基于PCA和SVM的遥感影像水体提取方法及验证  

PCA and SVM-based algorithm of water area extraction from remote sensing images and its verification

作　　者：周婷[1] 汪炎[2] 邹俊 李辰 崔玉环[3] 王笑宇 谢传流 夏萍[1] ZHOU Ting;WANG Yan;ZOU Jun;LI Chen;CUI Yuhuan;WANG Xiaoyu;XIE Chuanliu;XIA Ping(School of Engineering,Anhui Agricultural University,Hefei 230036,China;East China Engineering Science&Technology Co.,Ltd.,Hefei 230024,China;School of Science,Anhui Agricultural University,Hefei 230036,China)

机构地区：[1]安徽农业大学工学院,安徽合肥230036 [2]东华工程科技股份有限公司,安徽合肥230024 [3]安徽农业大学理学院,安徽合肥230036

出　　处：《水资源保护》2023年第2期180-189,共10页Water Resources Protection

基　　金：国家自然科学基金(U2243228);安徽省自然科学基金(2008085ME158,1808085ME158,2108085QE220);安徽农业大学稳定和引进人才科研资助项目(K2141004);工业废水及环境治理安徽省重点实验室开放基金(DHSZ202202)。

摘　　要：针对内陆湖泊水质及光谱特性空间差异性大、支流水系结构复杂而导致的遥感影像水体提取精度低的问题,提出了结合光谱主成分分析(PCA)及支持向量机(SVM)的PCA-SVM水体提取算法。基于GF-1卫星遥感影像,对原始影像光谱波段特征进行PCA降维,从中优选熵、方差、差异性纹理特征向量,结合原始波段及归一化差分水体指数(NDWI),构建了8维特征向量,并基于SVM算法提取湖泊水体。以巢湖洪水期与非洪水期影像为研究实例,分别采用NDWI法、传统SVM算法及PCA-SVM算法对水体进行提取,并进一步基于PCA-SVM算法对2020年汛期巢湖洪水期淹没演变过程进行反演和跟踪,定量解析特征向量组合及SVM惩罚系数C对水体提取性能的影响。结果表明:PCA-SVM算法提取的湖泊完整、支流连续,显著改善了含蓝藻水体漏提、建筑物误提等问题;洪水期和非洪水期提取结果的F1分数分别为95.08%和97.95%,虚警率分别为5.43%和1.13%,提取精度显著高于NDWI法和SVM算法。Considering the low accuracy of water area extraction from remote sensing images due to great spatial difference of water quality and spectral characteristics in inland lakes as well as complex tributary structures,a water area extraction algorithm combining spectral principal component analysis(PCA)and support vector machine(SVM)was proposed.Based on GF-1 satellite images,PCA dimensionality reduction analysis was conducted to derive texture feature vectors,including entropy,variance,and differentiation.Together with original 4-band spectrums and normalized difference water index(NDWI),an 8-dimensional optimal feature vector was constructed,and then water area was extracted using the SVM algorithm.With the GF-1 remote sensing image of Chaohu Lake area in flood and non-flood periods used as a case study,the NDWI method,traditional SVM algorithm,and PCA-SVM algorithm were applied to extracting lake water area.Furthermore,based on the PCA-SVM algorithm,the flooding evolution process of Chaohu Lake in the flood period of 2020 was inverted and tracked,and the influences of feature vector combination and the penalty parameter C of SVM were quantitatively analyzed.Results show that the lake area extracted with the PCA-SVM algorithm is complete with continuous tributaries,and mis-extractions due to blue algae and building confusions are significantly overcome;the F1 scores of the PCA-SVM algorithm in the flood and non-flood periods are 95.08%and 97.95%,and the false alarm rates are 5.43%and 1.13%,respectively,demonstrating a significant improvement of the PCA-SVM algorithm in extraction accuracy,as compared with the NDWI method and SVM algorithm.

关 键 词：遥感影像水体提取 归一化差分水体指数(NDWI) 支持向量机(SVM) 主成分分析(PCA) 纹理特征向量 巢湖 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]