A denoising method based on kernel principal component analysis for airborne time domain electromagnetic data

Airborne time domain electromagnetic (ATEM) data are often contaminated by natural noise and culture noise during the flight survey. If an inappropriate filter is taken to denoise for airborne electromagnetic data, it will result in low quality of data, less accuracy of the inversion and even wrong interpretation. This paper presents a denoising method based on kernel principal component analysis. Firstly, kernel principal component analysis is applied to extract the principal component of the raw decay curves; then the components reflecting subsurface media distribution and noise may be separated using the energy ratio method; finally, these components reflecting subsurface media are employed to reconstruct the decay curves. The denoising method proposed not only can remove natural noise such as spikes or sferics, but also effectively suppress the culture and shaking noise. In the experiments, we use the proposed denoising method and AeroTEM software respectively, to process the same helicopter time domain electromagnetic data measured in site. The experimental results show that the denoising method based on the kernel principal component analysis (KPCA) proposed in this paper is superior to the filtering method of the AeroTEM.