Abstract: In order to realize the individual identification of cattle in the actual breeding environment of small and medium-sized pastures and achieve the purpose of precision breeding, a dictionary learning algorithm based on image multi-channel K-singular value decomposition（K-SVD） was proposed for cattle face recognition. The algorithm is combined with sparse representation theory to obtain more component details of the image through R,G,B three-channel decomposition, and divide the three-channel components into n×n grids to reconstruct the input matrix. The reconstruction matrix is sparsely represented by orthogonal matching pursuit（OMP） algorithm, the dictionary is updated by the K-SVD, and a learning dictionary corresponding to each channel is constructed. The algorithm was used to test a dataset of 400 cattle face images from 20 cattle and compared with K-SVD algorithm and SRC algorithm. The results showed that the recognition rate of the image multi-channel K-SVD algorithm was 92.9%, which was better than K-SVD algorithm and SRC algorithm in the recognition accuracy and sparse representation ability. It is shown that by combining image processing algorithms and sparse representation theory, more image features can be extracted using multi-channel information to further improve the recognition accuracy.