Fundamental Research on Electronic Image Recognition of Cylindrical Zno Nanorods Based on Deep Learning

Abstract

ZnO is recognized as one of the most important photonic materials in the blue-violet region due to its straight-width band gap and large excitation bonding energy. Since ZnO nanorod array performs superior optical and field emission properties, a lot of efforts have been made in the fabrication of a vertically ordered ZnO nanorod array. The shape and size of ZnO nanorods have a significant effect on PEC property. In order to efficiently recognize and measure the shape and size of ZnO nanorods, a new method based on deep learning model mask r-cnn is proposed to detect cylindrical ZnO nanorods. The SEM images of ZnO nanorods were used as a data set for training. Adjust the size of the bounding boxes that model generated to make it more suitable for the data set. At the same time, improve the NMS (non-maximum suppression) algorithm to reduce the missing detection rate, and achieve a good detection effect on the SEM images of ZnO nanorods.