Image Fusion Approach Based on Heterogeneous Dense Network

In this research, we propose a novel algorithm for fusing visible and infrared images using a heterogeneous dense network and CBAM, generating fused images with more texture details while preserving salient targets. Firstly, the encoder and decoder are constructed based on the heterogeneous dense network to achieve deep feature extraction and reconstruction of the source image. Secondly, to guide network optimization, we design a loss function using the salient target mask. Under the constraint of this loss function, the encoder calculates pixel and gradient losses for background areas and salient target areas, respectively, with the decoder then fusing the features of corresponding areas to reconstruct desired results. Additionally, CBAM is added to the fusion part to process infrared image features by focusing more on salient targets and reducing redundant information introduction. Finally, experiments demonstrate that our algorithm enhances detailed features while preserving target features consistent with human visual characteristics. The proposed algorithm achieves a superior fusion effect.