YOLO-GCV: A Lightweight Algorithm for Ship Object Detection in Complex Inland Waterway Environments

Abstract Lightweight ship detection offers the dual benefits of rapid detection and low computational cost, making it particularly advantageous for inland waterway safety monitoring. This study introduces YOLO-GCV, a lightweight ship detection algorithm based on YOLOv7-tiny. The proposed algorithm strikes an effective balance between detection accuracy and speed. First, the ELAN-Ghost lightweight module is integrated into the backbone network, while VoVGSCSP, another lightweight module, is introduced into the neck to further streamline the model structure. Coordinate convolution is utilized to enhance the model’s ability to capture the spatial features of ship targets. Furthermore, the WIoU loss function is incorporated to improve convergence during training and significantly bolster the model’s generalization capability. Experimental results indicate a 19.2 \(%\) reduction in model parameters and an 18.9$%$ decrease in GFLOPs, with mAP0.5 and mAP0.5:0.95 increasing by 0.8$%$ and 0.5$%$, respectively, over the baseline model. The model achieves a processing rate of approximately 42 images per second, meeting real-time detection requirements. This lightweight ship detection algorithm effectively addresses real-time detection needs in complex inland waterway environments and offers notable advancements in inland navigation safety.