A Deeply Supervised Convolutional Neural Network for Pavement Crack Detection With Multiscale Feature Fusion.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Civil Engineering

Background:

• Road maintenance relies on accurate crack detection for efficiency and cost-effectiveness.
• Convolutional Neural Networks (CNNs) are increasingly used for crack detection.
• Existing methods may struggle with the complex geometric structures of cracks.

Purpose of the Study:

• To develop an improved deep learning model for automatic crack detection.
• To enhance the robustness and accuracy of crack identification in road infrastructure.
• To address limitations of single-scale feature extraction in current methods.

Main Methods:

• Proposed a deeply supervised convolutional neural network (CNN).
• Introduced a novel multiscale convolutional feature fusion module.
• Integrated high-level features into low-level features at various convolutional stages.
• Applied deep supervision to guide feature fusion and improve model convergence.

Main Results:

• The proposed network demonstrated superior performance on three public crack datasets.
• Achieved state-of-the-art results in crack detection, edge detection, and image segmentation.
• Outperformed existing methods in terms of F1-score and mean Intersection over Union (mIU).

Conclusions:

• The novel multiscale feature fusion and deep supervision enhance CNN performance for crack detection.
• The method provides a robust and generalizable solution for road crack identification.
• This approach offers significant improvements for automated road maintenance.