Turning Rocks into Kaleidoscopes, Devdigvijay Singh, UG '24 (3959058)
Accurately assessing the shape, size, and modality of features in rock samples is a longstanding problem in geology. Recent advances in machine learning have made it possible to improve upon point counts with automated image classification; however, these techniques perform best with enhanced contrast to help differentiate classes. To leverage these methods for geological applications, we need a way to acquire high-resolution images of thin sections with a field of view large enough to resolve entire crystals, fossils, bedforms, etc. We present a novel multispectral light table equipped with 5-band (470-940 nm) spectral resolution and computer controlled broadband polarizers that can rotate in 0.1 degree increments in both a plane-polarized and cross-polarized transmitted light configuration. Paired with a 150 MP panchromatic camera that can acquire images at ∼3.76 μm per pixel spatial resolution over a 4 cm by 4 cm field of view, the new system is a high-throughput thin section imager. The additional spectral bands outside the visible range, combined with cross-polarized rotations, encode rock properties that heighten image contrast through wavelength-dependent birefringence and differential extinction. Our setup provides an efficient way to (1) build reproducible image archives of petrographic thin sections that complement field observations, (2) classify and segment those images, and (3) quantitatively compare lithofacies and fossil assemblages with complementary geochemical measurements.