Semantic Image Encoding and Communication for Earth Observation With LEO Satellites

The substantial volume of data generated by Earth observation (EO) satellites poses a significant challenge to the limited-rate satellite-to-ground links. This paper addresses the downlink communication problem of change detection in multi-spectral satellite images for EO purposes. The proposed method is based on a cohesive strategy capable of eliminating clouds and performing semantic encoding during image processing. This approach is a manifestation of semantic communication, as it encodes vital information for the target application, in the form of changed multi-spectral pixels (MPs) to minimize energy consumption. The proposed method is based on a three-stage end-to-end scoring mechanism, which quantifies the significance of each MP before determining its transmission. Specifically, the sensing image is (1) normalized and passed through a high-performance cloud filtering via the Cloud-SLR model, (2) passed to the proposed scoring algorithm that uses Change-Net to identify MPs that have a high likelihood of being changed, compress them, and forward to the ground station, and (3) reconstructed at ground gateway based on the reference image and received data. The numerical results show the effectiveness of the proposed framework in achieving energy savings of up to 58% while upholding the transmission of high-quality data for satellite-based EO applications.

I. Introduction

Remote sensing satellites are vital for environmental monitoring as they swiftly provide comprehensive coverage of targeted regions, allowing for land use surveys, urban studies, and hazard management through image acquisition [2]. However, a major problem is that the use of high-resolution sensors results in large volumes of data, which necessitates significant communication resources and on-board data storage capacity for transmitting data to ground-based end-users. For instance, the Sentinel-2 system acquires an extensive amount of data (2.4 Terabits per day) for transmitting to the terrestrial gateway, with each surface location being captured at periodic intervals of five days [3]. The grow of Low Earth Orbit (LEO) satellite deployments in Earth Observation (EO) applications, coupled with the constrained communication capacity of LEO satellites, poses a limitation on the handling of daily generated EO data. While forwarding the complete dataset can be advantageous for accurately detecting any changes or anomalies, it is not regarded as efficient in terms of data storage and transmission capabilities. Therefore, the conventional approach of transmitting captured images to the ground for analysis and distribution may be inefficient. Thus, the vast amount of data generated by EO satellites requires novel techniques for high-spectral image processing and transmission.