A lightweight deep learning network based on knowledge distillation for applications of efficient crack segmentation on embedded devices

Abstract

Timely crack detection of pavement helps inspectors access road conditions and determine the maintenance strategy, which can reduce repair costs and safety risks. Deep learning has greatly advanced the development of automated crack detection, but there are still challenges that hinder the application of crack segmentation networks in engineering practice such as the bloated models, the class imbalance problem, and the high-performance device dependency. For efficient crack segmentation tasks, this paper proposes a novel high-performance lightweight network termed multi-path convolution feature fusion lightweight network (MCFF-L Net) and utilize the concept of knowledge distillation. The MCFF-L Net with only 1.18 M parameters achieves F1 score of 85.70% and intersection over union (IoU) of 78.22%, which surpasses the popular heavyweight networks and lightweight networks. The proposed network is further implemented on an embedded device of Jetson Xavier NX and the detection speed of pavement cracks can reach 9.71 frames per second (FPS). The combination of embedded device and proposed lightweight networks is suitable for application scenarios where the portable and efficient crack segmentation is needed and the reliable data transmission through network is not accessible.

A shallow 2D-CNN network for crack detection in concrete structures

“…But as shown in the Related work, less research has emphasized the aspects of execution time and reducing the number of calculations. There are different approaches to achieving this goal, and shallow architectures and deep learning techniques are among these solutions (Chen et al ., 2023; Xu et al ., 2023). For structural health monitoring, light techniques based on deep learning have been done for different aims, including localization of damages (Dipietrangelo et al ., 2023), structural component detection (Agyemang et al ., 2022), semantic segmentation of complex structural damage recognition (Xu et al ., 2023a), facilitating the monitoring and management of structural health monitoring (Martín et al ., 2022), damage detection on a large space antenna (Iannelli et al ., 2022), two-class crack detection (Chen et al ., 2023) and multisensor monitoring tasks (Li et al ., 2022).…”

Fine‐grained crack segmentation for high‐resolution images via a multiscale cascaded network

“…Regarding computational efficiency, Chen et al. (2023) proposed a lightweight network named the Multi‐Path Convolutional Feature Fusion Light Net (MCFF‐L Net). By using knowledge distillation techniques, they deployed MCFF‐L Net on an embedded device, achieving real‐time crack segmentation.…”

A lightweight encoder–decoder network for automatic pavement crack detection

“…For example, J. Chen et al. (2023) designed a lightweight model called MCLD with a parameter size of 0.43 M. Jiang and Zhang (2020) proposed a method that used a wall‐climbing UAV system to acquire crack images and then used a wireless data transmission method to meet real‐time detection requirements. However, even though these models have been specifically designed for practical crack detection scenarios, they can only achieve the detection speed of 9.7 and 6 frames per second (FPS), respectively, which cannot meet the requirement for real‐time applications.…”