您好，欢迎访问北京市农林科学院 机构知识库！
北京市农林科学院机构知识库

全部

  • 全部
  • 标题
  • 学者
  • 机构
  • 关键词

Semi-supervised fish school density estimation and counting network in recirculating aquaculture systems based on adaptive density proxy

文献类型： 外文期刊

作者： Zhu, Kaijie 1 ; Yang, Xinting 1 ; Yang, Caiwei 1 ; Fu, Tingting 1 ; Ma, Pingchuan 1 ; Hu, Weichen 1 ; Zhou, Chao 1 ;

作者机构： 1.Beijing Acad Agr & Forestry Sci, Res Ctr Informat Technol, Beijing 100097, Peoples R China

2.Natl Engn Lab Agriprod Qual Traceabil, Beijing 100097, Peoples R China

3.Natl Engn Res Ctr Informat Technol Agr, Beijing 100097, Peoples R China

4.Guangxi Univ, Sch Mech Engn, Nanning 530004, Peoples R China

关键词： Fish school density estimation; Aquaculture; Machine vision; Adaptive density proxy

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：8.9； 五年影响因子：9.3 ）

ISSN： 0168-1699

年卷期： 2025 年 230 卷

页码：

收录情况： SCI

摘要： In aquaculture, the precise quantification of fish school density holds importance for the optimization of culture management. Nevertheless, the fish occlusion, their rapid movement, as well as the size variations attributable to fluctuating distances, poses great challenges to density estimation. To overcome the above issues, this study introduces a novel semi-supervised detection network, termed FNSE (Fish Net Semi-Supervised Estimator). Specifically, the network initially employs the VGG19 backbone network to extract the global features. Subsequently, the network employs a density-proxy-guided adaptive semi-supervised fish counting network to connect different image regions, so as to obtain reliable supervision signals. Furthermore, the network incorporates a regression head with a transformer. Which combines learnable regional and local regularization. It employs attention mechanisms for enhanced the processing ability. This addresses the considerable size variations in fish caused by the change of observation distance, thereby further refining foreground features. The experimental result demonstrates that, with only 40 % of the data labeled, the FNSE network attains an accuracy of 95.81 %, alongside a Mean Absolute Error (MAE) of 7.23 and a Root Mean Square Error (RMSE) of 11.70. In comparison to DACount and MRCount, the FNSE network reduces the MAE from 9.49 and 10.27 to 7.23, signifying a substantial reduction in the MAE by approximately 24 % and 30 %, respectively, and concurrently surpassing other fully supervised methods. Therefore, the proposed FNSE can be effectively utilized for fish counting and density estimation in aquaculture and provides useful input for the development management system.

  • 相关文献
作者其他论文 更多>>
置顶
购物车
反馈

© 京ICP备09089781号-29 主办单位：北京市农林科学院

学术资源： 中国农业科学院 农业专业知识服务系统 农科联盟资源共建共享平台 中国农业科技文献与信息服务平台 中国农业期刊集成服务平台