10.16 The proposed algorithm has a significant advantage in running time, good generalization ability and stability, and provides a reference for subsequent quantitative image detection.
