Automated highway pavement crack recognition under complex environment.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Civil Engineering

Background:

• Pavement damage from various factors necessitates effective monitoring for timely repairs.
• Recognizing highway cracks in complex environments is challenging due to interferences.
• Computer vision offers promising solutions for pavement crack detection.

Purpose of the Study:

• To develop and evaluate a novel computer vision approach for highway pavement crack recognition.
• To improve the accuracy and efficiency of pavement crack detection using deep learning.
• To assess the performance of a ResNet34 model integrated with a convolutional block attention module (CBAM).

Main Methods:

• Trained ResNet18, ResNet34, and ResNet50 models using transfer learning.
• Selected ResNet34 as the base model and integrated the CBAM module.
• Conducted further training and evaluated model performance using precision, average recall, and class-specific recall.

Main Results:

• The ResNet34 model with CBAM integration showed improved test accuracy and average recall.
• The proposed model outperformed other evaluated models in performance metrics.
• Achieved high recall rates for transverse (88.8%), longitudinal (86.8%), map cracks (88.5%), repairing (98.3%), and pavement markings (99.9%).
• Attained the highest precision of 92.9% and average recall of 92.5%.

Conclusions:

• Integrating CBAM into ResNet34 significantly enhances pavement crack recognition capabilities.
• The proposed model demonstrates strong potential for highway maintenance applications.
• Further research is needed to address limitations in detecting mesh cracks.