DBSD: DUAL branches network using semantic and detail information for infrared and visible image fusion

Abstract

Infrared and visible image fusion is of great importance in all-weather target detection and surveillance work. However, the existing fusion algorithms have problems such as missing information and vague details. Aiming at the above problems, this paper proposes a dual-branch fusion network based on semantic and detail information (DBSD). The dual-branch structure is used to fully acquire the structural and textural features of multi-modal images to obtain the images with complete features and clear details. The network adds multi-receptive blocks and large kernel blocks in semantic branches and uses full-scale skip connections to achieve multi-directional and multi-scale feature extraction. Densely connected blocks are used in detail branches to improve the reusability of features. The fusion block based on channel and spatial attention mechanisms is used to integrate multi-scale semantic and detail features to enhance key features in the fusion results. Through a series of experiments conducted on three publicly available datasets to verify the superiority of DBSD.