Detection of Potholes using Convolutional Neural Network Models: A Transfer Learning Approach

The most serious issue with the road and highway transportation system is accidental deaths. A pothole is the most prominent cause of poor road conditions, and as a result, road damage is booming in Bangladesh and throughout the world. Using image processing techniques to identify and classify potholes is a very operative term for preventing road accidents and providing a safe journey. As a result, the objective of this study is to identify the pothole using image processing techniques, which is the Transfer Learning methodology. For identifying potholes, two conventional neural network models, Visual Geometry Group (VGG16) and Residual Networks (ResNet50), were used. The major goal of this paper is to recognize potholes and furnish the Road and Highway Department (RHD) with a Computer Vision based solution. For model feeding, several image preprocessing approaches are utilized, and our suggested models achieve a considerable boost in performance, with VGG16 gaining 96.31% and ResNet50 acquiring 98.66% accuracy, respectively. As a consequence, the high accuracy of the proposed models implies that the research study is indeed successful.