Infrared target detection ODD-YOLO algorithm for complex backgrounds

Complex backgrounds, inconspicuous target features and many interferences are typical characteristics in infrared images. In this paper, we propose a real-time ODD-YOLO algorithm based on the YOLOv8n network. First, the Omni-Dimensional Dynamic Convolution (ODConv) is used in the backbone network to selectively replace the regular convolution in the original model, optimizing the network structure so that the model can adapt to the diversity of input samples and reducing the computational cost of the model, and the feature fusion network constructs a lightweight C2f_DASI module based on the dimension-aware selective integration module, which significantly enhances the ability to express the features of the weak target through effective interaction and selective inheritance of multiscale features. In addition, Focal-EIoU is introduced to improve the generalization capability of the model. The experimental results show that on the SIRST dataset, the ODD-YOLO algorithm significantly improves the detection accuracy to 95.4% while maintaining real-time performance (253.47FPS), the inference speed of the model meets the real-time requirement of the infrared target detection task.