Machine Learning for Microscopy Data Analysis: Toward Real-time Optical and Electrical Characterization of Sub-micron Materials

(Micro)spectroscopy often generates various output signals due to intrinsic inhomogeneity of material arrangement at low dimensions or machinery drift, albeit the bulk composition and experimental parameters remain constant. In fact, such diversity can be harnessed to measure material’s purity, unveiling various concealed features via statistical inspection of heterogeneous signals acquired from several microscopy scans. However, the approach requires efficient categorization of a substantial number of signals, which is currently encumbered by laborious calculations, computational hurdles, and manual intervention. This necessitates a programmed interface to perform time-efficient big data analytics, lack of which has perpetually widened the schism between laboratory and industrial-scale microscopy-based assessment of nanomaterials. We present a robust technique - an unsupervised machine learning driven module for automatic clustering and class-wise power spectral density calculation of real-time microscopy signals. Our methodology has been tested across different aspects of wide-field fluorescence imaging and scanning tunnelling spectroscopy, demonstrating the versatility. Additionally, we investigated the impact of data-processing on the clustering efficiency and optimized the methodology. We anticipate that our futuristic workflow package for contemporary microscopes is the initial endeavor toward fast data analytics and instant material characterization, spanning a diverse spectrum of interests.