The result of feature extraction on multi-spectral image is related to the complex degree of target recognition algorithm directly，and also has deep influence on the performance of target recognition. On the aspect of algebra feature methods，the basic principles of 1DPCA，2DPCA，1DSVD，and 2DSVD are studied，then they are used as the feature extraction part of image target recognition algorithm. Whether is suitable for the multispectral image feature extraction is verified through the size of the recognition rate. The experiment results show that，① the feature of PCA and SVD on infrared image has better recognition results than that of visible image; ②when training samples are classified，the improvement of recognition performance based on the feature of PCA and SVD is not obvious; ③under the condition of little training sample，the features of SVD reconstructing image，2DSVD and 1DPCA have good recognition performance.