A novel transfer learning for recognition of overlapping nano object

Yuexing Han, Yuhong Liu, Bing Wang, Qiaochuan Chen, Leilei Song, Lin Tong, Chuanbin Lai, Akihiko Konagaya

Abstract: While the science and technology of nanostructure have rapidly been developed in many fields, it is still hard to obtain sufficient samples of nano objects due to high cost, thus, impeding the development of deep learning approaches to the material fields. Here, we develop a novel approach to recognize nano objects in Atomic Force Microscope (AFM) images. First, a noise reduction method based on the Laplacian of the Gaussian(LoG) is represented to denoise the AFM images. Then, two improved methods based on the watershed algorithm are proposed to segment the overlapping objects. Finally, a CNN recognition model based on transfer learning which is pre-trained on a large scale of shapes of handwritten numbers and letters is built to recognize the nano objects in AFM images. These methods can resolve effectively the small sample problem of AFM image processing.