GAN-Based Fusion of Infrared and Visible Images in Dark Light Environments

In low light conditions, the fused image generated by fusion of infrared and visible images is dark and lacks detailed texture information, resulting in low target detection accuracy, based on which a target detection network for fusion of infrared and visible images in low light conditions is proposed. Firstly, the dark light perception network is designed to enhance the brightness of the dark light image; secondly, the feature extraction module of the image fusion network introduces the gradient residual dense block to strengthen the ability of the image fusion network to extract the depth features and finegrained detail features of the source image; subsequently, the loss function incorporates the intensity loss and the structural similarity loss to enhance the ability of the image fusion network to extract the pixel intensities and the structural texture details of the source image; Finally, the optimal solution of the fusion network for target detection is achieved by joint training of the image fusion network and the target detection network. The experimental results show that the fusion network in the paper improves the mutual information, average gradient, visual fidelity, gradient-based fusion performance, and structural similarity by 0.142, 0.135, 0.043, 0.067, and 0.082 compared to the other network models, and improves the average precision mean and accuracy by 2.6% and 2.5%, respectively, in the YOLOV8 detection network.