Software Implementation of the Algorithm for Monitoring Technogenic Objects by Airborne Robotic Systems

The paper provides a substantiation of the relevance of the task of monitoring technogenic objects by airborne robotic systems: an overview of existing recognition algorithms and implementations is made; given a mathematical description of algorithms and machine learning methods used for quick recognition of specified objects in aerial photographs; own solution is proposed based on the analysis of the positive aspects of the current methods. The software implementation of the algorithm has been carried out, which can be conditionally divided into several parts: selection of the most efficient model; creation of a special data structure; training sample preparation; training the model. This algorithm was tested on specific images taken as a result of shooting from an unmanned aerial vehicle, and a comparative analysis was carried out using the data obtained. The object of testing was a trained Faster R-CNN model on two classes of objects: airplanes and helicopters.