Clustering Segmentation Of Visible Remote Sensing Image Based On Local Correlation

Remote sensing images contain a wealth of ground object information.Segmentation and extraction of key targets in remote sensing images can provide great research value in many fields.Image segmentation technology can achieve the segmentation of key targets.Clustering is widely used as a classic algorithm in the field of image segmentation.However,when processing remote sensing images with interlaced features and complex texture information,it is often segmented at the adjacent locations of different features.The effect is blurry.At the same time,before using the clustering algorithm to segment the remote sensing image,it is necessary to specify the number of clustering categories based on human experience,and the program cannot be automated.This paper proposes that the clustering segmentation technology based on local correlation can effectively make up for the shortcomings of the existing algorithms and improve the algorithm’s segmentation accuracy of remote sensing images.The specific research mainly includes two parts:(1)Propose an initial cluster center selection method based on the mean distance.Aiming at the two problems to be improved,the random generation method of the initial cluster center point of the clustering algorithm and the need to manually specify the number of cluster categories.This method selects two sample points with the highest sample density in the sample high-density area,and selects the initial cluster center point at the average Euclidean distance of the two sample points.At the end of the algorithm,the final number of cluster center points is used as the number of clusters in the image.Compared with the comparison algorithm,this method can accurately characterize the distribution of sample points,and the selected initial cluster center point is closer to the true value,thereby reducing the number of iterations of the algorithm,improving the efficiency of the algorithm,and achieving program autonomy Determine the number of clusters.Through repeated experiments to verify the two data sets of SCCTS and 2D15,the improved method proposed in this article is superior to the comparison method introduced in the article in terms of algorithm effectiveness and operating efficiency.(2)Propose a local correlation model.When the clustering algorithm processes remote sensing images with complex distribution of ground objects,there are serious blurring problems at the edge of ground object segmentation.This method uses the constructed local correlation model to characterize the correlation degree of the neighboring pixels in the local area to the central pixel.This model can fully mine the local correlation between pixels,and then eliminate the ambiguity at the boundaries of different types of remote sensing images.Based on the improved algorithm fused with the local correlation model,the objective function of the algorithm is modeled into different sub-problems.The weighted control factor is used to balance the two control items of image detail preservation and image noise suppression,and then a balanced solution in terms of image detail preservation and noise suppression is obtained.Through a large number of remote sensing image segmentation experiments,it is verified that the algorithm proposed in this paper can effectively solve the blur problem at the segmentation boundary,and it is suitable for most remote sensing images.