基于视频目标跟踪算法的储粮害虫活跃程度判别研究  

Identification of Stored-Grain Insects'Activity Based on Video Object Tracking Algorithm

作　　者：刘思琪 李江涛 钱荣荣[1] 周慧玲[1] Liu Siqi;Li Jiangtao;Qian Rongrong;Zhou Huiling(College of Automation, Beijing University of Posts and Telecommunications, Beijing 100876)

机构地区：[1]北京邮电大学自动化学院,北京100876

出　　处：《中国粮油学报》2021年第11期179-186,共8页Journal of the Chinese Cereals and Oils Association

摘　　要：针对储粮害虫杀虫效果评估实验,提出了一种多目标跟踪算法,通过跟踪评估区域内多头储粮害虫的运动给出每头害虫的活跃程度,对害虫存活状态进行自动评估。算法基于Faster R-CNN框架的目标检测技术,融合Mean Shift和Deep SORT目标跟踪算法,实现了对储粮害虫运动位置的连续跟踪,且减少了两头害虫相遇再分离后身份错位问题发生的情况。对于20头以内的同种害虫,平均多目标跟踪准确率为95.89%,多目标跟踪精度为83.18%。而且在目标跟踪算法中记录了每头储粮害虫的速度变化,通过分析一定时长的害虫移动速度,提出了一种评估储粮害虫活跃程度为低、中或高的分级方法,可以辅助实验人员对实验效果的分析。A multiple object tracking algorithm was proposed to evaluate the survival state of stored-grain insects automatically during fumigation experiments,which tracked the movements of stored-grain insects in the target area,giving the definition of activity degree of each insect.Based on the object detection framework Faster R-CNN,the proposed algorithm was combined with Mean Shift and Deep SORT tracking algorithms,that realized the continuous position tracking of the moving insects,and reduced the occurrence of identity switches after two insects touching each other.For the video containing less than 20 insects,the accuracy and precision of multiple object tracking is 95.89%and 83.18%respectively.The object tracking algorithm recorded the velocity change of each stored-grain insect,and analyzed the moving speed of tracked insects during the time period of a certain length.Then,a grading method was proposed to evaluate the activity degree of stored-grain insects as low,medium or high to assist the analysis of fumigation experiments.

关 键 词：储粮害虫 目标跟踪算法 害虫活跃程度 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]