GPR-based detection of internal cracks in asphalt pavement: A combination method of DeepAugment data and object detection. (30th June 2022)

• Record Type: Journal Article
• Authors: Liu, Zhen; Gu, Xingyu; Wu, Wenxiu; Zou, Xiaoyong; Dong, Qiao; Wang, Lutai
• Is Part Of: Measurement. Volume 197(2022)
• Publication Date: 2022-06-30
• Subjects: Internal cracks of asphalt pavement -- Data augmentation -- DeepAugment strategy -- Object detection -- Bayesian optimization; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2022.111281
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A DeepAugment method for searching augmentation strategies tailored to GPR images using Bayesian search is proposed. Geometric and color transformations are selected to realize data augmentation based on the features of GPR images. The improvement effect for detecting internal cracks using data augmentation is analyzed by a self-built GPR dataset. Abstract: The lack of data and poor quality of ground penetrating radar (GPR) images have limited the development of the object detection for internal cracks in asphalt pavement. To address this issue, this paper proposed a 'DeepAugment' data augmentation strategy combined with object detection models. First, the characteristic of internal cracks was determined with numerical simulation and GPR field test, which was in accordance with the coring results. Subsequently, the proposed DeepAugment method was used to enhance the crack features. Object detection results showed that the recognition accuracy and confidence level of internal crack detection improved by using the object detection model to test the enhanced GPR images, which was more noticeable in the YOLOv3 model. The proposed method is found to be of significance for accurately identifying internal cracks in GPR images, and the recognition accuracy after data enhancement can meet the needs of road maintenance engineering.