Efficient SAR Vessel Detection for FPGA-Based On-Satellite Sensing

2025-07-07

AI Summary

To address the challenges of large model size, high power consumption, and difficulty in on-orbit real-time deployment for ship detection in spaceborne SAR imagery, this paper proposes a lightweight YOLOv8 architecture tailored for FPGA implementation, coupled with hardware-aware optimization techniques—including model pruning, quantization, and custom hardware mapping—enabling efficient deployment on the Kria KV260 MPSoC platform under a strict <10 W power budget. Evaluated on the large-scale, real-world SAR dataset xView3-SAR, the optimized model achieves detection and classification accuracies within approximately 2% and 3%, respectively, of the GPU-based baseline, while reducing model size by two to three orders of magnitude. To the best of our knowledge, this work represents the first demonstration of high-performance, on-orbit ship detection from diverse, large-scale SAR data on low-power satellite hardware—significantly reducing downlink latency and advancing autonomous, real-time intelligent remote sensing.

Technology Category

• Machine Learning: Hardware-aware ML
• Computer Vision: Learning & Optimization for CV
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation

Abstract

Rapid analysis of satellite data is vital for many remote sensing applications, from disaster response to environmental monitoring, but is becoming harder to achieve with the increasing volumes of data generated by modern satellites. On-satellite machine learning (ML) offers a potential solution, by reducing latency associated with transmission of these large data volumes to ground stations, but state-of-the-art models are often too large or power-hungry for satellite deployment. Vessel detection using Synthetic Aperture Radar (SAR) is a critical time-sensitive task for maritime security that exemplifies this challenge. SAR vessel detection has previously been demonstrated only by ML models that either are too large for satellite deployment, have not been developed for sufficiently low-power hardware, or have only been developed and tested on small SAR datasets that do not sufficiently represent the real-world task. Here we address this issue by developing and deploying a new efficient and highly performant SAR vessel detection model, using a customised YOLOv8 architecture specifically optimized for FPGA-based processing within common satellite power constraints (<10W). We train and evaluate our model on the largest and most diverse open SAR vessel dataset, xView3-SAR, and deploy it on a Kria KV260 MPSoC. We show that our FPGA-based model has detection and classification performance only ~2% and 3% lower than values from state-of-the-art GPU-based models, despite being two to three orders of magnitude smaller in size. This work demonstrates small yet highly performant ML models for time-critical SAR analysis, paving the way for more autonomous, responsive, and scalable Earth observation systems.

Problem

• Reducing latency in satellite data analysis for real-time applications
• Optimizing ML models for low-power FPGA-based satellite deployment
• Improving SAR vessel detection accuracy on large diverse datasets

Innovation

• Customized YOLOv8 for FPGA processing
• Optimized for under 10W satellite power
• Trained on large xView3-SAR dataset