Abstract:Aiming at the over-segmentation and fuzzy edge sensitivity problems of the R-cut (Ratio cut) edge detection segmentation model for highresolution remote sensing image segmentation, a multi-scale R-cut (Multi-scale ratio cut, MSR-cut) method was proposed for remote sensing image edge detection and segmentation. Firstly, the watershed segmentation algorithm of morphological reconstruction was used to over-segment the image to form multiple super-pixel regions; then the texture feature information entropy value, spectral feature and neighborhood mean difference normalized value of each region of the image were calculated and extracted, and the same was performed respectively. Effective measurement of qualitative and heterogeneity was made; and an evaluation function was constructed to obtain the optimal segmentation scale, and initially merged these super-pixel regions to obtain the rough segmentation result of the image; finally, combined the boundary weight information of various objects, and R was used from a global perspective. The R-cut method further merged the coarse segmentation results, the fine segmentation of the image was completed, and the final segmentation result was generated. The experiment selected high-resolution remote sensing images of different scenes, and the method was compared and analyzed with traditional R-cut edge detection segmentation, Spectral-Rcut edge detection segmentation and Textured-Rcut edge detection segmentation methods by using qualitative and quantitative methods. Experimental results showed that the MSR-cut edge detection segmentation method can effectively improve segmentation accuracy, enhance noise robustness, and achieve better segmentation visual effects.
