Ultra-Low Complexity On-Orbit Compression for Remote Sensing Imagery via Block Modulated Imaging

TL;DR

This work introduces Block Modulated Imaging (BMI), an ultra-low-encoding-complexity approach for on-orbit remote-sensing image compression that uses optical mask modulation and a single exposure to generate measurements. Recovery is achieved via a deep unfolding BMNet decoder that employs gated 3D convolutions and a Two-Way Cross-Attention mechanism to exchange information across stages, with a final 2D refinement. The authors validate BMI across multiple datasets and tasks, demonstrate a hardware prototype, and show that BMI can outperform SPI in both encoding and decoding efficiency while maintaining robust reconstruction and acceptable downstream-task performance. The practical impact lies in enabling high-resolution, energy-efficient, onboard compression that reduces bandwidth and storage demands for satellite platforms. The work also provides a concrete hardware path and codebase to accelerate adoption and further research in optical-computing-based remote sensing compression.

Abstract

The growing field of remote sensing faces a challenge: the ever-increasing size and volume of imagery data are exceeding the storage and transmission capabilities of satellite platforms. Efficient compression of remote sensing imagery is a critical solution to alleviate these burdens on satellites. However, existing compression methods are often too computationally expensive for satellites. With the continued advancement of compressed sensing theory, single-pixel imaging emerges as a powerful tool that brings new possibilities for on-orbit image compression. However, it still suffers from prolonged imaging times and the inability to perform high-resolution imaging, hindering its practical application. This paper advances the study of compressed sensing in remote sensing image compression, proposing Block Modulated Imaging (BMI). By requiring only a single exposure, BMI significantly enhances imaging acquisition speeds. Additionally, BMI obviates the need for digital micromirror devices and surpasses limitations in image resolution. Furthermore, we propose a novel decoding network specifically designed to reconstruct images compressed under the BMI framework. Leveraging the gated 3D convolutions and promoting efficient information flow across stages through a Two-Way Cross-Attention module, our decoding network exhibits demonstrably superior reconstruction performance. Extensive experiments conducted on multiple renowned remote sensing datasets unequivocally demonstrate the efficacy of our proposed method. To further validate its practical applicability, we developed and tested a prototype of the BMI-based camera, which has shown promising potential for on-orbit image compression. The code is available at https://github.com/Johnathan218/BMNet.

Figures (9)

• Figure 1: Satellite-side optical encoding for compression and ground-based deep unfolding decoding for reconstruction.
• Figure 2: (a) BMNet processes the measurement $Y$ alongside the mask $M$, and reconstructs the original signal $X$. $\textbf{v}^{(k)}$ denotes the auxiliary variable at the $k$-th stage (see Eq. \ref{['linearpj']}). (b) The proposed gated 3D convolution block. (c) The proposed Two-Way Cross-Attention module. $\textbf{u}^{(k)}$ denotes the latent variable generated by the encoder of the $k$-th 3D U-net. $\textbf{h}_{h}^{(k)}$ denotes the hidden variable of the $k$-th stage ($\textbf{h}^{(0)}=\textbf{u}^{(0)}$).
• Figure 3: Visualization of the reconstructed images on DOTA-v1.0 at different compression ratios (Cr).
• Figure 4: Visualization of different tasks under varying compression ratios (Cr).
• Figure 5: Comprehensive performance of BMNet under varying compression ratios (Cr): PSNR, SSIM for image reconstruction (DOTA-v1.0), mAP for object detection (DOTA-v1.0) and mIoU for semantic segmentation (Vaihingen).