Scholarlink

<p>Object detection plays a pivotal role in computer vision, particularly in marine applications such as ocean monitoring, vessel recognition, and marine life tracking. To improve detection accuracy in complex marine environments, this paper introduces a novel object detection model, Boat-YOLO, built upon YOLOv12n. The model incorporates a self-developed branch convolution module, EIEStem, which integrates a SobelConv branch for edge information extraction with a convolution branch for spatial information extraction. In addition, we propose the A2C2f-Dual Frequency Aggregation FFN-DynamicTanh-Mona (A2C2f-DDM) feature fusion and attention module, which enhances the model’s ability to capture both spatial and frequency-domain information. To further improve precision, Boat-YOLO employs a Recalibrated Feature Pyramid Network (ReCalibrationFPN-P2345) with additional detection heads and integrates the lightweight and efficient Lightweight Shared Detail-Enhanced convolution Convolutional Detection Head (LSDECD) detection head. Experimental results demonstrate that Boat-YOLO achieves an accuracy of 90.39%, recall of 81.50%, F1 score of 85.62%, mAP@50 of 87.85%, and mAP@50–95 of 53.52%, while reducing the model file size to 5.2 MB, parameters to 2.51 million, and GFLOPs to 6.1. Compared with YOLOv12n, Boat-YOLO improves accuracy by 12.63%, F1 score by 10.22%, and mAP@50 by 8.69%, demonstrating outstanding performance in object detection tasks. Furthermore, experiments validate the robustness and generalization capability of Boat-YOLO. Compared with other state-of-the-art methods, Boat-YOLO consistently outperforms across multiple metrics, confirming its effectiveness and superiority in marine object detection.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Boat-YOLO: Efficient Inspection of Marine Vessels based on the improved YOLOv12n framework

  • 期刊论文

  • 发表日期：

    2025年10月25日

查看全文
  • Long Xu,
  • Shun An,
  • Qingyang Wang,
  • Longjin Wang

摘要

Object detection plays a pivotal role in computer vision, particularly in marine applications such as ocean monitoring, vessel recognition, and marine life tracking. To improve detection accuracy in complex marine environments, this paper introduces a novel object detection model, Boat-YOLO, built upon YOLOv12n. The model incorporates a self-developed branch convolution module, EIEStem, which integrates a SobelConv branch for edge information extraction with a convolution branch for spatial information extraction. In addition, we propose the A2C2f-Dual Frequency Aggregation FFN-DynamicTanh-Mona (A2C2f-DDM) feature fusion and attention module, which enhances the model’s ability to capture both spatial and frequency-domain information. To further improve precision, Boat-YOLO employs a Recalibrated Feature Pyramid Network (ReCalibrationFPN-P2345) with additional detection heads and integrates the lightweight and efficient Lightweight Shared Detail-Enhanced convolution Convolutional Detection Head (LSDECD) detection head. Experimental results demonstrate that Boat-YOLO achieves an accuracy of 90.39%, recall of 81.50%, F1 score of 85.62%, mAP@50 of 87.85%, and mAP@50–95 of 53.52%, while reducing the model file size to 5.2 MB, parameters to 2.51 million, and GFLOPs to 6.1. Compared with YOLOv12n, Boat-YOLO improves accuracy by 12.63%, F1 score by 10.22%, and mAP@50 by 8.69%, demonstrating outstanding performance in object detection tasks. Furthermore, experiments validate the robustness and generalization capability of Boat-YOLO. Compared with other state-of-the-art methods, Boat-YOLO consistently outperforms across multiple metrics, confirming its effectiveness and superiority in marine object detection.