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

This paper proposes a novel pavement transverse crack detection model based on time–frequency analysis and convolutional neural networks. The accelerometer smartphone installed in the vehicle collect vibration response between wheel road, such as cracks, manholes, normal pavement. Since original signal can only contain one-dimensional domain (time–acceleration). Time–frequency analysis, including Short-Time Fourier Transform Wavelet Transform, transfer into two-dimensional time–frequency-energy spectrum matrix. energy matrix obtained from STFT WT effectively obtain different features terms of time frequency features. If are further combined with CNN models, STFT-CNN WT-CNN, respectively, cracks be detected more accurately. In this study, reliability developed was evaluated data collected by conducting road driving test. Analysis results show that accuracies WT-CNN 97.2% 91.4%, respectively. F1 scores to analyse practicability adaptability 96.35% 89.56%,