Machine learning assisted graphene detection with polarized light

Machine learning has found a natural application in object identification, particularly when traditional algorithms based on heuristics show reduced performance because of variable ambient conditions. In this study we use machine learning to optimize fluorescence microscopy for graphene detection on silicon wafers. Various algorithms have been proposed in literature to incorporate machine learning to enhance the detection of graphene. However, the average precision (AP) of such graphene detection algorithms is still lower than human based detection. Our work is based on use of a polarized light to acquire images of graphene on Silicon wafer. The absorption of polarized light on graphene is anisotropic and depends on angle of incidence to the carbon-carbon bonds. By rotating the polarization of light, we seek to achieve better accuracy in images acquired to train the neural network. This can potentially be less prone to environmental factors such as changes in lighting conditions, wafer thickness and quality of graphite used. Our goal is to perform real time detection of graphene from such specially acquired polarization dependent microscopic images. This research could allow low-cost, high accuracy graphene detection available to be implemented on any fluorescence microscope.