林业工程学报2025，Vol.10Issue(5)：168-177,10.DOI:10.13360/j.issn.2096-1359.202404021
基于改进YOLOv7的松材线虫病疫木检测
Detection of pine wilt disease-infected trees based on an improved YOLOv7 model
摘要
Abstract
Pine trees,among the most widespread tree species globally,hold significant ecological and economic importance.However,pine wilt disease(PWD),a devastating forest disease caused by pine wood nematode,poses a severe threat to pine ecosystems worldwide.Characterized by strong pathogenicity,rapid onset,and fast propagation,PWD leads to substantial ecological and economic losses on a global scale.Photography by unmanned aerial vehicles(UAVs),followed by the processing of deep learning-based object detection algorithms,is currently a common approach for PWD detection.However,to cover large areas in short time periods,UAVs are often set to fly at high altitudes,thus pictures taken by UAVs are often subject to low ground resolution and frequently bear small targets,leading to low detection accuracy when the pictures are processed by object detection algorithms.To address this issue,this study proposed DSEN-YOLOv7,an object detection algorithm to detect PWD-infected trees in UAV images,by making four improvements over YOLOv7.Firstly,this algorithm replaced the CNN module in the MP block in YOLOv7's backbone network with a deformable convolution DCNv2,making the network more adaptable to shape variation of the PWD-infected trees.Secondly,this algorithm used the NWD loss function instead of the original loss function CIoU,improving the model's convergence speed and detection performance for small targets.Thirdly,the algorithm introduced the EMA attention mechanism,increasing the model's capability to extract the features of small targets.Fourthly,the algorithm replaced SPPCSPC with the SPPFCSPC space pyramid pooling module,improving the detection accuracy of the model.The experimental results showed that the detection performance of DSEN-YOLOv7 on mAP@0.5 and mAP@0.5∶0.95 reached 81.0%and 41.4%,respectively,counting 4.0 and 1.9 percentage points improvements over the original YOLOv7 model,respectively.Compared with other current mainstream object detection models,such as YOLOv8,YOLOv5i,Faster R-CNN and SSD,the proposed model achieved both increased F1 values and enhanced detection accuracy.The inference speed of the model was 175 frame/s and the size of the model was 71.89 MB.The proposed DSEN-YOLOv7 model meets the needs of large-scale real-time PWD detection and provides valuable technical support for efficient pine forest management.关键词
YOLOv7算法/目标检测/深度学习/无人机/松材线虫病Key words
YOLOv7 algorithm/object detection/deep learning/unmanned aerial vehicle/pine wilt disease分类
农业科技引用本文复制引用
陈冰雨,黄雷君,冯海林..基于改进YOLOv7的松材线虫病疫木检测[J].林业工程学报,2025,10(5):168-177,10.基金项目
浙江省公益技术应用研究计划(LGF21F020005). （LGF21F020005）
