Remote Sensing Image Compression Using ResNet and SEBlock
Han Wang, Qiang Zhang, Jun Fa Wu · 2024
This paper presents a remote sensing image compression method based on ResNet and the Squeeze-and-Excitation Block (SEBlock) to optimize bandwidth usage during satellite-to-ground transmission. By employing the ResNet model in the encoder and incorporating SEBlock to enhance feature representation, our model significantly reduces data size while preserving image quality. Experimental results demonstrate that the proposed method achieves an excellent trade-off between compression efficiency and image quality, with PSNR reaching 38.02 dB and SSIM reaching 0.9851. We tested scenarios with 2 to 5 ResNet layers, showing consistent performance improvements, with fewer layers resulting in higher PSNR and SSIM values. This approach offers a novel solution for efficient transmission of remote sensing imagery, suitable for high-resolution and bandwidth-limited satellite communications.