Separation of Cloud/No-Cloud Regions in Satellite Imagery Using a Variation ofHierarchical Clustering Analysis

摘要

This study investigated the usefulness of personal-computer-based softwareapplying hierarchical clustering theory to try to separate cloud-covered regions from clear regions using Automated Picture Transmission imagery from the National Oceanographic and Atmospheric Administration's Television Infrared Observation Satellite. The algorithms were developed in Turbo Pascal, Version 6, and are part of the Training Software Image Processing program developed by a professor at the Air Force Institute of Technology. The goal of the project was to see if hierarchical clustering could provide better separation of cloud/no-cloud regions than an existing technique, histogram thresholding, while running on a personal computer. Results of the research indicated that it was possible to use a centroid based clustering algorithm to separate cloud-covered regions from clear regions in APT imagery. Seed points were used to start the clustering process. The cloud seed point was chosen to be the brightest pixel in the clustering area. Typical results showed that the automated clustering approach provided results within 15 to 20 percent of those obtained from the histogram method.