基于特征优选的GF-6WFV影像湿地信息提取OACSTPCD

Taking the Hongze Lake wetland as the study area,we proposed an object-oriented wetland classification method based on feature op-timization,and used GF-6 WFV multispectral imagery to obtain wetland information.Firstly,GF-6 WFV data was preprocessed and multi-scale segmentation.Then,we extracted the characteristic variables,including spectral feature,vegetation index,water index,red edge index and tex-ture feature from GF-6 WFV images,and obtained the optimal feature set by using a recursive elimination algorithm with mean decrease accu-racy(MDA).Finally,we used the object-oriented classification method to extract the wetland information based on the optimal feature combi-nation.Comparing with four different feature optimization algorithms,we found that the recursive elimination algorithm with MDA can select feature variables more effectively.We used different feature combinations in the Hongze Lake wetland to carry out six classification experi-ments.The results indicated that the overall accuracy of proposed method up to 87.43%is the highest,improving 1.55%compared with the method without feature optimization.Additionally,the two red-edge bands and red-edge indexes of GF-6 WFV data play an important role in wetland classification.The research results will provide technical reference for the application of GF-6 WFV multispectral images in wetland monitoring.