Extraction of coastal cultivation areas based on Sentinel-2 remote sensing imagery
China's coastal aquaculture accounts for a large proportion across the world.Quickly obtaining information on the size and distribution of coastal aquaculture areas is conducive to the monitoring and planning,yield estimation and disaster prevention of aquaculture areas.In view of the difficulties in distinguishing raft aquaculture area from seawater,low recognition accuracy and"salt and pepper"noise in the process of raft aquaculture area extraction,this paper takes the raft aquaculture area near Changshan Islands as the research area,and uses Sentinel-2 satellite remote sensing image data to construct the features of spectrum,texture and geometric.The feature space optimization(FSO)is used to obtain the dominant features of raft aquaculture area extraction.The object-oriented random forest,decision tree and nearest neighbor algorithms are used to extract the raft aquaculture area in the study area.Based on the analysis and comparison of the extraction results.The optimal classification algorithm is summarized,and the reliability of FSO is verified.The results show that:(1)the normalized difference water index,geometric feature area and length,and gray level co-occurrence matrix correlation are the optimal features for identifying raft culture areas.(2)The classification after feature optimization ensures the extraction accuracy,reduces data redundancy,improves the operation efficiency,and has high reliability and applicability for the extraction of raft culture.(3)The classification method based on feature selection and object-oriented random forest have the best comprehensive evaluation.The overall classification accuracy is 88.8％and κ=0.801,this method can effectively avoid the occurrence of"salt and pepper"noise and can extract the thematic information of raft culture area efficiently and accurately.This study can provide technical and thematic data support for dynamic monitoring and yield estimation of raft culture.
Sentinel-2raft culture areaobject orientedfeature space optimization
国家科技期刊平台
NSTL
万方数据
