Synthetic Aperture Radar SLC data compression using Mean-Scale Hyperprior architecture

Synthetic Aperture Radar (SAR) images are becoming increasingly important in a variety of remote sensing applications, leading to new missions with higher resolution and coverage, ultimately resulting in an ever-increasing volume of data. This burden on SAR data storage and transmission has established a serious interest in developing compression methods that can obtain higher compression ratios, while keeping complex SAR image quality to an acceptable level. In computer vision, neural network-based RGB image compression has exceeded traditional methods such as JPEG, JPEG2000 or BPG. The Mean-Scale Hyperprior network [1] is an auto-encoder based architecture exploiting the probabilistic structure in the latents to improve compression performance. Auto-encoders are architectures particularly suited for the inherent rate-distortion trade-off of data compression. They also offer an intuitive solution to the on-board image compression problem, as demonstrate for the Φ-Sat-2 mission [2]. In this work, we explore efficient SAR image compression, in this regard, we adapt the Mean-Scale Hyperprior architecture to SAR data. We use Sentinel-1 IW mode VV polarization SLC images to build a dataset of diverse scenes: urban areas, forests, mountains and water bodies in dry as well as snow/ice conditions. The central idea being to create an open-source and general dataset of SAR images, in order to compare the performance of the studied architecture with traditional codecs and baseline models, such as the work in [3]. We will experiment with latent sizes, patch size as well as different SAR data representations for the network. References  [1] D. Minnen, J. Ball ́e, and G. D. Toderici, “Joint Autoregressive and Hierarchical Priors for Learned Image Compression,” in Advances in Neural Information Processing Systems, vol. 31, Curran Associates, Inc., 2018.  [2] G. Guerrisi, F. D. Frate, and G. Schiavon, “Artificial Intelligence Based On-Board Image Compression for the Φ-Sat-2 Mission,” IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, vol. 16, pp. 8063–8075, 2023.  [3] C. Fu, B. Du, and L. Zhang, “SAR Image Compression Based on Multi-Resblock and Global Context,” IEEE Geoscience and Remote Sensing Letters, vol. 20, pp. 1–5, 2023.