Enhancement over DBSCAN Satellite Spatial Data Clustering

Abstract

Image processing is a promising technique for enhancing images or extracting useful information from them. One commonly used density-based clustering algorithm is DBSCAN (Density-Based Spatial Clustering of Applications with Noise). However, DBSCAN struggles with satellite images due to their large sizes, often resulting in excessively long computation times. This research proposes an improved version of DBSCAN called “Enhanced DBSCAN-based Histogram” (EDBSCAN-H) to address these issues. EDBSCAN-H enhances DBSCAN by incorporating a histogram-based approach to better manage large datasets and reduce computation time. The key improvement lies in using the histogram of the input image and measuring the distance between data objects and histogram points to determine whether a region is dense or sparse, thereby selecting suitable parameters. EDBSCAN-H introduces an additional parameter, ε2, alongside the original DBSCAN parameters ε₁ and MinPts. This enhancement allows EDBSCAN-H to achieve improved performance across various metrics for clustering spatial data images.