基于无人机遥感和深度学习的布氏田鼠洞群识别  

UAV-based Remote Sensing and Deep Learning for the Identification of Burr’s Vole Hole Clusters

作　　者：堂格斯 包玉龙[1,2] 孙长青 特日格乐 包玉海[1,2] 图布新巴雅尔 金额尔德木吐[1,2] Tanggesi;BAO Yulong;SUN Changqing;Terigele;BAO Yuhai;Tubuxinbayaer;JIN Eerdemutu(College of Geographical Sciences,Inner Mongolia Normal University,Hohhot 010022,China;Inner Mongolia Key Laboratory of Remote Sensing and Geography Information System,Hohhot 010022,China)

机构地区：[1]内蒙古师范大学地理科学学院,内蒙古呼和浩特010022 [2]内蒙古自治区遥感与地理信息系统重点实验室,内蒙古呼和浩特010022

出　　处：《内蒙古民族大学学报（自然科学版）》2023年第5期448-457,共10页Journal of Inner Mongolia Minzu University：Natural Sciences

基　　金：内蒙古自治区科技重大专项课题(2021ZD004503);内蒙古自治区自然科学基金项目(2021MS04016);内蒙古自治区重点研发与成果转化计划项目(2022YFSH0070);内蒙古师范大学基本科研业务费专项资金资助项目(2022JBTD011)。

摘　　要：草原鼠害是典型的次生灾害,是草原退化引起的一种生物灾害,发生后不仅加剧草原退化程度,也会引起鼠疫等极度危险的传染病。因此,精确、及时监测鼠害状态对防治鼠害、草原退化工作具有重要实际意义。以内蒙古呼伦贝尔和锡林郭勒草原区发生的布氏田鼠灾害区为例,通过野外实地调查,采集了5个不同生境区无人机正射影像,分析每个区域布氏田鼠洞群结构和影像特征,并采集和建立样本库,分别构建深度学习(SSD、RetinaNet、YOLOv3和Faster R-CNN)识别模型,进行了布氏田鼠洞群识别。结果表明:Faster R-CNN模型在准确率上表现出更高的可靠性,其mAP达到了84.4%;SSD模型效果最弱,mAP只有45.4%,但在检测速度上较快。因此,可采用Faster R-CNN模型作为布氏田鼠洞群识别模型的核心算法。Grassland rodent damage is a typical secondary disaster and a biological disaster caused by grassland degradation.This disaster often aggravates grassland degradation and also causes extremely dangerous infectious diseases such as pestis.The breeding speed of mouse pests is fast,and the expansion of affected areas is also very rapid.Therefore,accurate and timely monitoring of mouse damage status is of important practical significance for the premise and basic work of preventing and controlling mouse damage and grassland degradation.In this paper,taking the outbreak of Brandt’s vole disaster areas in Hulunbuir and Xilingol grasslands in Inner Mongolia as an example,drone orthophoto images from five different habitat areas were collected through field investigations,the image characteristics of burrow groups of Brandt’s voles in each area were analyzed,and in-depth learning(SSD,RetinaNet,YOLOv3,and Faster R-CNN)recognition models were constructed to identify the burrow groups of Brandt’s voles.The results showed that the Faster R-CNN model shows higher reliability in accuracy,with its mAP reaching 84.4%.The SSD model had the weakest effect,with only 45.4%of mAP,but it is faster in detection speed.Therefore,the Faster R-CNN model can be adopted as the core algorithm for identifying the burrow group of Brandt’s voles.

关 键 词：无人机遥感 布氏田鼠 深度学习 洞群识别 

分 类 号：S443[农业科学—农业昆虫与害虫防治] V279[农业科学—植物保护]