Improvement of Segmentation Performance for Feature Extraction on Whirlwind Cloud-based Satellite Image using DBSCAN Clustering Algorithm

Images contain a lot of information that can be used in a variety of areas. One of the images that have much information inside is satellite image. In order to extract the information properly, the image processing step should be performed properly. The segmentation process plays an important role in image processing, especially for feature extraction. Many ways were developed to perform the segmentation image. In this study, we apply DBSCAN clustering to segment images on whirlwind cloud feature extraction problems. DBSCAN is a density-based classifier method which means it is suitable to group a density-based data. While the image used in the segmentation process is the Himawari 8 satellite image which also contains density-based data. It contains various information about clouds condition like cloud type, cloud temperature, cloud humidity, rainfall potential based on cloud temperature, etc. This study uses Himawari 8 satellite images as input where the images taken are images several hours before a wirlwind event in an area, while the cluster method used is the DBSCAN algorithm. Clustering is done to get the extraction features of a wirlwind in the form of centroid points that characterize the movement of a cloud. Segmentation performance was observed based on the number of centroid points as a result of clustering several types of clouds in an area before a wirlwind occurred. Based on segmentation testing using the DBSCAN algorithm for cloud data in an area for several hours before a wirlwind, better segmentation performance was obtained compared to the segmentation results of the Meng hee heng k-means algorithm for the same test data specifications. DBSCAN separates a type of cloud in more detail that makes it easier to record each centroid of each cluster around the scene. It is even able to cluster small groups of clouds independently so that these small groups of clouds can also be detected as features.