A segmentation method based on the deep fuzzy segmentation model in combined with SCANDLE clustering

Abstract

To enhance the low clustering accuracy of the fuzzy clustering segmentation algorithm for analyzing high spatial resolution remote sensing images (HSRRSIs), a deep fuzzy segmentation model (DFSM)combined with Spectral Clustering with Adaptive Neighbors for Deep Learning (SCANDLE) clustering is proposed. The DFSM is used to over-segment the image, and the automatic coding structure is used to adaptively fuse the image features, minimizing the internal compactness and maximizing the external separability of the clustering, yielding better results. Meanwhile, the SCANDLE clustering model is used to cluster the over-segmentation results, and the matrix construction algorithm for adaptive neighborhood allocation is used to map the frame of the connected layer and optimally combine the over-segmentation images to realize the final segmentation results. The new method can accurately segment HSRRSIs with good segmentation performance.