Optimized AI methods for rapid crack detection in microscopy images

Date published

2024-12-06

Free to read from

2025-01-09

Supervisor/s

Journal Title

Journal ISSN

Volume Title

Publisher

MDPI

Department

Type

Article

ISSN

1450-5843

Format

Citation

Lou C, Tinsley L, Duarte Martinez F, et al., (2024) Optimized AI methods for rapid crack detection in microscopy images. Electronics, Volume 13, Issue 23, Article number 4824

Abstract

Detecting structural cracks is critical for quality control and maintenance of industrial materials, ensuring their safety and extending service life. This study enhances the automation and accuracy of crack detection in microscopic images using advanced image processing and deep learning techniques, particularly the YOLOv8 model. A comprehensive review of relevant literature was carried out to compare traditional image-processing methods with modern machine-learning approaches. The YOLOv8 model was optimized by incorporating the Wise Intersection over Union (WIoU) loss function and the bidirectional feature pyramid network (BiFPN) technique, achieving precise detection results with mean average precision (mAP@0.5) of 0.895 and a precision rate of 0.859, demonstrating its superiority in detecting fine cracks even in complex and noisy backgrounds. Experimental findings confirmed the model’s high accuracy in identifying cracks, even under challenging conditions. Despite these advancements, detecting very small or overlapping cracks in complex backgrounds remains challenging. Our future work will focus on optimizing and extending the model’s generalisation capabilities. The findings of this study provide a solid foundation for automatic and rapid crack detection in industrial applications and indicate potential for broader applications across various fields.

Description

Software Description

Software Language

Github

Keywords

40 Engineering, 4009 Electronics, Sensors and Digital Hardware, Machine Learning and Artificial Intelligence, 4009 Electronics, sensors and digital hardware, crack detection, YOLOv8, deep learning, microscopic images, structural integrity

DOI

Rights

Attribution 4.0 International

Relationships

Relationships

Resources

Funder/s