Dilated Separable Convolution network for image steganalysis

With the continuous development of convolutional neural networks (CNNs), new techniques based on convolutional neural networks continue to emerge, such as dilated convolution and depthwise separable convolution, which have played a significant role in deep learning and provided ideas for new model design. This article proposes a new image steganalysis method that combines dilated convolution with channel convolution, resulting in a novel dilated separable convolution module (DS). This hybrid model utilizes the advantages of the two aforementioned techniques to improve the performance of image steganalysis tasks. Firstly, in the preprocessing stage, we use high pass filters (HPF) and channel attention mechanisms to enhance relevant features and filter out redundant information. Next, we use the designed dilated separable convolution for further feature extraction. Extended convolution effectively expands the receptive field, enabling it to capture a larger range of steganographic features and reduce computational complexity, while separable convolution improves efficiency by separating spatial and channel operations. In addition, to improve the performance and robustness of the model, we introduce residual connections to better aggregate features from different layers, promote information flow, and alleviate the problem of gradient vanishing. In order to maintain the lightweight nature of our network, we have adopted a fully connected (FC) layer. Using global average pooling operation before the FC layer to generate multi view features further improves detection performance. The experimental results show that compared with the previous classical image steganalysis network model, our model has significantly improved in accuracy.