AISAR
Artificial Intelligence for Synthetic Aperture Radar
AISAR is a research project aimed at revolutionizing the processing of Synthetic Aperture Radar (SAR) data directly onboard satellites.Traditionally, SAR image processing is performed on the ground, requiring satellites to downlink large volumes of raw data, leading to high communication costs and limiting the effectiveness of near–real-time applications. AISAR addresses these challenges by integrating artificial intelligence–based SAR focusing and compression algorithms directly onboard the satellite. This approach significantly reduces data transmission requirements while preserving high image quality.
By enabling intelligent onboard processing, AISAR enhances satellite efficiency, reduces dependence on ground-based infrastructure, and supports faster, more responsive applications in environmental monitoring, security, and emergency management.
Achievements:
SAR Focusing Algorithm Development: A hybrid SAR focusing algorithm has been developed, combining traditional range compression with deep learning–based azimuth compression. This approach reduces onboard computational load while preserving image quality and enables efficient SAR data processing directly onboard space-grade embedded devices.
Onboard Compression with CAE: A Convolutional Autoencoder (CAE) has been designed for onboard data compression. The model learns compact latent representations and reconstructs SAR images with high perceptual quality, achieving a compression ratio of 8. Training and evaluation were performed using SSIM-based loss functions, with performance assessed through SSIM and PSNR metrics.
Onboard Ship Detection:
A deep learning–based model has been developed for ship detection directly onboard the satellite, using SAR data focused through a DL-based approach. The model enables near–real-time ship localization in orbit, supporting applications such as maritime surveillance and rapid response to vessel activities.
