基于改进YOLOv4的储粮害虫检测算法  

Stored Grain Pest Detection Algorithm Based on Improved YOLOv4

作　　者：孙妍 宋雪桦[1] 陈静[1] 蒋思玮 SUN Yan;SONG Xuehua;CHEN Jing;JIANG Siwei(School of Computer Science and Communication Engineering,Jiangsu University,Zhenjiang 212013)

机构地区：[1]江苏大学计算机科学与通信工程学院,镇江212013

出　　处：《计算机与数字工程》2023年第6期1217-1222,1255,共7页Computer & Digital Engineering

基　　金：国家重点研发计划(编号:2017YFC1600804);国家自然科学基金项目(编号:61902156);江苏省自然科学基金项目(编号:BK20180860)资助。

摘　　要：针对储粮害虫检测中害虫目标较小,且存在重叠和遮挡现象,检测难度大的问题,提出了基于改进YOLOv4的储粮害虫检测算法。首先,该算法以多尺度检测的YOLOv4为基础算法,满足了小目标检测的要求;其次,引入K-means算法,对储粮害虫数据集聚类分析,调整先验框大小,加强对储粮害虫目标的检测能力;最后,在YOLOv4主干网络中加入空间金字塔池化结构,增加感受野,增加网络的特征提取能力。实验结果表明,在储粮害虫检测中,论文提出的算法的平均精度均值达到93.99%,与未改进的YOLOv4算法相比提高了4.49%,且检测速度达到了26.36帧/s。与YOLOv3相比,改善了储粮害虫的漏检情况。In order to solve the problems of small pest targets,overlapping and shielding,and difficult detection in the detec-tion of stored grain pests,a stored grain pest detection algorithm based on improved YOLOv4 is proposed.Firstly,the algorithm is based on the multi-scale detection of YOLOv4,which meets the requirements of small target detection.Secondly,the K-means al-gorithm is introduced to cluster the stored grain pest data set,and the size of the prior box is adjusted to enhance the detection abili-ty of stored grain pest targets.Finally,the spatial pyramid pooling structure is added to the backbone network of YOLOv4 to in-crease the receptive field,enhance the network feature extraction ability.The experimental results show that in the detection of stored-grain pests,the mean average accuracy of the proposed algorithm reaches 93.99%,which is 4.49%higher than the unim-proved YOLOv4 algorithm,and the detection speed reaches 26.36 frames per second.Compared with YOLOv3,the missed detec-tion effect of stored grain pests is improved.

关 键 词：储粮害虫 YOLOv4 K-MEANS 空间金字塔池化 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]