Adaptive Pavement Crack Segmentation with Multitask Learning

Computer Science and Engineering (CSE)

Crack segmentation in pavement infrastructure is crucial for maintaining road safety and optimizing maintenance strategies. Traditional manual inspection methods are labor-intensive and subjective, necessitating automated deep learning solutions. This research introduces a multitask learning framework leveraging the YOLOv8 architecture for accurate and efficient crack segmentation. The methodology includes data collection, preprocessing, model training, and evaluation using key performance metrics such as precision, recall, F1-score, and mean Intersection over Union (mIoU). Domain adaptation techniques are integrated to enhance the model’s generalization across diverse pavement surfaces. The proposed model employs adaptive hyperparameter tuning and a semi-supervised learning approach to improve segmentation performance. Experimental results demonstrate an mIoU of 89.4%, outperforming conventional methods in both accuracy and robustness. The findings highlight the potential of deep learning-based adaptive segmentation for real-world applications in pavement monitoring and maintenance, enabling proactive maintenance planning and cost reduction.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.