Investigation of the Efficiency of Detecting Road Defects Using Neural Networks

This work will allow us to evaluate the applicability of neural networks based on the Unet architecture for the tasks of segmenting pavement defects. The purpose of the article is to show the effectiveness of using this neural network to solve this problem, as well as to compare the quality of image segmentation for this problem. The leading research method is the empirical method, which allows you to compare the results of different algorithms. In this work, it was shown that the difference in the efficiency of using Unet with attention gates without them is different. Today, various methods of machine learning are increasingly being used to find solutions to a variety of problems. One of such important problems is the analysis of images and finding certain signs on them. For this article, the actual problem is the segmentation of pavement defects, which include various types of pavement cracks. Study of the effectiveness of neural network technologies for road surface image segmentation. The paper explores several types of neural networks aimed at image segmentation. In other works, this type of neural networks has shown its high efficiency for solving the tasks. This method will allow evaluating the applicability of neural network data for solving the problem, as well as comparing the effectiveness of these methods for test images of the roadway. It will also allow you to outline the direction of further work to improve neural networks. In this paper, the quantitative result of the work of neural networks on domain-specific data is obtained, as well as conclusions are drawn on the use of a neural network in this area. This work is of practical importance, making it possible to evaluate the efficiency of defect segmentation in roadway images.