Deep learning spectral infrared digital holography for phase analysis of shale characterization

Abstract

Micrometer-scale pores, crucial for hydrocarbon transport in unconventional reservoirs, mainly consist of cracks and intergranular pores. Optical microscopy lacks refractive index measurement capability, whereas traditional methods cannot adequately represent depth-dependent particle size features in phase images. Here, an approach integrates deep learning with a Transformer model for shortwave infrared spectral digital holography, enabling spectral, amplitude, and phase imaging. The infrared band is more sensitive to organics and water-bearing minerals, which contain more light elements, than advanced imaging techniques such as X-ray tomography. Shortwave infrared spectroscopy utilizes the characteristic absorption bands of hydroxyl groups and water molecules to quantitatively analyze the densified pore structure and mineral dissolution (plagioclase, orthoclase, microcline, and illite) through spectral phase changes. Furthermore, the pore distribution varies among shale types, with sandy shale exhibiting an amplitude difference approximately 14.3% higher and a phase difference about 91.2% greater than those of mud shale. Additionally, the band's low absorption and high penetration reveal structural characteristics in transparent and metallic minerals. Through quantitative analysis of structural distribution patterns, this method significantly improves mineral phase discrimination and elucidates formation dynamics, making it a valuable tool for the structural and compositional characterization of unconventional geological reservoirs.