Research on 1D Signal Fast Trend Extracting via Multi-scale Extrema

Current 1D signal trend extracting methods have such disadvantages as low efficiency, poor flexibility and so on. To overcome these problems, a new method of 1D signal fast trend extracting based on multi-scale extrema is proposed. By making full use of time sequence extrema information to establish a binary tree of multi-scale extrema, it avoids the time-consuming process of obtaining Intrinsic Mode Functions (IMFs) via iteratively sifting in traditional Empirical Mode Eecomposition (EMD) method. While obtaining similar results, it greatly improves the computation speed, and it could extract the trend of different scales directly. Simulated and practical signal experiments demonstrates the effectiveness of this approach. By comparing with traditional EMD method and trend filtering method, the results show that the approach could achieve 1 or 2 order of magnitude speedups.