Scholarlink

<p>To have a quick detection in underwater environment, 3D detection algorithm has developed very fast, but due to limited computing power in underwater robots the detection algorithm has to be light. To address this, this paper presents a novel 3D target recognition method, Lightweight Object Detection-Classification and Regression Trees (LiteDet-CART), specifically tailored for underwater target classification and optimized for lightweight devices. The method comprises two primary components: LiteDet, a lightweight feature extraction network built upon a feature pyramid structure, and a CART decision tree module that evaluates and integrates extracted features for precise target classification. In experiment on the URPC dataset and self-conducted dataset, LiteDet-CART shows good strong performance, having 49 FPS better than YOLOv5’s 36 FPS, 1.93% lower mAP with 85% fewer parameters.</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

Real-time 3D underwater target recognition using a lightweight LiteDet-CART network

  • 期刊论文

  • 发表日期：

    2025年10月14日

查看全文
  • Zhijie Tang,
  • Jiajun Chen,
  • Tian Guo,
  • Ke Meng

摘要

To have a quick detection in underwater environment, 3D detection algorithm has developed very fast, but due to limited computing power in underwater robots the detection algorithm has to be light. To address this, this paper presents a novel 3D target recognition method, Lightweight Object Detection-Classification and Regression Trees (LiteDet-CART), specifically tailored for underwater target classification and optimized for lightweight devices. The method comprises two primary components: LiteDet, a lightweight feature extraction network built upon a feature pyramid structure, and a CART decision tree module that evaluates and integrates extracted features for precise target classification. In experiment on the URPC dataset and self-conducted dataset, LiteDet-CART shows good strong performance, having 49 FPS better than YOLOv5’s 36 FPS, 1.93% lower mAP with 85% fewer parameters.