Deep Learning-based Pothole Detection and Severity Classification with Location Mapping

Abstract: This paper introduces a deep learning-powered system designed to detect potholes from road images, assess their severity, and map their location. At the core of the system is a Convolutional Neural Network (CNN) trained on a curated dataset of road images containing both potholes and non-potholes. The model is able to accurately identify potholes and further categorize their severity into three levels-Minor, Moderate, or Major-based on its confidence score. To demonstrate the system's functionality, detected potholes are assigned random locations within Bengaluru, and a detailed PDF report is generated. The report includes the detection results, supporting images, a severity distribution chart, and location information. The entire solution is deployed as a Flask-based web application, offering a simple interface where users can upload road images, receive real-time predictions, and download comprehensive reports.

Keywords: Pothole Detection, Deep Learning, Convolutional Neural Networks (CNN), Image Classification, Road Safety, Web Application (Flask).