Development of an Ontology-Based Technique for Labeling Land Cover Classes with Minimum Utilization of SAR Features

The availability of satellite imagery to "domain" experts, along with advancements in image processing and analysis techniques, has revolutionized numerous fields, enabling better understanding, planning, and management of our planet's resources. However, formalizing the knowledge gained from domain experts is essential for preserving, sharing, and leveraging their expertise. In this context, knowledge representation methods like ontology play a significant role in the development of applications based on satellite image analysis. Land cover labeling is one of the significant applications of satellite image analysis and plays a vital role in various domains by providing valuable information about the Earth's surface. Although several works have been reported to focus on land cover land use classification and labeling, very few are properly documented or formalized. Therefore, in this paper, an LCL (land cover labeling) ontology has been proposed for labeling the land cover classes using PALSAR-I satellite data. The ontology is based on an adaptive algorithm, as the labeling criterion is independent of specific range values and depends on feature image statistics. For algorithm development, four types of features, namely, polarimetric features, texture features, color features, and wavelet features were examined. For selecting optimal feature set, random forest was utilized and, consequently, for further labeling the classes a set of rules has been formed by applying Otsu thresholding on the selected class-wise feature sets. The derived rules were finally formalized to develop the ontology for labeling the land cover classes. The proposed ontology was applied to distinct study sites using PALSAR-I data which resulted in a satisfactory classification accuracy of around 85%.