ALFusion: Adaptive fusion for infrared and visible images under complex lighting conditions

Hanlin Xu, Gang Liu, Yao Qian, Xiangbo Zhang, Durga Prasad Bavirisetti

Abstract: Highlights•A mixed feature module with dynamic convolution and Transformer is embedded in U-Net, enabling effective extraction.•Convolutional kernels of various sizes expand the receptive field, combined with attention to improve feature representation.•A pre-trained network generates luminance probabilities to adjust pixel loss, refining pixel ratios in the fused image.