您好，欢迎访问重庆市农业科学院 机构知识库！

A novel detection model and platform for dead juvenile fish from the perspective of multi-task

文献类型： 外文期刊

作者： Zhang, Pan 1 ; Zheng, Jishu 5 ; Gao, Lihong 5 ; Li, Ping 5 ; Long, Hanwei 5 ; Liu, Hongbo 5 ; Li, Daoliang 1 ;

作者机构： 1.China Agr Univ, Natl Innovat Ctr Digital Fishery, Beijing, Peoples R China

2.China Agr Univ, Key Lab Smart Farming Technol Aquat Anim & Livesto, Minist Agr & Rural Affairs, Beijing 100083, Peoples R China

3.China Agr Univ, Beijing Engn & Technol Res Ctr Internet Things Agr, Beijing 100083, Peoples R China

4.China Agr Univ, Coll Informat & Elect Engn, Beijing 100083, Peoples R China

5.Chongqing Acad Agr Sci, Chongqing 400000, Peoples R China

关键词： Dead fish; Detection; Attention; Lightweight; Monitoring platform

期刊名称：MULTIMEDIA TOOLS AND APPLICATIONS （ 影响因子：3.0； 五年影响因子：2.9 ）

ISSN： 1380-7501

年卷期： 2023 年

页码：

收录情况： SCI

摘要： The cultivation stage of juvenile fish is the starting point of the whole cultivation process. In this stage, the vitality of juvenile fish is weak and they are extremely vulnerable to environmental stress and death. Therefore, there is an urgent need for an automated method to replace the aquaculture personnel to achieve the growth monitoring of juvenile fish in the cultivation stage. However, the dead juvenile fish are generally small, and there are interference signals with high similarity to the color of the dead fish, such as reflection and foam, on the water surface of the aquaculture area. To solve this problem, this study constructs dead fish detection model from two perspectives of improving model accuracy and realizing model lightweight. First, the dead fish detection data set is constructed using visual technology. Secondly, the Efficient Channel Attention module (ECA) is used to strengthen the feature auxiliary branch of YOLOv4 backbone feature extraction network output (E-YOLOv4), the E-YOLOv4 dead fish detection model is constructed (mAP = 96.91%, model-size = 244 M). In addition, this study replaced YOLOv4's cross stage partial network (CSPDarkNet53) with Densenet169 (D-YOLOv4), the D-YOLOv4 dead fish detection model is constructed (mAP = 95.93%, model-size = 84.7 M). The results show that E-YOLOV4 and D-YOLOV4 are superior to YOLOV4 in model effect and model-size. Finally, Python+PyQt5 is used to build a dead fish monitoring platform, and the above model is deployed on the platform to achieve real-time monitoring of dead fish in the breeding process. The dead fish monitoring platform established in this study can not only assist the aquaculture personnel in daily aquaculture supervision, but also assist the water quality sensor to jointly monitor the safety of the growth environment of juvenile fish. The platform is simple in design and has certain practical application value.

  • 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：重庆市农业科学院

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