面向鱼水分离作业的鱼群智能视觉计数方法  

Intelligent visual fishes counting method for fish-water separation operation

作　　者：李雨宸 左浩博 李升晨 符长虹 王志勇[3] 刘祥勇[3] LI Yuchen;ZUO Haobo;LI Shengchen;FU Changhong;WANG Zhiyong;LIU Xiangyong(School of Mechanical Engineering,Tongji University,Shanghai 201804,China;Department of Computer Science,University of Hong Kong,Hong Kong 999077,China;Fishery Machinery and Instrument Research Institute,Chinese Academy of Fishery Science,Shanghai 200092,China)

机构地区：[1]同济大学机械与能源工程学院,上海201804 [2]香港大学计算机科学系,中国香港999077 [3]中国水产科学研究院渔业机械仪器研究所,上海200092

出　　处：《传感器与微系统》2025年第3期102-106,共5页Transducer and Microsystem Technologies

基　　金：中国水产科学研究院中央级公益性科研院所基本科研业务费专项资金资助项目(2023TD87)。

摘　　要：针对养殖鱼群鱼水分离过程中现有人工计数方法存在计数效率不足、成本过高等问题,设计一种基于动态状态感知的智能视觉鱼群多目标跟踪与计数方法。具体地,分析鱼的姿态和运动状态变化,设计动态状态感知关联模块,以实现对具有复杂运动状态鱼群目标的高精度关联跟踪;引入自适应轨迹初始化修正模块对比轨迹状态特征,以过滤错误轨迹,提高跟踪计数精度;采用差值过线计数规则缓解重复计数问题。将本文所提方法在大型自采集真实挑战性鱼水分离场景下的鱼群多目标跟踪数据集FishMOT上开展实验,实验结果表明:本文方法在跟踪与计数精度方面均优于已有多目标跟踪计数方法。同时将本文方法在真实鱼水分离场景进行计数测试,其计数精度和实时性速度均满足实际应用需求。Aiming at the problems of low efficiency and high cost of manual fishes counting methods in the process of aquaculture fish-water separation,an intelligent visual fishes multi-targets tracking and counting method based on dynamic state-aware is proposed.Analyzing pose and motion state changes of fishes,the dynamic state-aware association module is designed to realize the high-precision association and tracking of fishes target which have complex motion states.An adaptive correction of trajectory initialization module is introduced to compare the features of trajectory state to filter the ghost trajectory and enhance the tracking and counting precision.The differential line-crossing counting rule is used to alleviate double counting problem.Experiments are carried out on FishMOT,a self-collected large fishes multi-targets tracking dataset from real challenging fish-water separation scenarios.The experimental results show that the proposed method outperforms the existing multi-targets tracking and counting methods in terms of tracking and counting precision.The counting test is carried out using this method in real fish-water separation scenario show that the proposed method meets the requirements of practical applications in counting precision and real-time speed.

关 键 词：鱼水分离 鱼群计数 智能视觉感知 多目标跟踪 动态状态感知 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]