Automatic Feature Selection for Operational Scenarios of Satellite Image Understanding using Measures of Mutual Information

Abstract

The Earth Observation processing tools operating in the recent scenario need to be tailored to the new products offered by the sub-meter spatial resolution imaging sensors. The new methods should provide the image analysts the essential automatic support to discover relevant information and identify significant elements in the image. We advocate an automatic technique to select the optimum number features used in classification, object detection and analysis of optical satellite images. Using measures of mutual information between the target classes and the available features, we investigate the criterions of maximum-relevance and maximum-relevance-minimumredundancy for automatic feature selection at very-low cost. Following a comprehensive set of experiments on multiple sensors, applications and classifiers, the results demonstrate the possible operational use of the method in future scenarios of humanmachine interactions in support of Earth Observation technologies.