This paper introduces a novel deep learning architecture, DenseFuse, for the fusion of infrared and visible images. Unlike traditional convolutional networks, DenseFuse combines convolutional layers, a fusion layer, and dense blocks, where each layer's output is connected to every other layer. This architecture aims to extract more useful features from source images during the encoding process. Two fusion strategies (addition and $l_1$-norm) are designed to combine these features, and the fused image is reconstructed by a decoder. Compared to existing methods, DenseFuse achieves state-of-the-art performance in both objective and subjective assessments. The paper also discusses the training process, experimental results, and comparisons with other fusion methods, demonstrating the effectiveness of DenseFuse in preserving detailed information and reducing artificial noise in the fused images.This paper introduces a novel deep learning architecture, DenseFuse, for the fusion of infrared and visible images. Unlike traditional convolutional networks, DenseFuse combines convolutional layers, a fusion layer, and dense blocks, where each layer's output is connected to every other layer. This architecture aims to extract more useful features from source images during the encoding process. Two fusion strategies (addition and $l_1$-norm) are designed to combine these features, and the fused image is reconstructed by a decoder. Compared to existing methods, DenseFuse achieves state-of-the-art performance in both objective and subjective assessments. The paper also discusses the training process, experimental results, and comparisons with other fusion methods, demonstrating the effectiveness of DenseFuse in preserving detailed information and reducing artificial noise in the fused images.