Knowledge Bank

With the rapid development of 2D materials and their applications in the industrial area in recent years, the automatic search for mechanically exfoliated van der Waals layers via optical imaging has become a crucial part of improving efficiency. However, currently the Deep Learning method widely used by computer vision and the traditional Machine Learning method cannot be adapted to materials whose color is too similar to the background. Also, very large image data sets are required for training a model. In this paper, we introduce the Number of Cluster (NoC) algorithm to quickly calculate the target k value required by the K-Means clustering algorithm. This research can not only be used for the identification of two-dimensional materials on SiO2/Si substrate, but also widely used for color extraction of various continuous/discontinuous areas under a single-color background. Eight pictures with different number of discrete color areas were used to test the time and accuracy of k value calculation by the gap statistic method, elbow method, and NoC method respectively. The result shows that the NoC method runtime only takes 0.23 seconds to test with 52*52 discrete area images.