Due to the expanding geography of highways, an urgent task is to assess the condition of their tracks. In recent years, various methods of training neural networks have been increasingly used to analyze images and improve their quality. In this regard, it is of interest to compare the capabilities of various neural networks in terms of obtaining a high-resolution image according to the criterion of average time to achieve an acceptable result. The neural networks ESRGAN, EDSR, ESPCN, FSRCNN, LapSRN were selected for analysis, each of which can increase the resolution simultaneously in width and height of the frame by 4 times, and, accordingly, the number of pixels by 16 times. For this purpose, 5 experiments were conducted for the listed networks with 5 different photos in each experiment, while the number of pixels in the image doubled each time. It was found that the ESPCN network has the best indicators in terms of time spent, the FSRCNN network demonstrates comparable results.