Pothole Classification Using DenseNet Model: An Empirical Analysis with CNN and InceptionResNetV2

Road conditions have a big impact on the effectiveness and safety of road transport. Potholes, cracks, and bumps in the road may impact driving behaviors and increase the risk of accidents and casualties. To improve the driving experience, proper road maintenance procedures are required. This research paper focuses on the DenseNet model to classify potholes using road image data. A self-generated dataset of 1721 road images with 4 classes is used to train the model. The model is tested with different dataset division strategies and epoch counts. The model provided a maximum accuracy of 98.3% on 30 epochs with a 90% training size and a 10 percent testing size. To further check the effectiveness, the DenseNet model is compared with the Convolutional Neural Network (CNN) and InceptionResNetV2 with the same dataset.