Classification of Pothole Laser Images Based on the Improved ResNet18 Network

This paper improves the pothole laser image classification model based on the ResNet18 network through transfer learning and the introduction of the SE channel attention mechanism. The study utilizes a self-made road inspection vehicle to collect a laser image dataset of road surfaces, which is then cropped and labeled. Experiments show that the improved SEResNet18 model outperforms the original ResNet18 and other mainstream networks in terms of classification accuracy and loss values, verifying the effectiveness of the SE module in enhancing classification performance. The research aims to improve the accuracy of pothole detection and provide technical support for road maintenance.