Scholarlink

<p>To address the inefficiency and high labor intensity associated with the manual measurement of growth characteristic parameters in Micropterus salmoides, this study proposes an automated non-contact detection system based on DeepLabCut. An RGB-D image acquisition system was developed using an Intel RealSense D415 depth camera, and seven neural network models, including ResNet and MobileNet series, were trained for growth feature point detection in images. Through comparative performance evaluation, MobileNet_v2_0.35 was selected as the backbone neural network for feature point identification in DeepLabCut. Subsequently, the detected key points were projected onto a 3D point cloud using intrinsic camera parameters. A combination of point cloud slicing, B-spline curve fitting, and Catmull-Rom curve interpolation was employed to accurately compute ten growth characteristic parameters of Micropterus salmoides. To validate the feasibility of the proposed method, experiments were conducted on 30 specimens of Micropterus salmoides. When compared to manual measurements, the system achieved mean absolute errors (MAE) of 7.04&#xa0;mm, 14.87&#xa0;mm, 2.00&#xa0;mm, 2.10&#xa0;mm, 1.61&#xa0;mm, 3.04&#xa0;mm, 1.11&#xa0;mm, and 1.24&#xa0;mm for total length, body length, head length, caudal peduncle length, head height, body height, caudal peduncle height, and body width, respectively. The corresponding mean absolute percentage errors (MAPE) were 2.49%, 6.19%, 2.86%, 5.47%, 3.01%, 4.16%, 4.29%, and 3.68%, indicating that the proposed system is effective for non-contact detection of growth parameters in Micropterus salmoides. The system can be applied in real-world aquaculture practices, including fish growth monitoring, health assessment, and precision feeding management, thereby supporting the development of intelligent aquaculture.</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

Automatic measurement system for growth characteristic parameters of micropterus salmoides based on DeepLabCut

  • 期刊论文

  • 发表日期：

    2025年6月9日

查看全文
  • Hongcheng Li,
  • Zhijie Xu,
  • Zhiyou Niu,
  • Zhaoxia Liu

摘要

To address the inefficiency and high labor intensity associated with the manual measurement of growth characteristic parameters in Micropterus salmoides, this study proposes an automated non-contact detection system based on DeepLabCut. An RGB-D image acquisition system was developed using an Intel RealSense D415 depth camera, and seven neural network models, including ResNet and MobileNet series, were trained for growth feature point detection in images. Through comparative performance evaluation, MobileNet_v2_0.35 was selected as the backbone neural network for feature point identification in DeepLabCut. Subsequently, the detected key points were projected onto a 3D point cloud using intrinsic camera parameters. A combination of point cloud slicing, B-spline curve fitting, and Catmull-Rom curve interpolation was employed to accurately compute ten growth characteristic parameters of Micropterus salmoides. To validate the feasibility of the proposed method, experiments were conducted on 30 specimens of Micropterus salmoides. When compared to manual measurements, the system achieved mean absolute errors (MAE) of 7.04 mm, 14.87 mm, 2.00 mm, 2.10 mm, 1.61 mm, 3.04 mm, 1.11 mm, and 1.24 mm for total length, body length, head length, caudal peduncle length, head height, body height, caudal peduncle height, and body width, respectively. The corresponding mean absolute percentage errors (MAPE) were 2.49%, 6.19%, 2.86%, 5.47%, 3.01%, 4.16%, 4.29%, and 3.68%, indicating that the proposed system is effective for non-contact detection of growth parameters in Micropterus salmoides. The system can be applied in real-world aquaculture practices, including fish growth monitoring, health assessment, and precision feeding management, thereby supporting the development of intelligent aquaculture.