Infrared ship detection in challenging maritime environments: A YOLOv12-based robustness and benchmark evaluation

Infrared imaging plays an important role in maritime surveillance, especially under low-light, nighttime and visually degraded sea-surface conditions. However, ship detection in infrared imagery remains challenging because of blurred target contours, weak thermal signatures, dynamic maritime backgrounds and significant scale variations among targets. To address these practical difficulties, this study develops a YOLOv12-based infrared ship detection framework for challenging maritime environments and establishes a task-oriented infrared maritime dataset covering seven representative ship categories and diverse scenarios, including nighttime operations, reflective sea surfaces, partial occlusions and dynamic wave interference. The dataset was reorganized from publicly available infrared maritime image resources and further completed through manual annotation, verification, preprocessing and augmentation. To improve robustness and generalization, geometric augmentation, class-rebalancing and iterative label refinement were applied during training. Extensive experiments were conducted under a unified training protocol to evaluate the applicability of the YOLOv12 series in infrared maritime ship detection. The final model achieved a precision of 0.9174, recall of 0.8889, F1 score of 0.9029, mAP 50 of 0.9276 and mAP 50–95 of 0.8013, while the convergence behaviors of Box_loss, Cls_loss and Dfl_loss confirmed stable optimization. Beyond standard quantitative evaluation, the study further investigated application performance in long-range small-object detection, multitarget classification, dynamic maritime scenarios and comparative detector analysis across different detection paradigms. In addition, distance-based attention heatmap analysis was conducted for nighttime infrared ship detection to provide interpretability support for model behavior. The results demonstrate that the proposed YOLOv12-based pipeline provides a reliable and practical solution for infrared ship detection in complex maritime environments with promising potential for future deployment in intelligent maritime monitoring systems.