Low-Light Image Enhancement Method for Consistency Judgment of Multi-view Multi-target for UAV Swarm

To address the challenges posed by low-light images in multi-drone detection tasks, which lead to significant differences in target image features across different perspectives and poor accuracy in target association under multiple viewpoints, this paper proposes an enhanced network architecture based on MIRNet for low-light image enhancement. The goal is to improve the stability of target features in multi-view scenarios. By incorporating Sobel convolution to capture target edges and texture features, and by designing a residual contextual block combined with an attention mechanism and a pixel loss function that emphasizes the central region, the model not only acquires richer features but also focuses more on the central region's target characteristics. Experimental results demonstrate that the proposed approach effectively enhances image brightness, contrast, and target texture, exhibiting substantial advantages in both objective and subjective evaluation metrics.