Denoising method for terahertz signal using RBF neural network with adaptive projection learning algorithm

Radial basis function (RBF) neural network has the ability to eliminate the terahertz (THz) spectrum’s noise via its robust feature removal ability. Unfortunately, this method has some disadvantages, such as difficulty in obtaining clean training data, smaller quantity of training data, lower spatial resolution, and an inadequate restoration impact. Aiming at the problem of low spatial resolution in the imaging process caused by the beam diffraction of terahertz signals, this paper uses the RBF neural network of adaptive projection learning algorithm to process the image. The proposed method extracts the edge image and non-edge image of the terahertz pulse via an adaptive projection learning algorithm using the RBF neural network to denoise the non-edge images. In addition, it synthesizes the image to obtain the terahertz pulse image. Experiments show that using this method for processing the images, the difference between the denoised image and the original image is much smaller, and it has a higher spatial resolution, which can effectively identify the target.