Scene Recognition Technology for Unmanned Surface Vehicles Based on Electro-Optical Image Fusion

Unmanned Surface Vehicles (USVs) have important applications in fields such as marine monitoring, military reconnaissance, and autonomous navigation. However, their scene recognition capabilities are limited by factors such as illumination changes and target occlusion in complex marine environments. This paper utilizes the infrared and visible light dual imaging channels and integrates them with a segmentation-based visual attention module. We propose a deep learning-based infrared and visible light fusion algorithm and a scene classification algorithm based on visual attention. Through an improved feature extraction and fusion strategy, we fuse the multimodal information of visible light and infrared images, thereby compensating for the limitations of single-modality images in different environments and enhancing their perception abilities in complex scenarios. The visual attention mechanism enables the model to focus on key information in the images, thereby improving both the accuracy and robustness of the classification task. Experiments demonstrate that on our self-constructed dataset of infrared and visible light image pairs, this method achieves a recognition accuracy of 93.3% for three typical scenes namely, open water, shallow water areas, and obstacle detection. Furthermore, it exhibits satisfactory performance under disturbances such as vibrations, rain, fog, and light/shadow variations. Visualization results confirm the role of visual attention. The proposed optoelectronic fusion method significantly enhances the recognition capability of USVs in complex environments, providing theoretical and technical foundations for the development of intelligent USV systems.