Research on water seepage detection technology of tunnel asphalt pavement based on deep learning and digital image processing

Abstract

To improve the safety of road tunnel pavement, the research established road tunnel pavement water seepage recognition models based on deep learning technology, and a water seepage area extraction model based on image processing technology to finally achieve accurate detection of water seepage on tunnel pavements. First, the deep learning models EfficientNet water seepage recognition model and MobileNet water seepage recognition model were built, the models were trained with the self-collected pavement seepage data set, and the F1 score was introduced to evaluate the accuracy and comprehensive performance of the two models in predicting different categories of water seepage characteristics. Then three grayscale processing methods, the cvtColor function, mean method and maximum method, six global threshold segmentation methods, Otsu thresholding method, THRESH_BINARY, THRESH_BINARY_INV, THRESH_TRUNC, THRESH_TOZERO and THRESH_TOZERO_INV, three filtering methods, namely Gaussian filtering, median filtering and morphological open operation, as well as small connected domain removal, were used to reduce the noise of the images. Finally, the seepage area image calculation method was proposed based on the processed images to predict the actual pavement seepage area. The results show that the recognition accuracy of the EfficientNet water seepage recognition model is 99.85% and 97.53% in the training and validation sets respectively, which is 2.85% and 0.76% higher than the 97% and 96.77% of the MobileNet model. The average F1 score of the EfficientNet model is 95.22%, which is 5.05% higher than that of the MobileNet model, for the four types of seepage feature images: point seepage, line seepage, surface seepage and no seepage. The cvtColor function for grayscale processing, THRESH_BINARY for threshold segmentation and a combination of median filtering and morphological open operation for image noise reduction can effectively extract the seepage features. The area calculation is performed by the seepage area image calculation method, and the average error between the predicted value and the actual seepage area is 8.30%, which can better achieve the accurate extraction of the seepage area.

Performance Evaluation of Convolutional Neural Network Models for Classification of Highway Hidden Distresses with GPR B-Scan Images

“…In the classification task, Wu et al adopted DenseNet and deconvolution network framework for pixel-level detection and combined it with a support vector machine classifier to classify cracks, achieving 98% accuracy [31]. Li et al constructed EfficientNet and MobileNet models, respectively, to identify pavement water seepage problems, and the research results showed that EfficientNet performed well and could accurately detect pavement water seepage [32]. Martinez-Ríos et al used generalized Morse wavelet (GMWs) to perform continuous wavelet transform on vehicle vertical acceleration data for the detection of transverse cracks in the road surface and fine-tuned GoogLeNet, SqueezeNet, and ResNet18 [33].…”