<p>In mineral resource exploration, improved mapping is critical for enhancing the accuracy of mineralization identification and displaying ore body distributions. However, hyperspectral data are limited by medium resolution and mixed pixel effects, making it difficult to accurately identify lithology and boundaries, whereas traditional methods rely on geologists’ subjective interpretation of image data, resulting in lower mapping accuracy and consistency. As a result, this research offers an improved lithology mapping approach based on high-resolution remote sensing data and machine learning models, which is then applied to optimize the identification and mapping of granite pegmatites in the Altay area. The study employs Jilin-1 satellite imagery with a resolution of 0.5 meters and visible spectral bands to build a random forest model with 97% accuracy. The findings indicate that the granite pegmatites in the research region have a clear northwest-striking strip distribution, which is compatible with the geological background of the area. Comparative tests reveal that this technique outperforms standard methods and hyperspectral data for lithology identification and boundary characterization, demonstrating the efficacy and stability of (high-resolution) imagery integrated with machine learning. The study’s findings provide useful data for the exploration of rare metal deposits in the Altay region, as well as direct evidence for the study of tectonic evolution, magmatism, and mineralization processes in the region, all of which are important for mineral resource exploration.</p>