基于YOLOv5的松材线虫感染检测优化  

Detection of Pine Wood Nematode Infestation Based on an Improved YOLOv5 Model

作　　者：苗家明 张幻 张学俊[1] Miao Jiaming;Zhang Huan;Zhang Xuejun(Zhejiang International Maritime College,Zhoushan 316021,China;The Open University of Zhoushan,Zhoushan 316021,China)

机构地区：[1]浙江国际海运职业技术学院,浙江舟山316021 [2]舟山开放大学,浙江舟山316021

出　　处：《浙江国际海运职业技术学院学报》2024年第3期8-14,共7页Journal of Zhejiang International Maritime College

基　　金：浙江省教育厅一般科研项目“基于复合翼无人机海岛松材线虫病精准监测的研究与应用”(项目编号:Y202250763)。

摘　　要：松材线虫病是由松木线虫寄生在松树内部引起的一种毁灭性森林病害。一旦感染,由于无法排水,松树会迅速凋萎死亡,因此也被称为松树枯萎病。主要由松虫传播,具有快速传播、快速发作和高死亡率的特点,是一种重要的全球植物病害。如果不能及时有效地控制松材线虫病,将导致森林地区大量松树在短时间内死亡。随着人工智能和无人机技术的不断发展和普及,结合各种方法及时检测病树,进行处理和保护,可以大大节省生物防治的时间和人力成本,同时推动诸如生物工程和绿化工程等多个相关学科的工作。通过无人机对森林区域进行了大量的航空勘查,收集了相关数据,并对所获取到的数据进行了预处理,包括数据扩充、清洗、过滤、去重、格式化等,以确保数据质量和准确性。然后对数据进行了标记,对原始的YOLOv5进行了一定的改进。并为原始网络模型添加了一个新的RRAM(递归残余注意模块),使得网络能够及时关注冗余数据中的重要信息,从而提高了网络的性能。与原始的YOLOv5相比,优化后的网络具有更强的性能。Pine wilt disease,caused by the parasitism of pine wood nematodes within pine trees,is a devastating forest disease.Infected trees quickly wither and die due to their inability to transport water,earning the disease the alternate name“pine tree withering disease.”The disease is primarily spread by pine sawyer beetles and is characterized by rapid transmission,fast onset,and a high mortality rate,making it a major global plant threat.Without timely and effective control,pine wilt disease can cause widespread pine tree mortality in forested areas.With advancements in artificial intelligence(AI)and unmanned aerial vehicle(UAV)technology,integrating these methods for early detection,treatment,and protection of affected trees can significantly reduce the time and labor costs associated with biological pest control.Furthermore,these technologies support advancements in fields like bioengineering and environmental management.In this study,we conducted extensive aerial surveys of forested areas using UAVs,collected relevant data,and preprocessed it through techniques such as data augmentation,cleaning,filtering,deduplication,and formatting to ensure high data quality and accuracy.We then labeled the data and made significant improvements to the original YOLOv5 model by introducing a new Recurrent Residual Attention Module(RRAM).This addition allows the network to focus on critical information in redundant data,thereby improving its performance.Compared to the original YOLOv5,our improved model demonstrates superior performance.

关 键 词：绿化工程 目标检测 YOLOv5 害虫管理 无人机航拍 

分 类 号：TP29[自动化与计算机技术—检测技术与自动化装置]