基于Relief F与卷积神经网络的湿地植物群落精细分类  

Wetland plant community classification based on Relief F and convolutional neural network

作　　者：张永彬 刘雅辉 刘明月 满卫东 宋唐雷 李春雨 ZHANG Yongbin;LIU Yahui;LIU Mingyue;MAN Weidong;SONG Tanglei;LI Chunyu(College of Mining Engineering,North China University of Science and Technology,Tangshan 063210,China;Tangshan Key Laboratory of Resources and Environmental Remote Sensing,Tangshan 063210,China;Hebei Industrial Technology Institute of Mine Ecological Remediation,Tangshan 063210,China;Hebei Key Laboratory of Mining Development and Security Technology,Tangshan 063210,China)

机构地区：[1]华北理工大学矿业工程学院,河北唐山063210 [2]唐山市资源与环境遥感重点实验室,河北唐山063210 [3]河北省矿区生态修复产业技术研究院,河北唐山063210 [4]河北省矿业开发与安全技术重点实验室,河北唐山063210

出　　处：《测绘通报》2023年第2期58-64,共7页Bulletin of Surveying and Mapping

基　　金：国家自然科学基金(41901375,42101393);河北省自然科学基金(D2019209322,D2022209005);河北省高等学校科学技术研究项目青年拔尖人才项目(BJ2020058);河北省引进留学人员资助项目(C20200103);唐山市科技计划重点研发项目(19150231E)。

摘　　要：湿地是陆地生态系统和水生生态系统之间的重要过渡带,准确高效地获取湿地植物群落分布信息对于保护湿地具有深远的意义。本文以无人机多光谱影像为数据源,首先构建包含光谱特征、植被指数和纹理特征的多维特征数据集,并采用Relief F算法进行特征优选,确定最优特征数据集;然后构建基于特征优选的卷积神经网络(CNN)分类模型,对最优特征数据集进行分类,并与基于原始多光谱影像的CNN和随机森林(RF)分类方法进行对比。结果表明:(1)随着特征个数的增加,分类精度先增加后下降,当特征数为32时分类精度最高;(2)窗口为13×13的GLCM提取的信息熵和同质性等纹理特征及GNDVI、MSAVI2、RVI等多光谱植被指数重要性较高;(3)基于最优特征数据集的CNN分类模型,能够有效提取空间光谱信息,抑制“椒盐现象”的产生,分类效果最佳,总体精度达93.40%,与未进行特征优选的RF和CNN分类模型相比分别提高了9.80%和7.40%。Wetland is an important transition zone between terrestrial and aquatic ecosystems. Accurate and efficient acquisition of wetland plant community distribution information is of profound significance for wetland conservation. We use the UAV multispectral images as the data source, construct a multidimensional feature dataset containing spectral features, vegetation indices, and texture features, and determine the optimal feature dataset by the Relief F algorithm. Then, we construct a convolutional neural networks(CNN) classification model based on the feature selection and classify the optimal feature dataset with the CNN based on the original multispectral images. The results show that:(1) The classification accuracy increases and then decreases as the number of features increases, and the classification accuracy is highest when the number of features is 32.(2) Texture features such as information entropy, homogeneity, and multispectral vegetation indices such as GNDVI, MSAVI2, and RVI extracted by GLCM with a window of 13×13 have higher importance.(3) The CNN classification model based on the optimal feature dataset can effectively extract the spatial-spectral information and suppress the “salt-and-pepper noise”, with the best classification effect and the overall accuracy of 93.40%, which is 9.80% and 7.40% higher than the RF and CNN classification models without feature optimization, respectively.

关 键 词：无人机多光谱 卷积神经网络 特征优选 湿地精细分类 

分 类 号：P23[天文地球—摄影测量与遥感]