Enhancing Road Safety: A CNN-Based Approach to Road Damage Classification

Road damage detection technology is widely utilized as a real-time and reliable detection system for users' safety, to prevent inconvenience for commuters and transportation services, preventing delays and vehicle damage. The use of road damage detection is seemingly progressively essential due to its comprehensive infrastructure management. The objective of this study is to create the best model utilizing Convolutional Neural Network within the set parameters of three classifications of damage. The proposed work involves construction of three (3) CNN models using Python programming including Keras and Tensorflow libraries projected to classify road damages. Each model was constructed with different configuration details which were more focused on the dataset input sizes and convolutional layers. The proponents found out that Model 3 showed the best overall performance overall with highest accuracy, lowest error rate, and best specificity, false positive rate, precision, and F1 score. Model 2 follows closely but is less accurate. Model 1 lags but still makes reasonably accurate predictions. To further assess their performances, the proponents used 9 road images to classify them as normal, cracks, or potholes. Model 1 correctly classified 7 images, Model 2 classified 6 accurately, and Model 3 classified 7 correctly.