基于匈牙利算法和LSTM网络的储粮害虫轨迹跟踪及行为研究  

Trajectory Tracking and Behavior Analysis of Stored Grain Pests via Hungarian Algorithm and LSTM Network

作　　者：闵超 詹炜[1] 张豫麒 吕建华[2] 洪胜兵 董天豫 佘金惠 黄华姿 Min Chao;Zhan Wei;Zhang Yuqi;LüJianhua;Hong Shengbing;Dong Tianyu;She Jinhui;Huang Huazi(School of Computer Science,Yangtze University 1,Jingzhou 434023;School of Food Science and Technology,Henan University of Technology 2,Zhengzhou 450001)

机构地区：[1]长江大学计算机科学学院,荆州434023 [2]河南工业大学粮油食品学院,郑州450001

出　　处：《中国粮油学报》2023年第3期28-34,共7页Journal of the Chinese Cereals and Oils Association

基　　金：中国高校产学研创新基金项目(2020ITA03012)。

摘　　要：长期以来储粮害虫给粮食储物带来严重损失,为寻找一种安全高效、环境友好的储粮害虫防治方法替代化学防治,本研究采用计算机视觉技术对视频图像逐帧处理,运用匈牙利算法对多个储粮害虫目标轨迹跟踪,并结合长短期记忆网络(LSTM)与余弦相似度方法解决多个储粮害虫在运动过程中目标重叠的问题,研究储粮害虫的运动行为。实验结果表明,该方法在137个储粮害虫正常运动视频中达到98.5%的跟踪准确度,在58个含有目标重叠的视频中达到89.7%的准确度。最后计算实时运动行为指标,如行走距离、行走速度、行走方向等,这为分析在热处理、熏蒸、气调等杀虫过程中储粮害虫运动行为,推算其移动范围并依此调整杀虫强度,从而控制成本以达到最好的杀虫效果提供支持,为建立高效的储粮害虫防治方案提供一种策略参考。Stored grain pests have brought serious losses to grain storage for a long time.In order to find a safe,efficient and environment-friendly prevention method to replace chemical control methods,the movement behavior of stored grain pests was studied in this paper.Computer vision technology was used to process video images frame by frame,Hungarian algorithm was used to track multiple stored grain pest targets,and LSTM(long short-term memory)network prediction and cosine similarity method were combined to solve the problem of target or overlap during the movement of multiple stored grain pests,thus,better tracking effect was obtained.The experimental results indicated that the tracking accuracy of this method was 98.5%in 137 videos with normal movement of stored grain pests and 89.7%in 58 videos with target overlap.Finally,the real-time motion data in the whole video,such as walking distance,walking speed and walking direction,etc.,would be calculated,providing theoretical support for the analysis of the movement behavior of grain pests stored during heat treatment,fumigation,air conditioning and other insecticidal processes,estimating their movement range and adjusting the intensity of insecticidal accordingly,so as to control the cost to achieve the best insecticidal effect,and then,providing a strategic reference for establishing an efficient control scheme of stored grain pests.

关 键 词：储粮害虫 匈牙利算法 长短期记忆网络(LSTM) 轨迹跟踪 余弦相似度 

分 类 号：TP399[自动化与计算机技术—计算机应用技术]