A new global multiplexing structure of original features for road crack detection

ABSTRACT

This paper proposes a novel approach, referred to as Global Multiplexing of Original Features (GMOF), to detect cracks in images. The proposed approach leverages the shallow layers of a network to extract crack-edge details, which are then fused with deep features in later stages to improve detection accuracy. Specifically, GMOF utilises the first stage of a backbone network to extract shallow edge information, which is then resized to match the resolution of the subsequent stages. The original feature is then concatenated with the deep feature map, followed by a one-dimensional convolution to ensure channel consistency between stages. The experimental results on image classification and segmentation tasks show that GMOF helps the network learn fine edge details of cracks, resulting in a maximum improvement of 6.06% (classification) and 15.83% (segmentation) in accuracy. This crack detection method is easy to integrate into existing deep learning frameworks.

KEYWORDS:

• Crack detection
• deep learning
• feature fusion
• backbone networks
• feature multiplexing

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability statement

The data that support the findings of this study are openly available in: https://drive.google.com/drive/folders/1P_QQSDw-d8ErXfrL2kiN0T0CTBdDEUuN.