Learning-Based Autofocus for Scanning Electron Microscopy

Fast and reliable autofocusing methods are important for automatic nano-objects positioning tasks with a Scanning Electron M.lcroscopy (SEM). However, many operating parameters of SEM cannot be set before image capture begins so that It's hard to Implement an autofocus In this paper, we proposed a best focus image prediction method based on machine learning. To predict the best focus in SEM, the performances of different sharpness functions including Sobeledge detector, Canny edge detector and discrete cosme transform were compared; the gradi ent of curves was used as the input features of the prediction model. Least squares support vector machines (LS- SVM) and back-propagation neural network (BPNN) methods were appl ied to develop.quantitative models. By comparing the results of different models, the LS- SVM was better than BPNN to predict the best focus in SEM for silicon sample, and chessboard sample. It can be concluded that image processing combined with learning methods IS attractive for standard prediction of the best focus performing tasks using SEM.