Improve Semantic Segmentation of Remote sensing Images with K-Mean Pixel Clustering: A semantic segmentation post-processing method based on k-means clustering

Semantic image segmentation has been used to detect objects and label pixels in images. It has been applied to high-resolution remote sensing images to detect different types of terrains and landforms. However, the accuracy of the existing methods is not always satisfactory. Here we propose a semantic segmentation post-processing method using K-mean clustering. Our method aggregates the predictions from network training algorithms such as Unet and HrNet [1], and then performs postprocessing using K-Mean clustering iteratively [2] [3]. The accuracy of our method improves as the number of iterations increases. Source code is at https://github.com/carlsummer/SSK.