A Discriminative Enhancement and Selective Fusion Method for Low-Light Cross-Spectral Object Detection

Under low-light conditions, visible-spectrum images are prone to detail loss and contrast degradation, which substantially limits object detection performance. Although infrared imagery can provide complementary cues, direct fusion often introduces noise interference and thus undermines detection stability. To address this issue, this paper proposes a discriminative enhancement and selective fusion method for low-light cross-spectral object detection. Specifically, a task-oriented discriminative Retinex enhancement module is introduced at the front end to mitigate illumination interference while strengthening structural information. Meanwhile, a spectral-selective cross-scale fusion module is designed to suppress noise propagation through adaptive weighting and cross-scale interaction. In addition, mutual information loss and cross-scale consistency constraints are incorporated to enhance cross-spectral feature representation and prediction stability. Experimental results on multiple public datasets demonstrate that the proposed method can consistently improve the accuracy and robustness of object detection under low-light conditions.