Weak feature signal extraction for small leakage in pipelines based on wavelet

For solving the difficult problem to distinguish weak signal of small leakage in pipelines, a method was presented to extract feature weak leakage signal from signal with much noise based on wavelet entropy. Not only the Signal to Noise Ratio (SNR) could be increased and the feature was clearer, but also it is not sensitive with the form of signals. The nonlinear adaptive filtering was realized to weak signals according to the different characteristics between the useful signal and noise. Additionally, using wavelet packet elaborate frequency division the signal frequency bandwidth could be decomposed more subtly. The feature vectors of pipeline leakage and normal operation states could be formed based on frequency segment power, which could be used as input samples of neural network to improve detection accuracy to pipeline leakage. The experiment results showed the small leakage would be distinguished and located effectively.