Methods and Tools for Developing Decision Rules for Classifying Objects in Aerial Images

Automatic processing and identification of objects in aerial images is in demand in a wide range of practical tasks, for example, in the land use control process. One of the popular approaches to automatic object detection is the object-based classification using decision rules considering various features of image objects. The effectiveness of this approach is largely determined by the reliability of the created decision rules. The purpose of this work is to increase the degree of automation of the process of analysis and search for decision rules by experts and users of aerial image processing systems. The proposed approach consists in the search for dependencies between feature sets, constraints, and the degree of reliability of the decision rules by converting quantitative feature values into qualitative ones based on histograms, fuzzy clustering of a training sample set, and identifying the “class-features” dependencies. We experimentally show that the proposed approach allows obtaining more reliable classification rules compared to the traditional method, in which the means of supporting the analysis process are limited.