Classification of Potholed Roads Using Deep Convolutional Neural Network Model Based on AlexNet

This research is dedicated to addressing the detection and classification of potholed roads, a problem faced in the field of traffic safety and urban maintenance today. By adopting the AlexNet model implemented in PyTorch based on deep learning, a network structure with 5 convolutional layers and 3 pooling layers has been successfully constructed, which can efficiently and accurately extract features of potholed road images. The model introduces nonlinear processing, including the Relu activation function, LRN local response normalization function, and Dropout layer, achieving precise extraction of pothole features. Furthermore, the data preprocessing of road images and optimization of the training model have achieved satisfactory accuracy on the validation set. Finally, the images to be predicted are preprocessed and input into the trained model for classification through a loop structure, ultimately achieving an accuracy rate of 91.752%. This comprehensive solution provides feasibility for automated detection and classification of potholed roads, and is expected to yield significant benefits in the fields of urban maintenance and traffic safety.