Data-Efficient Machine Learning for Layer Number Identification of 2D Materials Using Color Difference Analysis

The identification of the number of layers in two-dimensional (2D) materials is critical for exploiting their unique quantum and optoelectronic properties. In this study, we propose a machine learning-assisted approach to classify the number of layers of five different 2D materials based on pixel-level microscopic image data, accommodating diverse experimental conditions. Our methodology leverages RGB and YIQ color spaces to capture the color differences between 2D material flakes and their substrates, with YIQ demonstrating superior performance. Among various models, such as support vector machines (SVM), k-nearest neighbors (k-NN), and decision tree models, we demonstrate that the decision tree algorithm with YIQ data provides the highest classification accuracies across all materials, particularly effective for intermediate layers with values of 2–5. Even with the limited data for training, our pixel-level approach demonstrates a superb layer-number identification capability for up to five layers. Our study highlights the versatility of machine learning-assisted layer-number classification methods with minimal image data, offering a time-efficient and cost-effective solution for identifying various 2D materials with diverse imaging environments.