Brief

Researchers have developed a novel segmentation-based detection method for multi-task spacecraft perception, addressing challenges like limited annotated data and difficult visual conditions. Their compact architecture, featuring a MobileNetV3 encoder and a U-Net-style decoder, achieved strong performance in classification, detection, and segmentation tasks. This approach ranked second in the SPARK 2026 Challenge, demonstrating the effectiveness of lightweight models for practical onboard space vision systems. AI

Researchers have developed a new knowledge distillation framework called SURGE to create more efficient SAR ship detection models. This framework transfers relational geometry from a larger teacher model to a smaller student model using a contrastive learning objective. Experiments on benchmark datasets show significant improvements in detection accuracy, with the student model even surpassing the teacher's performance in some cases. AI