Multi-view Road Disease Detection Based on Attention Fusion and Distillation

Mo, Jiadi; Wang, Yue; Yu, Zhi; Wang, Yangyang; Yan, Shoujing

doi:10.1109/eebda53927.2022.9744763

Cited by 2 publications

(2 citation statements)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Model comparison experiment. The heavyweight networks and lightweight networks which are often used for comparison of crack segmentation networks are trained on the same dataset for the comparison with MCFF-L networks 16,30,34,52,[58][59][60][61][62][63][64][65][66] and the detection effects and evaluation indicators are shown in Figures 11 and 12 and Tables 9 and 10, respectively. Three parameters for the evaluation of model size are also listed in Tables 9 and 10, where the model parameter and the file size are measures of model volume, and the floating-point operations (FLOPs) is a measure of model complexity.…”

Section: Evaluation Experimentsmentioning

confidence: 99%

“…30 Even lightweight networks are still very difficult to deploy directly and perform efficient segmentation tasks on embedded devices. [31][32][33] Moreover, the lightweight network for crack segmentation has not received enough attention as in the crack classification and crack object detection tasks, 34,35 so the development of lightweight crack segmentation network suitable for embedded devices is anticipated.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

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

Chen

Liu

Hou

2023

Structural Health Monitoring

View full text Add to dashboard Cite

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.

show abstract