基于改进YOLOv5注意力模型的农田害虫图像识别  

Farmland Pest Image Recognition Based on Improved YOLOv5 Attention Model

作　　者：石璐莹 童顺延 吴婷 冯媛 刘海华[1] SHI Luying;TONG Shunyan;WU Ting;FENG Yuan;LIU Haihua(South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学,湖北武汉430074

出　　处：《现代信息科技》2023年第10期70-73,79,共5页Modern Information Technology

基　　金：国家自然科学基金项目资助项目(61773409)。

摘　　要：农田害虫防控是一项争分夺秒的挑战,在此过程中害虫种类的正确识别是一项极为重要的环节。针对传统识别害虫检测过程中准确率低、检测目标较小的问题,文章提出了一种基于YOLOv5s和注意力机制的农田害虫图像识别模型。将自注意力机制引入YOLOv5s网络,对上下文信息进行建模,通过建立非局部模型提高网络解决图像远距离和多层次依赖关系的能力。实验结果显示,基于YOLOv5注意力模型的农田害虫图像识别具有较高的检测精度,可以有效识别和定位各类害虫。Pest prevention and control in farmland is a race against time challenge,and the correct identification of pest species is an extremely important link in this process.In response to the problems of low accuracy and small detection targets in traditional pest detection processes,a farmland pest image recognition model based on YOLOv5s and attention mechanism is proposed in this paper.Introduce the self attention mechanism into the YOLOv5s network to model contextual information,and improve the network's ability to solve long-distance and multi-level dependency relationships in images by establishing non local models.The experimental results show that the recognition of farmland pest images based on the YOLOv5 attention model has high detection accuracy and can effectively identify and locate various pests.

关 键 词：农田害虫识别 目标检测 YOLOv5 注意力机制 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术] TP18[自动化与计算机技术—计算机科学与技术]