Two-dimensional small leak detection of pipeline based on time sequence coding

In this study, a two-dimensional pipeline small leak detection based on time series imaging is introduced. First, the sampling points and data of pipeline signals are converted into polar diameters and polar angles in polar coordinate system to form vectors, and then the inner product between vectors is mapped to 2-dimensional pixels through the use of Gram angular field (GAF). The resulting image preserves both the temporal and spatial features of the original signal. Wavelet threshold (WT) denoising and adaptive segmentation constant approximation (APCA) are used to denoise and fit the original signal in order to ensure the noise immunity and computational efficiency of the imaging. The image generated after WT-APCA-GAF processing is input into Swin Transformer for training, and the final identification accuracy of small leakage can reach 100%, effectively avoiding the problem of missing and false positive in detecting small leaks.