In order to improve the video compression capability in railway emergency communication, a video compression algorithm based on service oriented architecture (SOA) is proposed. A railway emergency communication model based on SOA is constructed, and the original data of AI video is compressed by principal component analysis (PCA) algorithm for the first time, and the valuable feature vectors are retained. An AI video image which is compress for that first time is decomposed into a plurality of wavelet coefficient by adopting a set partitioning in hierarchical trees (SPIHT) compression algorithm, and the AI video image is output to a compressed AI video bit stream through Huffman code after the wavelet coefficients are compressed, The AI video image is reconstruct after bee compressed by that inverse Huffman coding and the inverse SPIHT compression algorithm, and the AI video after bee compressed is viewed through the application layer under the support of the information service layer and the SOA architecture layer. Experimental results show that the average video compression ratio of the proposed algorithm is 48.34, under different compression ratios, the SSIM values of AI video images compressed by the proposed algorithm are all above 0.955, and the PSNR values are all above 37.13. The proposed algorithm can effectively improve the quality of AI video images and has strong compression capability.