Road Flaw Detection using CNN and Transfer Learning Techniques

Road cracks can have very dangerous results. Road flaw detection is a crucial aspect of ensuring road safety and maintenance. Identifying cracks and taking the necessary steps are requirements for preventing such dangers. Manual approaches are costly and time consuming. Using pricey laser equipment vehicles for operation and the associated upfront costs is also not very feasible. This paper proposes an automation-based approach to identify various types of flaws like potholes, cracks, and patchy terrain on road using machine learning techniques. The convolutional neural network (CNN) and transfer learning techniques are utilized to carry out the mentioned work. To get the precise features from the image data, first the data is preprocessed and then transfer learning is applied using different pre-trained CNN models. Thereafter, to accurately detect the road flaws, the model is fine-tuned using the transfer learning and the dataset collected by us. The proposed approach is evaluated on metrics such as precision, recall, and F1 score. The performance results show that the proposed approach performs well while achieving an F1 score of 95%. Therefore, reducing the risk of accidents and improving road safety.