Study on intelligent recognition of urban road subgrade defect based on deep learning

Area of Science:

• Geotechnical Engineering
• Civil Engineering
• Artificial Intelligence

Background:

• Urban road subgrades face increasing defects and safety risks due to diversifying types.
• Non-destructive testing (NDT) is crucial for assessing subgrade health and preventing incidents.

Purpose of the Study:

• To develop an intelligent system for identifying urban road subgrade defects using geological radar.
• To evaluate the effectiveness of deep learning algorithms for analyzing geological radar data.

Main Methods:

• Utilized GprMax software for forward simulation of multi-layer subgrade models with defects.
• Created a geological radar subgrade defect image database using simulated and field data.
• Applied and compared four improved Faster R-CNN deep learning algorithms for defect detection and classification.

Main Results:

• The faster_rcnn_inception_v2 algorithm demonstrated superior performance in recognizing subgrade defects.
• Key metrics like loss value, region identification, and accuracy were used for algorithm comparison.
• A robust database was established for training and validating the deep learning models.

Conclusions:

• Intelligent identification of urban road subgrade defects is achievable using geological radar and deep learning.
• The faster_rcnn_inception_v2 model is well-suited for NDT of road subgrades.
• This approach enhances road safety and infrastructure management through advanced diagnostics.