草地学报 ›› 2026, Vol. 34 ›› Issue (10): 3750-3760.DOI: 10.11733/j.issn.1007-0435.2026.10.021

• 研究论文 • 上一篇    下一篇

基于无人机遥感影像的荒漠草原鼠害地特征识别算法研究

范雨荃1, 蔡斌1, 孙舰1, 董隆明1, 刘济泽1, 薛飞扬1, 袁帅2, 黄子杰3, 刘振宝4, 花立民1   

  1. 1. 甘肃农业大学草业学院/草业生态系统教育部重点实验室/草原鼠兔害治理国家林业和草原局重点实验室, 甘肃 兰州 730070;
    2. 内蒙古农业大学草业学院, 内蒙古 呼和浩特 010040;
    3. 民乐县林业和草原局, 甘肃 张掖 734500;
    4. 靖远县林业和草原局, 甘肃 白银 730600
  • 收稿日期:2025-07-21 修回日期:2026-01-11 出版日期:2026-10-15 发布日期:2026-10-10
  • 通讯作者: 花立民, E-mail: hualm@gsau.edu.cn
  • 作者简介:范雨荃(2000-),女,藏族,甘肃兰州人,硕士研究生,主要从事草地保护研究,E-mail:3195764089@qq.com
  • 基金资助:
    草业生态系统教育部重点实验室“揭榜挂帅”项目(KLGE-2024-06);鼠害高效、精准监测基础研究及新技术开发项目(2024YFD1400500);甘肃省教育厅优秀研究生“创新之星”项目(2023CXZX-664)资助

Investigation of the Algorithm for Detecting Rodent-Damaged Areas in Desert Grasslands Using Unmanned Aerial Vehicle Remote Sensing Images

FAN Yu-quan1, CAI Bin1, SUN Jian1, DONG Long-ming1, LIU Ji-ze1, XUE Fei-yang1, YUAN Shuai2, HUANG Zi-jie3, LIU Zhen-bao4, HUA Li-min1   

  1. 1. College of Pratacultural Science, Gansu Agricultural University/Key Laboratory of Grassland Rodent and Pika Pest Control of National Forestry and Grassland Administration/Engineering and Technology Research Center for Alpine Rodent Pest Control, National Forestry and Grassland Administration, Lanzhou, Gansu Province 730070, China;
    2. College of Grassland Science, Inner Mongolia Agricultural University, Hohhot, Inner Mongolia, Inner Mongolia 010040, China;
    3. Forestry and Grassland Bureau of Minle County, Zhangye, Gansu Province 734500, China;
    4. Forestry and Grassland Bureau of Jingyuan County, Baiyin, Gansu Province 730600, China
  • Received:2025-07-21 Revised:2026-01-11 Online:2026-10-15 Published:2026-10-10

摘要: 荒漠草原鼠害监测对于生物多样性保护和鼠害防控具有重要意义。本研究以甘肃省靖远县荒漠草原鼠害地为研究对象,在研究区内选取20块鼠害地,利用无人机采集可见光(RGB)影像,采用5种监督分类模型识别鼠害地植被、裸地及鼠洞3种地物,训练样本1578个,验证样本683个。以总体分类精度(Overall Accuracy, OA)、Kappa系数(Kappa)、生产者精度(Producer’s Accuracy, PA)和用户精度(User’s Accuracy, UA)评价分类模型精度。结果表明,SVM模型在整体分类表现上最优,其总体精度达99.68%,Kappa系数为0.99,鼠洞识别的生产者精度与用户精度分别为83.96%和87.51%。SVM模型在3类地物中均表现出较高的分类精度和稳定性,特别在裸地与植被识别中精度均达到99%以上。综上,在本研究区内,SVM模型表现出更优的识别效果,显示出其在荒漠草原鼠害地识别中的应用潜力,为荒漠草原鼠害监测与防控提供技术支撑。

关键词: 荒漠草原, 鼠害地, 无人机, 监督分类

Abstract: Monitoring rodent damage in desert grasslands is of great significance for biodiversity conservation and rodent pest control. This study selected rodent-damaged desert grasslands in Jingyuan County, Gansu Province, China, as the research area. A total of 20 rodent-damaged plots were established, and UAV(unmanned aerial vehicle)-based visible-light (RGB) imagery was acquired for analysis. Five supervised classification models were employed to identify three land-cover types, including vegetation, bare ground, and rodent holes. A total of 1,578 training samples and 683 validation samples were used. The classification performance was evaluated using overall accuracy (OA), Kappa coefficient (Kappa), producer’s accuracy (PA), and user’s accuracy (UA). The results showed that the support vector machine (SVM) model achieved the best overall performance, with an overall accuracy of 99.68% and a Kappa coefficient of 0.99. For rodent hole recognition, the producer’s accuracy and user’s accuracy reached 83.96% and 87.51%, respectively. The SVM model demonstrated high classification accuracy and stability across all three land-cover types, particularly achieving accuracies above 99% for bare ground and vegetation classification. Overall, within the study area, the SVM model exhibited superior performance in identifying rodent-damaged features, indicating its potential application in desert grassland rodent monitoring and management and providing technical support for rodent pest prevention and control.

Key words: Desert grassland, Rodent pest area, Drone, Supervised classification

中图分类号: