Rapid Method Using Deep Learning with Multi-focus Microphotographs to Measure Submicrometric Structures and Its Evaluation

Confocal laser scanning microscopy (CLSM) and scanning electron microscope (SEM) systems are commonly used for measuring the dimensions of laser-processed objects. Nevertheless, both methods require some time for preprocessing and measurement, thereby entailing high costs. We propose a simple, fast, and inexpensive method for measuring submicrometric structures using deep learning with multi-focus microphotographs taken using an optical microscope. The average errors in depth and height for a laser-processed groove and a laser-processed ridge are, respectively, 0.1667 m and 0.4349 m. The estimation time is 971.50 ms for the 64 64 m 2 area.