The paper proposes a deep neural network architecture based on the integration of the convolutional neural network Faster R-CNN with the Feature Pyramid Network module. Based on this approach, an algorithm for detecting and classifying vehicles in images and a corresponding model have been developed. A cross-platform environment ML.NET was used to train the proposed model. The results of comparing the effectiveness of the proposed approach and convolutional neural networks YOLO v4 and Faster R-CNN are presented. The improvement of the accuracy of detection and localization of different types of vehicles in ultra-high resolutions images is shown. Examples of processing ultra-high resolutions remote sensing images and appropriate recommendations are given.

No abstract