关键词: 高分卫星影像, 内陆盐沼湿地, 卷积神经网络, 多尺度叠置分割, 最近邻分类法

Abstract: Based on Landsat 8 image, GF-2 image, DEM data generated by LiDAR data interpolation and census data of geographical conditions in Yanchi, Ningxia, in this paper, the best time phase of inland salt marsh wetland extraction is determined by using Landsat 8 image. Then the best time phase of GF-2 fusion image is segmented by multi-scale overlay, and NDVI, DEM, Tasseled Transformation and other features are selected. The nearest neighbor classifier is used to obtain the information of inland salt marshes, and the inland salt marshes sample database is constructed. On this basis, the convolution neural network method for extracting inland salt marshes from high-resolution satellite images is discussed. The experimental results show that the convolution neural method is suitable for inland salt marsh wetland extraction. Ccompared with the nearest neighbor classification method, the extraction effect of inland salt marsh wetland is significantly improved.

Key words: high-resolution satellite image, inland salt marsh wetland, convolution neural network, multi-scale stack analysis, KNN