Road reconstruction or restoration is amongst the most challenging difficulties to elude collisions ,dramatically increased obstruction and minimizing or maintaining upkeep costs .Potholes are generated or created as a result of poor natural situation and significantly very high traffic on highways. Only manual identification of potholes is now applicable which is highly slow and delayed process. The identification of potholes in this work is using on 2 methods which are spectral clustering (sc) and deep learning methods .In one approach, sc and morphological procedures are employed to process the input picture and then the road pothole is identified by making use of a threshold classifier. For spotting road potholes, this method will not require any training. Making use of cnn and alexnet is the other method for identifying road potholes. To test both strategies a balanced and proportional dataset of Three hundred non-pothole and pothole photographs was used. As higher number of photos are needed for deep learning ,training data augmentation is employed for enhancing the dataset size. In comparison to the spectral clustering method the accuracy of lenet and cnn was significantly higher.

Road Pothole, deep learning, TensorFlow, CNN