现代电子技术2025，Vol.48Issue(11)：151-155,5.DOI:10.16652/j.issn.1004-373x.2025.11.023
PCASort:融合粒子滤波和注意力机制的鱼类跟踪算法
PCASort:Fish tracking algorithm incorporating particle filtering and attention mechanism
谭华超 1袁贵鸿 1江彦博 1毕海 2程远 3刘丹1
作者信息
- 1. 大连海洋大学 信息工程学院,辽宁 大连 116023
- 2. 杭州云栖智慧视通科技有限公司,辽宁 大连 116000
- 3. 大连理工大学 宁波研究院,浙江 宁波 315000
- 折叠
摘要
Abstract
In fish multiple object tracking,the feature differences among individuals are not obvious due to the fact that most of the fish have similar appearance within the fish class,which leads to low accuracy and poor robustness of fish tracking algorithms based on appearance features for data association.In view of the above,a DeepSort-based improved multiple object tracking algorithm for fish is proposed.The algorithm is named as PCASort.The movement trajectories of fish may exhibit nonlinear characteristics due to external perturbations,so the original Kalman filtering method is replaced with particle filtering applicable to nonlinear and non-Gaussian problems,so as to improve the accuracy of trajectory prediction.An improved coordinate attention mechanism is added to the original feature extraction network,and the location information is embedded into the generated feature vectors used to compute the minimum cosine distance,so as to improve the accuracy rate of data association.The experimental results show that the accuracy of IDF1 and multiple object tracking accuracy(MOTA)on open-source video dataset is 63.4%and 96.9%,respectively,and the number of ID switches is 31 times.关键词
鱼类跟踪/多目标追踪/粒子滤波/注意力机制/最小余弦距离/交并比Key words
fish tracking/multiple object tracking/particle filtering/attention mechanism/minimum cosine distance/intersection over union分类
电子信息工程引用本文复制引用
谭华超,袁贵鸿,江彦博,毕海,程远,刘丹..PCASort:融合粒子滤波和注意力机制的鱼类跟踪算法[J].现代电子技术,2025,48(11):151-155,5.
