Auto-selected rule on principal component analysis in ground penetrating radar signal denoising

Since the principal component analysis algorithm has no robust threshold and dependent on artificial selection of the principal components, an auto-selected rule which is composed of two parts is proposed. One is mean cumulated energy percent standard which makes the mean energy as stable threshold value to discard principal component connected with ground bounce wave. The other part is an improved local energy percent ratio rule to discard the principal component related to background signals. The remaining principal components are reconstructed to denoise image to suppress the ground bounce wave and remove background signals. Experimental results show that the rule can solve the problem and the threshold value is more robust than the one of whole energy percent.