RESUMO
Hyperspectral stimulated Raman scattering (SRS) microscopy is a label-free technique for biomedical and mineralogical imaging which can suffer from low signal-to-noise ratios. Here we demonstrate the use of an unsupervised deep learning neural network for rapid and automatic denoising of SRS images: UHRED (Unsupervised Hyperspectral Resolution Enhancement and Denoising). UHRED is capable of "one-shot" learning; only one hyperspectral image is needed, with no requirements for training on previously labelled datasets or images. Furthermore, by applying a k-means clustering algorithm to the processed data, we demonstrate automatic, unsupervised image segmentation, yielding, without prior knowledge of the sample, intuitive chemical species maps, as shown here for a lithium ore sample.
RESUMO
Hyperspectral stimulated Raman scattering (SRS) microscopy is a powerful label-free, chemical-specific technique for biomedical and mineralogical imaging. Usually, broad and rapid spectral scanning across Raman bands is required for species identification. In many implementations, however, the Raman spectral scan speed is limited by the need to tune source laser wavelengths. Alternatively, a broadband supercontinuum source can be considered. In SRS microscopy, however, source noise is critically important, precluding many spectral broadening schemes. Here we show that a supercontinuum light source based on all normal dispersion (ANDi) fibres provides a stable broadband output with very low incremental source noise. We characterized the noise power spectral density of the ANDi fibre output and demonstrated its use in hyperspectral SRS microscopy applications. This confirms the viability and ease of implementation of ANDi fibre sources for broadband SRS imaging.