Microscopic Recognition of Tight Sandstone Based on Deep Learning

HIGHLIGHTS (1) The sample collection procedure was designed to replicate the manual identification process by capturing full extinction cycles and interference color sequences under both plane-polarized and cross-polarized light, thereby constructing an intelligent thin-section identification database with multi-parameter information. (2) In view of the characteristics of fine grains and strong compaction in tight sandstone, the semantic segmentation algorithm was boosted by integrating the segment anything model (SAM), achieving 95% accuracy in mineral boundary segmentation and providing a basis for precise mineral identification and grain-structure analysis. (3) A “primary identification + secondary verification” method was developed for easily confused minerals, achieving 91% recognition accuracy for major minerals (quartz, plagioclase, potassium feldspar, etc.) in tight sandstone thin sections.