Unsupervised Hyperspectral Microscopic Image Segmentation Using Deep Embedded Clustering Algorithm.

Area of Science:

• Optics and Photonics
• Artificial Intelligence
• Materials Science

Background:

• Hyperspectral microscopy generates complex data, posing challenges for signal analysis and information extraction.
• Current algorithms struggle with uncertainty in evaluating biosignals.
• Supervised deep learning requires extensive labeled datasets, limiting its applicability.

Purpose of the Study:

• To develop an effective unsupervised deep learning method for hyperspectral image denoising and resolution enhancement.
• To overcome limitations of traditional machine learning and supervised deep learning in hyperspectral imaging.
• To enable intuitive chemical species mapping for mineral samples and advanced biosignal analysis.

Main Methods:

• Utilized nonlinear deep learning networks to transform hyperspectral data into a latent feature space (Z).
• Employed unsupervised learning, specifically k-means clustering, for intuitive chemical species mapping.
• Applied Kullback-Leibler divergence to monitor objective function convergence during training.

Main Results:

• Demonstrated that a single hyperspectral image is sufficient to train the unsupervised denoising and enhancement model.
• Successfully generated an intuitive chemical species map for a lithium ore sample.
• Showcased the potential for enhanced analysis of biosignals and mineralogical data.

Conclusions:

• Unsupervised deep learning offers a powerful approach for hyperspectral image processing, overcoming data limitations.
• The developed method enhances resolution and reduces noise in hyperspectral microscopy.
• This technique holds promise for diverse applications in biology, mineralogy, and AI-driven signal analysis.