Abstract: The existing deep neural network steganography detection technology was mainly proposed for digital image steganography． Since there are great differences between image steganography and video steganography，the deep neural networks designed for image steganalysis cannot be simply extended to video steganalysis． Therefore，a motion vector-modification video steganography was taken as the target example and a deep neural network for video steganalysis was proposed based on state-of-the-art image steganalysis network SＲNet． A data input matrix which can well reflect the steganographic modification of motion vectors was also introduced． Experimental results demon-strate that the proposed method outperforms two traditional video steganalysis algorithms in detection accuracy for low and middle-bitrate videos，and behaves well for videos at different bitrates．

Key words: video steganalysis, neural networks, motion vector, data input matrix, two-input network