Regional evidential patch and restored attribute extraction technology based on FCN-DARG
In the process of surface observation,evidential patches are complex and difficult to distinguish typical features,resulting in poor performance in extracting restored attributes.In order to improve the accuracy of surface observation and monitoring,regional evidential patch and restored attribute extraction technology was studied.By using the fully convolutional network with dual attention residual grouping(FCN-DARG)segmentation algorithm and region growing segmentation method,the minimum rectangular box of the patches was determined.The adaptive region growing method was used to perform secondary segmentation within the rectangular box for coarse and fine segmentation of the evidential patches.Based on the segmentation results,the optimal classification function was applied to classify the corresponding restored attributes of the evidential patches.The peak valley threshold method and error function were combined to extract the normalized features of the blue-green band,spatial features of hue-intensity-saturation(HIS),equalization features,and grayscale corrected color tone features of the evidential patches.The restored attribute features were used as the marked content to complete the extraction of restored attributes.The experimental results of the case study show that the proposed method can accurately segment wetland patches,cultivated land patches,and forest patches in remote sensing images and accurately classify restored attributes.It has a high effectiveness in extracting regional evidential patches and restored attributes.
fully convolutional network with dual attention residual grouping(FCN-DARG)regional evidential patchadaptive region growing methodrestored attribute extraction
万方数据
维普
