Data collecting, analysis, classification methods, and approaches of the road pavement defects detection

Imagining different spheres such as industry, medicine, education, and others undigitized nowadays is impossible. Wide digitalization leads to the significant growth of various data amounts. The data processing and analysis issue became a real challenge. This work is intended to show the author’s vision of the probable methods and solutions for the data array analysis. The article touches on accelerometer data analysis. It was decided to take road pavement defects identification and classification issue as a practical task to show the implementation of the theoretical assumptions. In this paper, the authors demonstrate the application of Recurrent Neural Networks such as Long Short-Term Memory Networks (LSTM), Convolutional Neural Networks LSTM (CNN-LSTM), and Convolutional LSTM (ConvLSTM) in the context of road pavement binary classification (defect or not a defect). Experiments have shown that sophisticated architectures (CNN-LSTM and ConvLSTM) compared to the basic LSTM have an 8-11% larger recall value; at the same time, comparing CNN-LSTM and ConvLSTM, according to the results of experiments, ConvLSTM has up to 13% increase in the precision metric, that is the best result among three Neural Networks described in the paper. Moreover, such aspects as the influence of the accelerometer position on the sensitivity of the sensor and acceleration data sets were also discussed.