A Lightweight YOLOv8 Object Detection Method Based on Multimodal Fusion of Infrared and Visible Light

To enhance the accuracy and adaptability of target detection in complex environments, this paper proposes a lightweight YOLOv8 detection method based on multimodal fusion of infrared and visible light. This method constructs a dualbranch structure for extracting semantic and structural information, and designs a cross-modal feature fusion module (CMFM), combined with channel attention mechanism to improve feature expression ability. To achieve lightweight deployment, MobileOneNet is adopted to replace the YOLOv8 backbone network, and the GSConv module is introduced in the Neck to reduce computational load and complexity. Furthermore, in the design of the loss function, this paper incorporates the combination strategy of Varifocal Loss and CIoU Loss, respectively optimizing the classification and regression sub-tasks, thereby enhancing the accuracy of the model. Experiments are verified on the LLVIP multimodal dataset, and the results show that while the mAP reaches $90.3 \%$, the model size is compressed to 6.7 MB, and the frame rate is increased to 84.58 FPS, outperforming mainstream detection methods. This method provides an efficient and deployable solution for intelligent perception tasks in night or harsh conditions.