A novel pavement transverse cracks detection model using WT-CNN and STFT-CNN for smartphone data analysis

Figures & data

Figure 1. Framework of a novel pavement transverse cracks detection model.

Figure 2. The architecture of the proposed CNN model.

Figure 3. Experimental area.

Figure 4. Layout of equipment.

Figure 5. Schematic diagram of the signal labelling method.

Figure 6. Performance comparison between smartphone sensor and professional sensors.

Figure 7. Classification basis examples of vibration signal.

Figure 8. Raw acceleration data.

Figure 9. Spectrum of STFT time–frequency analysis.

Figure 10. Spectrum of WT time–frequency analysis.

Figure 11. Training and validation accuracy and loss with ImageNet pre-trained VGG-16 DCNN deep image extractors with STFT feature for transverse cracks detection.

Figure 12. Training and validation accuracy and loss with ImageNet pre-trained VGG-16 DCNN deep image extractors with WT feature for transverse cracks detection.

Table 1. Samples of each dataset.

Method	STFT-CNN			WT-CNN
Categories	Transverse cracks	Manhole	Normal	Transverse cracks	Manhole	Normal
Number of training samples	1792	1792	1792	1792	1792	1792
Number of validation samples	768	768	768	768	768	768
Total	2560	2560	2560	2560	2560	2560

Figure 13. Confusion matrix.

Table 2. Comparison of pavement transverse cracks detection results using WT-CNN and STFT-CNN.

Standard for evaluation	WT-CNN	STFT-CNN
Accuracy	97.53%	91.02%
Precision	94.72%	88.15%
Recall	98.05%	92.93%
F1-sore	96.35%	89.56%

Table 3. Comparison with pavement crack method.

Methods	Data type	Accurancy (%)	Equipment cost (estimate)
Zhang et al. (2018a)	RGB image (Kinect-Based)	89.09	$199
Zhou and Song (2020)	laser-scanned range images	99.6 (average)	none
Zhang et al. (2018b)	3D laser scaning	Over 98	none
Ours method	vibration signal	97.20	none

Zhang, Y., et al., 2018a. A kinect-based approach for 3D pavement surface reconstruction and cracking recognition. IEEE Transactions on Intelligent Transportation Systems, 19 (12), 3935–3946.

 Web of Science ®Google Scholar

Zhou, S. and Song, W, 2020. Deep learning-based roadway crack classification using laser-scanned range images: A comparative study on hyperparameter selection. Automation in Construction, 114, 103171. https://doi.org/10.1016/j.autcon.2020.103171.

 Web of Science ®Google Scholar

Zhang, D., et al., 2018b. Automatic pavement defect detection using 3D laser profiling technology. Automation in Construction, 96, 350–365.

 Web of Science ®Google Scholar