Layered Identification of Two-Dimensional(2D) Material Flakes

2D material is considered as the key to crack the chip performance bottleneck. After preparing 2D material flakes and obtaining their micro-optical images, it is still necessary to rely on manually experience to discriminate the material layers, which is inefficient, high workload, and poor accurate. Therefore, a method based on semantic segmentation is proposed to automatically identify the number of layers of 2D materials in this paper. First, the dataset images are color corrected, then a semantic segmentation model is trained to segment the flakes in the image, and finally a pre-trained Gaussian mixture model is used to fine-tune the semantic segmentation results. The experimental results show that the segmentation IoU results can reach 54.28%, 74.80% and 57.70% for the 1–3 layers of WSe2 2D material and 65.4%, 83.15%, 79.23% and 62.21% for the 1–4 layers of graphene 2D material, which can satisfy the needs of the study.