基于无人机遥感与深度学习的芨芨草识别方法  

Achnatherum splendens identification based on uav remote sensing and deep learning

作　　者：杨红艳[1,2] 杜健民[3] YANG Hongyan;DU Jianmin(School of Mechanical Engineering,Inner Mongolia University of Technology,Hohhot 010051,China;Inner Mongolia Key Laboratory of Special Service Intelligent Robotics,Inner Mongolia University of Technology,Hohhot 010051,China;School of Mechanical and Electrical Engineering,Inner Mongolia Agricultural University,Hohhot 010018,China)

机构地区：[1]内蒙古工业大学机械工程学院,呼和浩特010051 [2]内蒙古工业大学内蒙古自治区特殊服役智能机器人重点实验室,呼和浩特010051 [3]内蒙古农业大学机电工程学院,呼和浩特010018

出　　处：《内蒙古工业大学学报（自然科学版）》2024年第3期250-256,共7页Journal of Inner Mongolia University of Technology：Natural Science Edition

基　　金：国家自然科学基金项目(31660137);内蒙古工业大学科学研究项目(BS2020016)。

摘　　要：芨芨草是干旱、半干旱草原广泛分布的一种多年生杂草,具有极强的环境适应力和种群竞争力,芨芨草的分布状况和变化趋势对于维持区域生态系统平衡和稳定具有重要意义。利用无人机高光谱遥感技术采集内蒙古格根塔拉荒漠草原遥感影像,获得厘米级空间分辨率和纳米级光谱分辨率的图像。采用子区间波段选择法选择代表研究区地物的特征波段,实现数据去除冗余和降维。在ENVI 5.6.1中利用Deep learning模块构建基于U-Net网络的深度学习模型对研究区的芨芨草进行识别,总体分类精度为95.67%,Kappa系数为0.83,均高于其他四种机器学习算法。研究结果表明,基于特征波段选择的深度学习算法能更有效地提取地物的光谱、纹理和形状信息,无人机高光谱低空遥感和深度学习算法的结合为荒漠草原芨芨草的准确、快速识别提供了新途径。Achnatherum splendens is a perennial weed widely distributed in arid and semi-arid grasslands,with strong environmental adaptability and population competitiveness.Its distribution and changing trends are of great significance for the balance and stability of regional ecosystems.Unmanned Aerial Vehicle(UAV)hyperspectral remote sensing technology was used to acquire remote sensing images of Gegentala desert steppe in Inner Mongolia,and obtained centimeter level spatial resolution and nanometer level spectral resolution images.The subinterval band selection method was adopted to select the feature bands representing the characteristics of the research area,achieving data redundancy removal,dimensionality reduction,and feature extraction.In ENVI 5.6.1,a deep learning model based on U-Net network was constructed by using the deep learning module to identify Achnatherum splendens in the study area,and was compared with random forest,neural network,support vector machine and maximum likelihood classification.Among the five classification methods,the deep learning classification algorithm has the best recognition effect on Achnatherum splendens,with an overall classification accuracy of 95.67%and a Kappa coefficient of 0.83,all higher than the other four machine learning algorithms.The research results indicate that deep learning algorithms based on feature band selection can more effectively extract spectral,texture,and shape information.The combination of UAV hyperspectral low altitude remote sensing and deep learning algorithms provides a new approach for accurate and rapid identification of Achnatherum splendens in desert steppe.

关 键 词：无人机遥感 高光谱 深度学习 芨芨草 识别 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]