The use of generative neural network models in industry is becoming increasingly relevant and cost-effective. Considerable attention is paid to improving the quality of remote sensing and aerial photography using unmanned aerial vehicles in the field of Earth exploration and land management. This study aims to assess the accuracy of cartographic products obtained from aerial photography, the original images of which are processed by the Real-ESRGAN and SwinIR resolution enhancement models. A review of the most relevant and up-to-date studies has been conducted, the authors of which have asked the question of creating optimal neural network models to improve the quality of satellite and aerial photography. The architectures of the models used, a detailed algorithm for performing and photogrammetric processing of the survey are described. Particular attention is paid to the assessment of the obtained images - a comparative analysis of the metrics of the original and generated images is performed, it is found that good IQA metrics do not always provide a high-quality result during photogrammetric processing. The accuracy indicators of the constructed orthophoto plan and digital terrain model are considered separately. The obtained result allows us to state that the use of generative models possesses the potential in the field of aerial photography: the resolution of cartographic products increases, small terrain objects are detailed, and image distortions are eliminated.