An improved image steganalysis method based on convolutional neural network

With the continuous development of Steganalysis based on convolutional neural network, in order to obtain higher steganalysis accuracy, softpooling method is used to replace the original conventional pooling, which not only retains the advantages of conventional pooling in reducing the amount of calculation and compressing the size of feature map, but also prevents the loss of steganographic feature information. The Relu activation function is replaced by the Mish activation function. The accuracy of steganalysis is improved. The experimental results show that the accuracy of Xu and SRNET is improved by 0.53% and 1.27% respectively.