基于改进YOLOx模型的无人机影像城市树种识别  

Urban Tree Species Identification in UAV Images Based on Improved YOLOx Model

作　　者：王全福 唐丽玉[1,2] 焦怀瑾 何建国 WANG Quanfu;TANG Liyu;JIAO Huaijin;HE Jianguo(Key Laboratory of Spatial Data Mining&Information Sharing of Ministry of Education,Fuzhou University,Fuzhou 350108,China;National Engineering Research Center of Geospatial Information Technology,Fuzhou University,Fuzhou 350108,China)

机构地区：[1]福州大学空间数据挖掘与信息共享教育部重点实验室,福建福州350108 [2]福州大学地理空间信息技术国家地方联合工程研究中心,福建福州350108

出　　处：《测绘与空间地理信息》2025年第2期46-49,共4页Geomatics & Spatial Information Technology

基　　金：国家自然科学基金(41971344)资助。

摘　　要：利用高分辨率无人机遥感影像,结合深度学习算法,提出一种城市棕榈树树种识别的解决方法。以YOLOx-s模型为基础,在主干特征网络引入一种混合注意力机制模块CBAM;在Neck部分特征金字塔结构中加入Bottom-up结构;并改进IoU损失的计算方式,增大对网络误判的惩罚来优化模型。分别与Faster R-CNN、SSD、YOLOv3和原始YOLOx-s模型对实验区进行棕榈树识别,并对实验结果进行对比分析。实验结果表明,本文方法在基于高分辨率无人机影像的棕榈树识别中具有更高的精度,相较于原始的YOLOx-s有一定的提升。本文提出的改进后的YOLOx-s模型,实现了可见光影像的树种识别,可为城市棕榈树识别提供一定参考。Using high-resolution UAV remote sensing images,combined with deep learning algorithms,a solution of urban palm tree species identification is proposed.Based on the YOLOx-s model,a mixed attention mechanism module CBAM is introduced into the backbone feature network;the Bottom-up structure is added to the Neck partial feature pyramid structure;and the calculation method of IoU loss is improved to increase the risk of network misjudgment penalty to optimize the model.It identifies palm trees in the experimental area with Faster R-CNN,SSD,YOLOv3 and the original YOLOx-s model and compares and analyzes the experimental results.The experimental results show that the method in this paper has higher accuracy in palm tree identification based on high-resolution UAV images,and there is a certain improvement over the original YOLOx-s.The improved YOLOx-s model proposed in this paper realizes the tree species identification in visible light images and provides a certain reference for urban palm tree identification.

关 键 词：无人机 遥感影像 YOLOx-s 棕榈树 树种识别 注意力机制 

分 类 号：P237[天文地球—摄影测量与遥感]