SZR5: A Modern Technique to Detect Potholes Using Convolution Neural Network

Roads serve as transportation corridors and are a part of daily life, but irregularities in the surface have an influence on more than just the road itself, including driver safety, car mechanics, and fuel efficiency. In order to evaluate road roughness and find potholes, a number of methods for automated monitoring of the state of the road surface have been developed. The aim of this paper is to provide a deep learning-based CNN model to detect potholes with superior accuracy. The target of the model is to spontaneously detect potholes from normal road surfaces so that it can help reduce accidents and other unwanted situations. Different environments, with varying lighting and weather were also taken into consideration while building the model. Using deep learning methods, the model was trained and evaluated on a dataset of 3000 raw photos, 1500 of which are of smooth roads and 1500 of which are of potholes. To ensure that the model can comprehend any picture input in any circumstance, these photos were enhanced and created into 12000 new images. With our unique CNN model SZR5, we were able to achieve 97.58% accuracy on the training dataset and 97.35% accuracy on the validation dataset.