STFA-YOLO: An Algorithm for Enhanced Infrared Ship Detection in Complex Marine Environments

In the field of marine surveillance, infrared detection technology is crucial for the timely localization and identification of ships, serving vital security and monitoring functions. However, detecting, classifying, and accurately locating multiple ships in infrared images presents significant challenges due to typical characteristics of infrared ship targets such as low pixel resolution, low contrast, and cluttered backgrounds. To address these challenges, we propose a novel model STFA-YOLO which based on the YOL011 model. For small targets in infrared images with complex backgrounds and low resolution, we introduce a double-branch STFA module with cross-channel interaction. This module dynamically adjusts feature by extracting both local and non-local information, enhancing the model's ability to detect weak targets. Additionally, we integrate a dedicated small target detection head to further increase the detection ability of weak targets, while employing GhostConv convolution to reduce computational complexity without sacrificing performance. Experimental results demonstrated that our model outperformed state-of-the-art methods, this method increases precision by 0.2%, mAPSO by 1.0% and recall by 1.4%, while reducing the model size by 1.16% compared to the original YOLOv11 model.