SNCTD: a detector for space non-cooperative targets with low-light enhancement and small-scale feature fusion

Precise detection of components on space non-cooperative targets under low-light and small-scale conditions is critical for on-orbit servicing. To address the performance limitations of traditional detectors in space environments, an end-to-end detection framework named space non-cooperative targets detector (SNCTD) is proposed. SNCTD integrates a low-light enhancement and denoising (LLED) module, which is jointly optimized with the detection framework to improve image quality under variable illumination. Efficient feature extraction is achieved through the hybrid downsampling (HyDown) and range-constrained sample (RSample) modules while the improved cross stage partial (ICSP) module ensures robust multi-scale feature fusion with reduced computational overhead. A high-resolution dataset capturing diverse lighting and scale variations for non-cooperative targets is also introduced to support training and evaluation. Experimental results indicate that SNCTD achieves notable improvements in detection accuracy over representative state-of-the-art models, particularly for low-light and small-scale targets. With its robustness to illumination changes and computational efficiency, SNCTD is well suited for onboard visual systems in space missions.