Steganalysis, the process of finding secret information within JPEG images. It plays a vital role in digital forensics and security. JPEG steganalysis uses various techniques to identify concealed information, mostly using statistical methods. However, due to advanced steganographic algorithm the JPEG steganalysis techniques will not be able to identify the secret information within JPEG images. Therefore, Adaptive JPEG steganalysis techniques are used. Adaptive JPEG steganalysis techniques and transfer learning provide tailored detection procedures, however, these approaches encounter difficulties in achieving model generalization and stability across various operational situations. In this paper, we propose Siamese Augmented Network (SAuGNet) an innovative approach for improving the efficiency of JPEG steganalysis by leveraging the fundamental concept of Siamese CNN-based architecture in combination with self-augmented convolution. This design improves the extraction and analysis of global features such as statistical features and transform coefficients across different parts of the image using self-augmented convolution. Experimental results show that the model achieves a detection performance that is competitive with existing steganalysis models. We conducted experiments on the BOSSBase and BOWS2 datasets, selecting samples randomly from each. These experiments were conducted across distinct scenarios, incorporating JPEG quality factors such as 75, 85, and 95, and embedding rates (bpnzAC) ranges from 0.1 to 0.5. The effectiveness of SAuGNet demonstrates superior detection capabilities compared to current steganalysis methods.