Lightweight Bit-Depth Recovery Network for Gaofen Satellite Multispectral Image Compression

Abstract

The rapid increase in satellite imagery from Gaofen satellites has created an urgent need for efficient compression methods. Existing techniques are significantly limited in compression ratio when high reconstruction quality must be preserved, particularly with 16-bit multispectral images. In this article, we introduce a novel compression framework for Gaofen satellite multispectral images that leverages a lightweight bit-depth recovery network. Our approach divides the image into the most significant bits (MSB) image and the least significant bits (LSB) image. The MSB image is compressed losslessly using a conventional codec to preserve essential information. The LSB image, on the other hand, is compressed effectively by overfitting a lightweight neural network to predict the LSB image during encoding. The optimized network parameters are utilized as side information to ensure precise bit-depth recovery during decoding. Experimental results on three types of Gaofen multispectral images demonstrate that our approach significantly outperforms state-of-the-art methods, effectively addressing the demand for efficient Gaofen satellite image compression while preserving high reconstruction quality.

Keywords

• Bit-depth recovery
• Gaofen satellite multispectral image compression
• high-quality reconstruction
• lightweight neural network