Classification Methoed of Cement Concrete Pavement Conditions Based on Improved EfficientNet

Given the drawback of high complexity and low recognition rate of the existing cement concrete pavement state classification models and methods, this paper proposes an improved model based on EfficientNet for the classification of cement concrete pavement states in five cases. A more comprehensive and effective feature extraction ability for the model is provided by an attention module for the performance of convolutional neural networks introduced in the model. To reduce the model complexity, an efficient channel attention module is introduced. The improved network model has higher detection accuracy and better adapts to the classification of cement concrete pavement states. The image samples come from the open-source dataset (Road Surface Classification Dataset, RSCD). Through reclassification, a total of 58,568 images of five types of cement concrete pavements in different weather conditions are established as the dataset to train the proposed model and obtain the classification results of cement concrete pavement images. The experimental results show that the accuracy of the proposed model on the dataset reaches 95.57%, which is better than other comparison models and can accurately and effectively classify the cement concrete pavement states under different conditions.