An Intelligent Convolutional Neural Network based Potholes Detection using Yolo-V7
Madarapu Sathvik, G. Saranya, S. Karpagaselvi
Abstract
Road travel is one of the most common modes of transportation in the world, where more than 60% of the population commutes by personal or shared automobiles. According to a poll, potholes are a major contributing factor in several incidents. Pothole detection techniques have been created to address these issues, including the use of sensors and many others, but they are actually expensive to produce and difficult to put into practice. As a result, solid strategy that utilizes CNN is developed. According to the evidence presented, potholes are the principal factor responsible for the degradation of roadways. It is essential to give some thought to the question of how to locate potholes in the most efficient and economical way. Convolutional neural networks, often known as CNNs, have the ability to filter through vast volumes of data and extract the aspects that are most relevant to their purpose. YOLOv7 was used to annotate and train a pothole image dataset for this research, and the findings were analyzed in terms of recall, accuracy. The model was validated by examining a wide range of photographs relating to potholes. Our model had given a F1 score of 0.51 and this increases proportionally (more the value of F1 score more efficient is the training model) with the number of epochs increased.