This paper presents an improved object detection model, GT-YOLO, for nearshore infrared ship detection based on YOLOv5s. The model addresses the challenges of low-resolution infrared images and dense occlusions in coastal areas by incorporating a feature fusion module and SPD-Conv, and using Soft-NMS to enhance detection accuracy. The feature fusion module, based on a fusion attention mechanism, enhances the integration of high-level and low-level features while suppressing noise. SPD-Conv improves the detection of small targets and low-resolution images by replacing stride convolutions and pooling layers. Soft-NMS is introduced to handle dense occlusions more effectively, ensuring accurate detection even in crowded scenes. Experimental results on the Infrared Ship dataset show that GT-YOLO achieves a 1% increase in mAP0.5, a 5.7% increase in mAP0.75, and a 5% increase in mAP0.50.95 compared to the original YOLOv5s model. The improved algorithm demonstrates superior performance in detecting small and multi-scale targets, making it effective for nearshore ship detection tasks.This paper presents an improved object detection model, GT-YOLO, for nearshore infrared ship detection based on YOLOv5s. The model addresses the challenges of low-resolution infrared images and dense occlusions in coastal areas by incorporating a feature fusion module and SPD-Conv, and using Soft-NMS to enhance detection accuracy. The feature fusion module, based on a fusion attention mechanism, enhances the integration of high-level and low-level features while suppressing noise. SPD-Conv improves the detection of small targets and low-resolution images by replacing stride convolutions and pooling layers. Soft-NMS is introduced to handle dense occlusions more effectively, ensuring accurate detection even in crowded scenes. Experimental results on the Infrared Ship dataset show that GT-YOLO achieves a 1% increase in mAP0.5, a 5.7% increase in mAP0.75, and a 5% increase in mAP0.50.95 compared to the original YOLOv5s model. The improved algorithm demonstrates superior performance in detecting small and multi-scale targets, making it effective for nearshore ship detection tasks.