Road Surface Condition Identification with Deep Neural Networks and SVM Classifier

Roads are people's main transportation mode, deeming them an important aspect of worldwide everyday life. However, weather conditions increasingly impact road infrastructure, necessitating improved road safety measures. Identifying road types enhances traffic management and safety, particularly as roads often sustain damage during the rainy season and require restoration that takes time. In many countries, weather conditions also affect road usability. This study proposes a Deep Neural Network (DNN) for automatic road classification Road Surface Images (RSI). ResNet-50 is employed for feature extraction, while additional features, such as Gray-Level Co-Occurrence Matrix (GLCM), correlation factor, and Histogram of Oriented Gradients (HOG) are integrated to improve detection accuracy. These features collectively form the GHR50 model. Next, the collected features are classified using a Support Vector Machine (SVM) classifier and the parameters are evaluated. The proposed GHR50 model achieves 97.39% accuracy in detecting road types, such as dry mud, fresh snow, and water-asphalt smooth, representing a 0.95% improvement over conventional Convolutional Neural Networks (CNNs).