PAU-GAN: Object Detection Based on Infrared and Visible Image Fusion Under Low-Light Conditions

Object detection in low-light conditions poses significant challenges for autonomous systems, as traditional visible light cameras often fail to capture clear and informative images in such environments. To address this issue, we propose a novel generative adversarial network called PAU-GAN, which consists of aU-shaped generator and two patch discriminators. The generator incorporates pixel attention (PA) and convolutional mechanisms. Under the guidance of the two discriminators and the constraints of Fourier transform loss, gradient loss, and structural similarity (SSIM) loss, the generator effectively extracts features from the fused images and enhances high-frequency information, benefiting object detection. To verify the effectiveness of the proposed model, we conducted detection using advanced object detection model on our fusion results. Comparative experiments have demonstrated that the proposed model can effectively improve the detection accuracy in low-light environments.