Evaluation and optimization of image segmentation models in pavement crack detection

Yuheng Liu

doi:10.1117/12.3055739

9 January 2025 Evaluation and optimization of image segmentation models in pavement crack detection

Yuheng Liu

Author Affiliations +

Proceedings Volume 13486, Fourth International Conference on Computer Vision, Application, and Algorithm (CVAA 2024); 1348624 (2025) https://doi.org/10.1117/12.3055739
Event: Fourth International Conference on Computer Vision, Application, and Algorithm (CVAA 2024), 2024, Chengdu, China

Abstract

Pavement cracks are a critical aspect of road maintenance, and their timely and accurate detection is essential for road upkeep. This study aims to evaluate and optimize the application of image segmentation models in pavement crack detection to enhance the efficiency and accuracy of road maintenance. The research methods include experimental comparisons of five mainstream image segmentation models (U-Net++, MANet, FPN, LinkNet, and PAN), and further improvement of model performance through hyperparameter optimization. The main results indicate that the U-Net++ model performs best in terms of the Dice coefficient, with an average Dice score of 0.715 when the batch size is 32. After optimization of the encoder, optimizer, and learning rate, the final Dice coefficient increased to 0.734. The analysis shows that the superior performance of the U-Net++ model is attributed to its improved skip connections and cascade modules, as well as its effective feature fusion capability.

(2025) Published by SPIE. Downloading of the abstract is permitted for personal use only.

Citation Download Citation

Yuheng Liu "Evaluation and optimization of image segmentation models in pavement crack detection", Proc. SPIE 13486, Fourth International Conference on Computer Vision, Application, and Algorithm (CVAA 2024), 1348624 (9 January 2025); https://doi.org/10.1117/12.3055739

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

;

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE