A Deeply Supervised Convolutional Neural Network for Pavement Crack Detection and Pothole Detection with Multiscale Feature Fusion

Road crack and pothole detection is necessary for effective and economic road maintenance because the early identification of these phenomena can avoid additional degradation and improve driving safety. In this project, we present a deep learning-based method for the automatic identification of road cracks and potholes with the help of convolutional networks (CNNs). For improving detection precision, we present a new multiscale convolutional feature fusion module, which incorporates high-level features to low-level features at various convolutional levels. This fusion allows the model to extract the intricate geometric patterns of both cracks and potholes that are normally hard to find using single-scale features. The model is also deeply supervised, and integrated direct supervision is supported across various convolutional levels, which not only enhances convergence but also enhances the overall performance of the model. By combining multiscale features learned at different levels, the model is stronger in detecting road surface anomalies. To show the effectiveness and generalizability of our method, we test the model on a number of public datasets, demonstrating its capability to correctly identify both road cracks and potholes in varied road conditions. This automated detection system provides a useful tool for efficient road maintenance, minimizing manual inspection efforts and enhancing road safety.