Super-Resolution AFM Imaging Based on Enhanced Convolutional Neural Network

Atomic force microscope (AFM) is a powerful nanoscale instrument, which can obtain the true surface morphology of samples. There are strict requirements for the detailed features in AFM images, so it is necessary to mine the deep information in the images. Nevertheless, the standard AFM scanning process takes a very long time to obtain high-quality images. For most cases, the original images of AFM are with low resolution. In order to get the more detailed texture and feature information as much as possible, a super-resolution convolutional neural network algorithm based on enhanced data set is proposed in AFM imaging. By learning the mapping relationship between low-resolution images and high-resolution images from the image database, the high-resolution image is finally obtained. Aiming at the problem of long scanning time and too small training database of AFM image, adaptive histogram equalization is used to expand and enhance the training set of AFM images. Compared with the traditional super-resolution methods, the subjective and objective evaluation of the reconstructed image verifies the feasibility of the proposed algorithm. Aiming at the problem of long scanning time and too small training database of AFM image, a super-resolution convolutional neural network algorithm based on enhanced data set is proposed in AFM imaging. It can save a lot of time and hardware cost by using adaptive histogram equalization (AHE) and improve the generalization ability of deep learning model to further improve the reconstruction quality of AFM image.