Underwater Sick Fish Detection Based on an Improved YOLOv11n Approach

Read full text Summarise with VIRAT
Authors: Guangyu Du, Chuande Xu, Su Xu, Chong Zheng
Cited by: 0
References: 12

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

To mitigate the issues of missed and false detections arising from scale variations, image blurring and target occlusion in underwater diseased fish detection, this study develops an enhanced YOLOv11n-based algorithm designated as YOLOv11-DFL. The backbone network incorporates the lightweight MobileNetV4, which combines depthwise separable convolution, lightweight attention modules, and multi-scale feature enhancement to balance feature extraction capability with significant reductions in parameters and computation. For the detection layer, the original P5 branch is removed and a P2 branch is added to enhance shallow detail capture, thereby reducing missed and false detections of small targets. The neck network integrates SlimNeck, which uses GSConv for channel splitting to cut redundant computation and VovGSCSP to strengthen multi-scale feature aggregation and gradient transfer, achieving lightweight design while boosting feature expression efficiency and small-target detection accuracy. Finally, the WIM loss function is adopted to improve model generalization and accelerate convergence. Experiments on the self-built sick fish dataset show that compared with the original YOLOv11n, the proposed method increases Precision, Recall, and mAP50 by 4.7%, 9.2%, and 7.0% respectively. Tests on underwater target datasets including RUOD, URPC2019, UDD, and DUO also demonstrate improved precision and recall, verifying the superior performance of YOLOv11-DFL in sick fish detection.

Fields of Study & Subjects

Water Quality Monitoring Technologies Advanced Neural Network Applications Smart Agriculture and AI

Publication details

Journal / source
Journal of Image Processing Theory and Applications
Volume
9
Issue
1
Publication date
1 January 2026
Language
EN