A multispectral image compression and encryption algorithm that combines chaos, wavelet transform and KL transform is proposed for solving the security problem of multi-spectral image compression and transmission. Firstly, the K-means clustering scheme is used to cluster multi-spectral images into common pixels, and the performance of the algorithm is optimized by selecting the appropriate K value, and it is convenient for subsequent processing. Secondly, the multispectral image is clustered into general pixels, we will perform a two-dimensional discrete 9/7 wavelet transform on the general pixels, and then perform Arnold transform and encryption processing on the transformed coefficients to eliminate most of the spatial redundancy of the multispectral image and reduce the block effect of the compression process. Next, to eliminate residual spatial redundancy and spectral redundancy, the generated wavelet coefficients are performed by KL transform. Finally, differential pulse filters are used to encode the coefficients, and Tent mapping is used to implement confusion diffusion encryption on the code stream. Through experiments, it can be known that the information entropy of this algorithm reaches 11.794 3 (selecting 12-bit multispectral images), and the information entropy is closer to the maximum value of 12, which is better than the existing algorithm and can better hide the original image features. The NPCR and UACI are respectively 99.81% and 34.19, which are better than the existing other algorithms, which can better resist differential attacks. The output bit-stream change rate is maintained between 47.62%~47.71%, and the ciphertext bitstream change rate is maintained between 47.45%~47.52%, so this algorithm has good key sensitivity; In the range of 4∶1~32∶1, the system PSNR is above 42 dB, which has high compression performance. Within the range of 4∶1~32∶1, this compression algorithm achieves a very high peak signal-to-noise ratio, which is better than the existing compression algorithm. When the normal working compression ratio is 16∶1, it is better than the existing compression algorithm. The ratio is improved by more than 0.64 dB. In order to further verify the compression performance of the algorithm in the case of a high compression ratio, this paper tested the system’s signal-to-noise ratio of 31.28 when the compression ratio is 128∶1. The reconstructed image is clearer at this time, which is more than 1dB better than the existing algorithm. It can be seen that this algorithm is feasible and particularly suitable for occasions which require a high compression ratio and has a good effect in terms of spectrum fidelity.