摘要
在水产养殖中,水面残留饲料的实时检测可以有效减少饲料浪费和水污染,在经济效益和生态效益方面是双赢的局面。由于水面残留饲料的特殊性,如尺寸小、密集程度高等,使得水面残留饲料检测表现不佳。该研究提出了一种基于改进YOLOv8n的水面残留饲料检测算法,通过增添小目标检测层,融合多尺度特征以增强对小目标检测的精度;引入C2f_Faster_EMA模块,以降低模型的参数量,提高模型检测速度;构建ICBAM模块融入颈部网络,加强网络对小目标的特征信息融合,提升检测精度。结果显示:该算法相较于YOLOv8n的mAP@0.5提升10.3%;精确率P提升7.6%;召回率R提升10.2%;检测速度达到了125FPS。研究表明,该算法能有效实现对水面残留饲料快速、准确地检测。为实现水产养殖的智能化管理提供了技术支持,有望降低饲料浪费,改善水环境质量,提高养殖效益。
In aquaculture,the accumulation of leftover feed on the water surface not only leads to wastage but also contributes to deteriorating water quality,significantly impacting the well-being and growth of aquatic organisms.Conventional detection methods face challenges in accurately identifying small feed particles due to the intricate nature of aquatic environments.To tackle this issue,this research introduces an enhanced algorithm based on YOLOv8n for detecting residual feed on water surfaces.This algorithm improves the precision of detecting small feed particles by incorporating a specialized detection layer tailored for small targets.By amalgamating shallow and deep feature information,the algorithm enhances the network's ability to perceive targets across various scales,thereby boosting the accuracy of detecting small feed particles.Furthermore,the integration of the C2f_Faster_EMA module reduces model parameters,elevates detection speed,and fortifies the extraction of intricate features.Additionally,the devised ICBAM module bolsters the amalgamation of feature information for small targets,significantly enhancing detection accuracy.Experimental findings illustrate that the enhanced algorithm delivers exceptional performance across multiple evaluation metrics.Comparative to the original YOLOv8n,the mA P@0.5,precision,and recall rates have surged by 10.3%,7.6%,and 10.2%,respectively.Furthermore,the algorithm achieves a detection speed of 125 frames per second FPS,meeting the demands for real-time detection.These outcomes underscore the algorithm's capacity to swiftly and accurately identify residual feed on water surfaces,providing substantial technical backing for the intelligent administration of aquaculture.The implementation of this algorithm holds promise in efficiently curbing feed wastage,enhancing water quality,and amplifying the profitability of aquaculture operations.This advancement positions the aquaculture sector on a trajectory towards a more sustainable and efficient future.
作者
郑海锋
江林源
文露婷
周秀珊
介百飞
文家燕
ZHENG Haifeng;JIANG Linyuan;WEN Luting;ZHOU Xiushan;JIE Baifei;WEN Jiayan(College of Automation,Guangxi University of Science and Technology,Liuzhou 545006,Guangxi,China;GuangXi Academy of Fishery Sciences,Nanning 530021,Guangxi,China;GuangXi Aquatic Technology Promotion Station,Nanning 530021,Guangxi,China;Research Center for Intelligent Cooperation and Cross-application(Guangxi University of Science and Technology),Liuzhou 545616,Guangxi,China)
出处
《渔业现代化》
北大核心
2025年第1期80-88,共9页
Fishery Modernization
基金
国家自然科学基金(61963006)
广西自然科学基金面上项目(2018GXNSFAA050029,2018GXNSFAA294085)
广西科技重大专项(桂科AA22068064,桂科AA22068066)
广西重点研发计划(桂科AB23075093,桂科AB22035066)。
