The article deals with a key challenge in Geographic Object-Based Image Analysis (GEOBIA), lacking a unified approach to describing land cover classes, which leads to ambiguity and low reproducibility of results. To solve this problem, a unified framework for describing land cover classes was proposed; it provides a connection between the EAGLE semantic model and the interpretation features computed in the Trimble eCognition environment. The methodology is based on creating a two-component system: a registry of standard complex classes, ensuring compatibility with the core EAGLE model, and that of non-standard complex classes, which allows users to introduce more detailed, task-specific classes while maintaining a hierarchical relationship with the standard model. Furthermore, the authors present a comprehensive study of the eCognition features efficiency, aimed at reducing their number to be considered in the practical application of the methodology. As a result, a group of the 10 most informative features was identified, and a hierarchy to apply them was proposed. The findings enable avoiding a full-scale search of all available features, focusing on a limited but effective set instead, which significantly accelerates development of classification rules and improves the objectivity and reproducibility of the results