摘要
为了检测观赏鱼类的行为及其健康状况,设计了一种具有双层路由注意力机制的血鹦鹉(Vieja synspila♀×Amphilophus citrinellus♂)目标检测模型YOLOv8n-BiFormer,该方法在YOLOv8n模型基础上添加了双层路由注意力以减少计算量和内存,添加了新的视觉通用变换器BiFormer以提升计算效率,并采用ByteTrack算法追踪血鹦鹉的运动轨迹。结果表明:使用YOLOv8n-BiFormer模型对血鹦鹉的检测准确率达到99.2%,召回率为93.7%,平均精度均值(mAP@0.5)为99.1%,相较于YOLOv8n模型分别提升了0.8%、1.4%、1.0%;使用该模型对水族箱中的慈鲷(Chindongo demasoni)进行检测追踪同样取得了较好的效果,慈鲷的检测准确率达到97.0%,召回率为93.4%,平均精度均值为96.5%,相较于YOLOv8n模型召回率和平均精度分别提升了1.8%和1.9%。研究表明,本文中设计的YOLOv8n-BiFormer模型具有通用性,在检测和追踪血鹦鹉和慈鲷目标方面均表现优异,消耗的计算资源较少,可部署在水族箱监控系统中,为观赏鱼信息记录自动化和智能化提供了可行的解决方案。
In order to detect the behavior and health status of hybrid ornamental fish(Vieja synspila♀×Amphilophus citrinellus♂),a target detection model called YOLOv8n-BiFormer with a dual-layer routing attention mechanism was designed.In this method a dual-layer routing attention mechanism is added into the YOLOv8n model to reduce computation and memory requirements,and a new visual universal transformer called BiFormer is introduced to the YOLOv8n model for improvement of computational efficiency.The ByteTrack algorithm is employed to track the motion trajectory of the fish blood parrot.The results showed that the YOLOv8n-BiFormer model had a detection accuracy of 99.2%,a recall rate of 93.7%,and an average precision of 99.1%(mAP@0.5)for the blood parrot,increased by 0.8%,1.4%,and 1.0%compared to the YOLOv8n model,respectively.The model demonstrated good performance in the detection and tracking of the cichlid(Chindongo demasoni)in an aquarium,with a detection accuracy of 97.0%,a recall rate of 93.4%,and an average precision of 96.5%,increase by 1.8%in recall rate and 1.9%in average precision compared to the YOLOv8n model.The finding demonstrates that the designed YOLOv8n-BiFormer model performs excellently in detecting and tracking blood parrot fish and cichlid targets,with fewer computational resources,and that can be deployed in aquarium monitoring systems,providing feasible solution for the automation and intelligence of ornamental fish information recording.
作者
李鹏龙
张胜茂
沈烈
樊伟
顾家辉
邹国华
LI Penglong;ZHANG Shengmao;SHEN Lie;FAN Wei;GU Jiahui;ZOU Guohua(College of Navigation and Ship Engineering,Dalian Ocean University,Dalian 116023,China;Key Laboratory of Fisheries Remote Sensing,Ministry of Agriculture and Rural Affairs,East China Sea Fisheries Research Institute,Chinese Academy of Fishery Sciences,Shanghai 200090,China;Shanghai Junding Fishery Technology Company Limited,Shanghai 200090,China)
出处
《大连海洋大学学报》
CAS
CSCD
北大核心
2024年第2期318-326,共9页
Journal of Dalian Ocean University
基金
国家自然科学基金(61936014)
崂山实验室专项经费资助(LSKJ202201804)。
关键词
血鹦鹉
慈鲷
YOLOv8模型
检测追踪
ByteTrack算法
Vieja synspila♀×Amphilophus citrinellus♂
Chindongo demasoni
YOLOv8 model
detection and tracking
ByteTrack algorithm
