改进YOLOv5模型下农作物虫害图像识别算法仿真  

Simulation of crop pest image recognition algorithm based on improved YOLOv5 model

作　　者：陆娇娇[1,2] 段峰峰 LU Jiao-jiao;DUAN Feng-feng(School of Information and Intelligent Engineering,Sanya University,Sanya Hainan 572000,China;Academician Chunming Rong Workstation,Sanya Hainan 572000,China;Hunan Normal University,Changsha Hunan 410081,China)

机构地区：[1]三亚学院信息与智能工程学院,海南三亚572000 [2]容淳铭院士工作站,海南三亚572000 [3]湖南师范大学,湖南长沙410081

出　　处：《计算机仿真》2023年第11期156-160,共5页Computer Simulation

基　　金：2021海南省自然科学基金青年基金项目(621QN0900);2023年海南省高等学校科学研究项目(Hnky2023-40)。

摘　　要：由于环境和光线等条件影响,虫害图像识别难度较大,为此提出改进YOLOv5模型下农作物虫害图像识别算法。通过K-means算法改进并优化暗通道图像去雾算法,利用优化后的暗通道图像去雾算法对农作物虫害图像展开处理,以此避免因水汽或光照不均等因素造成的图像清晰度低的问题;对原始YOLOv5模型中的注意力机制和损失函数展开优化,通过优化注意力机制提高农作物虫害图像特征提取的精度,根据优化后的损失函数调整预测框的位置以此锁定虫害目标;将去雾后的农作物虫害图像输入改进后的YOLOv5模型中,实现农作物虫害图像的识别。仿真结果表明,经过所提算法改进后的YOLOv5模型IoU有所提高,且特征提取精度和图像识别效率高。Due to the influence of environmental and lighting conditions,it is difficult to recognize pest images.Therefore,this paper presented an algorithm for identifying crop pest images based on the improved YOLOv5 model.Firstly,we used the K-means algorithm to improve and optimize the dark channel image defogging algorithm,and pro-cessed the crop pest images by the optimized algorithm,thus avoiding the problem of low image clarity caused by water vapor or uneven lighting.Secondly,we optimized the attention mechanism and loss function in the original YOLOv5 model in order to improve the accuracy of feature extraction of crop pest images.Moreover,we adjusted the position of the prediction box according to the optimized loss function,thus locking the pest target.Finally,we input the defogged images into the improved YOLOv5 model,thus achieving the recognition of crop pest images.Simulation results show that the IoU of the YOLOv5 model improved by the proposed algorithm is improved.In addition,the fea-ture extraction accuracy and image recognition efficiency are high.

关 键 词：暗通道图像去雾算法 注意力机制 农作物虫害识别 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]