A Deep Joint Source Channel Coding Scheme with Adaptive Feature Enhancement

In this paper, a feature-enhanced deep source channel joint coding scheme called DJSCC-E is proposed, considering the varying importance of transmitted data in DJSCC schemes and the diversity of target scales for encoding and decoding images. The scheme is designed with a Weight Adjustment Module (WAM) that adjusts the weight values of the feature images based on varying Signal-to-Noise Ratio (SNR). Furthermore, a multi-scale encoder-decoder structure is proposed to optimize the encoding and decoding of target information at different scales. The experimental results demonstrate that the DJSCC-E scheme enhances the Peak Signal-to-Noise Ratio (PSNR) of the reconstructed image by approximately 0.7 dB compared to the current DJSCC scheme when the SNR of the AWGN channel falls within the range of 0 to 20 dB, while reducing the number of parameters.