Research on infrared ship detection algorithm based on deep learning

In order to solve the issue of missing detection of small targets and the severe occlusion of ships in the current methods for detecting infrared ships; the present work introduces an infrared vessel detection method called YOLOv8-OFW, which is an advancement of the YOLOv8 algorithm. Initially, this study incorporates the concept of full-dimensional dynamic convolution, enhancing C2f with a multi-dimensional attention mechanism, and improves it into the C2f-ODConv module, which boosts the model’s ability to extract features. Secondly, the algorithm optimizes the neck network architecture, adds BiFPN and fuses the P2 small target detection layer to better integrate low-level and high-level features. Ultimately, the WIOUv2 loss function is employed to enhance the model’s collective efficacy. The algorithm was tested on the infrared ship dataset, and the AP was increased by 1.5% and the mAP by 1.8%. Experimental results demonstrate that the enhanced algorithm significantly outperforms the initial version, highlighting the utility of the enhancements made.