Remote Sensing Image Compression Method Based on Implicit Neural Representation

Abstract

Oceanography-oriented remote sensing image is characterized by being multi-source, heterogeneous, and massive. In recent years, the scale of remote sensing data is exploding, necessitating efficient compression methods for storage. Traditional algorithms have low compression ratio, poor flexibility, and are unsuitable for storing large-scale remote sensing image data. To address this, we propose a remote sensing image lossy compression and storage method based on implicit neural representation. We used an implicit neural network to learn the mapping relationship between longitude, latitude coordinates, and values. Then, we compressed the network weights using quantization. Our method can significantly reduce storage space by approximately 80% for multi-source and heterogeneous remote sensing image data. And the reconstructed quality is capable of meeting the requirements of downstream tasks related to marine AI and visualization.