YOLOv8-FA: Multimodal Fusion Image Object Detection Algorithm

To address the issues of weak information expression capability of single-modal images and the limited receptive field of traditional object detection algorithms, we propose the YOLOv8-FA algorithm based on the YOLOv8 algorithm. In the backbone feature extraction network, the FasterNetBlock module is introduced to reduce the redundancy of model feature channels. Additionally, a depthwise separable convolution module is incorporated to enhance the receptive field of the network model's feature extraction. The loss function is improved to WIou to enhance the network's adaptive adjustment capability to the quality of input images. The training dataset is obtained by fusing visible light and infrared images based on the LLVIP dataset. Subjective and objective experimental results indicate that the proposed algorithm effectively improves the model's detection performance, reduces false detections and missed detections, and achieves object detection tasks in multimodal fused images.