Application of Natural Gas Pipeline Leakage Detection Based on Improved DRSN-CW
Hongcheng Liao, Wenwen Zhu, Benzhu Zhang, Xiang Zhang, Yu Sun, Cending Wang, Jie Li
Abstract
Aiming at solving the natural gas leakage detection issue, we propose an improved method based on deep residual network with channel-wise thresholds (DRSN-CW) to improve the detection accuracy with GPLA-12 dataset. In the approach, larger and unequal convolution kernel size are designed in all convolution layers to extend the receptive field in the process of extracting fault feature. Moreover, considering that datasets of natural gas pipeline leakage typically contain large amounts of ambient noise, the soft threshold module of DRSN-CW is combined with designed kernel size to reduce the influence of noise on accuracy of gas pipeline leakage detection. Compared with the-state-of-art techniques (e.g., CNN, DRSN-CW and DRSN-CS), experimental results show that our method outperforms the compared methods.