Maritime target detection algorithm based on fusion of visible and infrared images

Unmanned surface vessel target detection plays a key role in the fields of marine scientific research and maritime border security. Accurate target detection algorithms can enhance the efficiency of marine traffic management and strengthen maritime security monitoring. In order to effectively handle the unique situations encountered in real-world scenarios, we propose a target detection algorithm based on multimodal image fusion to detect ships at sea. We introduce the GAM (global attention mechanism) attention module into the traditional DenseFuse network to fuse visible images with infrared images, effectively reducing the influence of day and night variations on detection results. Additionally, we replace some modules of the original model with the SAConv (switchable atrous convolution), enabling the network to adapt more flexibly to features of different scales. For the detection of relatively small objects, the NWD (normalized Wasserstein distance) is used to optimize the IOU value. Experimental results show that the algorithm can accurately detect and identify targets such as ships, buoys and speedboats. In typical sea surface scenes, the algorithm achieves an \(F_{1}\) $F_1$ score of 83.5% and a mAP50 of 86.8%.