Research on visible and infrared image fusion network based on CNN-transformer

A single sensor on a drone is insufficient to meet task requirements under various lighting conditions. By utilizing multisource sensors to fuse visible and infrared images, the operational capability of drones in complex environments is significantly enhanced. Through analyzing the characteristics of Convolutional Neural Networks (CNNs) and Transformer models in the fusion process, a network architecture is designed that combines a multi-scale attention differential CNN module with a wavelet pooling Transformer module. This network architecture enhances the local feature extraction capability while increasing the model's ability to capture global information, resulting in fused images rich in key information. This approach addresses the limitations of existing models in local feature extraction and context information interaction, demonstrating superior performance compared to advanced methods in image fusion benchmark tests.