Abstract: A multiscale image fusion network with texture and edge feature enhancement was developed to overcome the problems imposed by the large number of model parameters, low efficiency, insufficient feature extraction, and loss of detail in texture and edge features when using infrared and visible image fusion algorithms. First, an edge enhancement operator and edge-guided attention module were developed to retain more edge texture information and attenuate bad artifacts. Second, deep separable convolution and channel shuffling operations were integrated into the multiscale channel attention module to enhance the interaction between channels and reduce the time complexity. Sequential attention feature fusion was changed to double-branch fusion to enhance the multiscale information interaction ability. Finally, a convolutional multilayer perceptron was developed to reduce the number of network parameters while providing location information and enhancing the representation ability by replacing the fully connected layer and adding a depth-separable convolution and gating mechanism. In fusion experiments on the TNO, MSRS, and road scene datasets, the SD, SF, MI, SCD, and SSIM indices of the proposed method increased by 4.7%, 7.9%, 2.9%, 1.7%, and 15.5%, respectively, compared with the results of seven other methods. The edge texture of the fused image was clear, the contrast was appropriate, and the fusion target was clearly visible.