Classification of Potholes using Convolutional Neural Network Model: A Transfer Learning Approach using Inception ResnetV2

Rough and damaged roads give poor ride quality which leads to poor transport experience, high travel cost, physical loss to vehicles and passengers, and high vehicle maintenance cost. Therefore, to plan a safe and optimal road trip, prior information of road conditions is most important. Poor road conditions are also responsible for traffic accidents. There exist many road conditions and potholes detected techniques and these techniques can be categorized into two basic categories named as vision-based techniques and vibration-based techniques. In this paper vision-based technique is used for road image classification. For the classification, a transfer learning-based Inception ResnetV2 transfer learning-based Convolutional Neural Network model is used. The selected model is trained and tested on 3211 road images. These images were collected from the public data available on the internet and captured using the camera. Data is classified into three categories plane road, large pothole, and small pothole. The accuracy of the classification is calculated in terms of precision, recall, F1-score, support, and accuracy percentage. According to performed analysis, the Inception ResnetV2 transfer learning-based Convolutional Neural Network model attained maximum accuracy of 94.42 percent with a 0.933 precision value. The performance of the model during the classification process is evaluated using the training and testing loss. Further, the accuracy of the proposed model is compared with Convolution Neural Network and Support Vector Machine using the same dataset. This paper also provides a comparative analysis of the proposed model with other published work.