Abstract:Crack is one of the most common road distresses which may pose road safety hazards. Generally, crack detection is performed by either certified inspectors or structural engineers. This task is, however, time-consuming, subjective and labor-intensive. In this paper, we propose a novel road crack detection algorithm based on deep learning and adaptive image segmentation. Firstly, a deep convolutional neural network is trained to determine whether an image contains cracks or not. The images containing cracks are then smoothed using bilateral filtering, which greatly minimizes the number of noisy pixels. Finally, we utilize an adaptive thresholding method to extract the cracks from road surface. The experimental results illustrate that our network can classify images with an accuracy of 99.92%, and the cracks can be successfully extracted from the images using our proposed thresholding algorithm.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is road crack detection. Specifically, traditional crack detection methods usually rely on certified inspectors or structural engineers, and these methods are time - consuming, subjective and labor - intensive. Therefore, this paper proposes a novel road crack detection algorithm based on deep learning and adaptive image segmentation, aiming to improve the efficiency and accuracy of crack detection. ### Main contributions of the paper: 1. **Image classification**: Use a deep convolutional neural network (DCNN) to determine whether an image contains cracks. The network structure in this part is shown in Figure 1. Through operations such as the convolutional layer, batch normalization, ReLU activation function, max pooling, and fully - connected layer, efficient classification of images is achieved. 2. **Image segmentation**: For the classified images containing cracks, use bilateral filtering to smooth the images and reduce the influence of noisy pixels. Then, extract the crack regions from the images by the adaptive thresholding method. ### Experimental results: - **Image classification**: The experimental results show that the proposed network has an accuracy rate of 99.92% in image classification. - **Image segmentation**: In the pixel - level segmentation performance evaluation, compared with the Otsu thresholding method, the proposed method performs better in precision, accuracy, and F1 - measure. Especially when τ is set to 1, the segmentation effect is the best. ### Formula display: 1. **Bilateral filtering**: \[ i_{bf}(u,v)=\frac{\sum_{x = u-\rho}^{u + \rho}\sum_{y = v-\rho}^{v + \rho}\omega_s(x,y)\omega_c(x,y)i(x,y)}{\sum_{x = u-\rho}^{u + \rho}\sum_{y = v-\rho}^{v + \rho}\omega_s(x,y)\omega_c(x,y)} \] where, \[ \omega_s(x,y)=\exp\left(\frac{(x - u)^2+(y - v)^2}{\sigma_s^2}\right) \] \[ \omega_c(x,y)=\exp\left(\frac{(i(x,y)-i(u,v))^2}{\sigma_c^2}\right) \] 2. **Adaptive thresholding method**: - Assume that the down - sampled image consists of two parts, the foreground (cracks) and the background (road surface), and they can be separated by a threshold δ. - Use the k - means clustering algorithm to find the optimal threshold δ by minimizing the within - cluster sum of squares: \[ \arg\min_S\sum_{i = 1}^2\sum_{m\in S_i}\|m-\mu_i\|^2 \] where \(\mu_i\) represents the mean of the vectors in cluster \(S_i\). ### Conclusions and future work: This paper proposes a new road crack detection method, including a fully - connected neural network for image classification and an image segmentation algorithm based on k - means clustering. The experimental results show that this method performs well in image classification and pixel - level segmentation. Future work can further explore the use of deep neural networks for image segmentation to improve the ability to detect cracks in complex backgrounds.

Road Crack Detection Using Deep Convolutional Neural Network and Adaptive Thresholding

Detection of Road Cracks Using Convolutional Neural Networks and Threshold Segmentation

Deep Discriminant Learning-based Asphalt Road Cracks Detection via Wireless Camera Network

Novel Road Crack Detection Technology Based on Deep Learning

Visual Detection of Road Cracks for Autonomous Vehicles Based on Deep Learning

Structural Crack Detection Using Deep Learning–based Fully Convolutional Networks

Automatic Pixel-Level Pavement Crack Recognition Using a Deep Feature Aggregation Segmentation Network with a scSE Attention Mechanism Module

Vision-Based Crack Detection of Asphalt Pavement Using Deep Convolutional Neural Network

A novel real-time pixel-level road crack segmentation network

Adaptive Canny and Semantic Segmentation Networks Based on Feature Fusion for Road Crack Detection

Deep Learning-based Pavement Cracks Detection via Wireless Visible Light Camera-based Network

Road Crack Detection Using Deep Neural Network Based on Attention Mechanism and Residual Structure

Automatic Crack Detection on Road Pavements Using Encoder-Decoder Architecture

Road Crack Detection by Combining Dynamic Snake Convolution and Attention Mechanism

A Cost Effective Solution for Road Crack Inspection using Cameras and Deep Neural Networks

Automated pavement distress detection using region based convolutional neural networks

Road Crack Detection and Recognition Based on Image Correction of Horizontal Camera and Convolutional Neural Network

A crack detection network with multi-channel attention and enhanced information interaction

Pavement Crack Detection and Segmentation Method Based on Improved Deep Learning Fusion Model

Road crack detection interpreting background images by convolutional neural networks and a self‐organizing map

Detection of Road Crack Images Based on Multistage Feature Fusion and a Texture Awareness Method