Objective Remote sensing lithology mapping is of great significance for both basic geological research and mineral exploration. Aiming at the problems of low efficiency and strong subjectivity of traditional lithology interpretation methods in complex bedrock areas, this study takes the Huangshan area of the Eastern Tianshan Mountains in Xinjiang as the research area, aiming to construct an automatic classification model integrating spectroscopic and spatial characteristics. Improve the lithology identification accuracy of ASTER data in bedrock exposure areas and provide technical support for mineral resource exploration.

Methods Propose a collaborative framework of watershed segmentation and regularized extreme learning machine: ①Extract spatial boundary features through the watershed algorithm and establish a spatial constraint rule base; ②Principal component analysis and L2 regularization are adopted to optimize the spectral feature space and simplify the hidden layer structure of ELM. ③Design the maximum voting mechanism to integrate spectral classification and spatial constraint results. And compare and verify the model performance with four traditional algorithms such as Support Vector Machine (SVM), maximum likelihood method, and Markov distance method.

Results The experiments show that: ①The overall accuracy of the fusion model reaches 92.13% (Kappa=0.91), which is significantly improved compared with traditional classification methods such as SVM; ② Spatial characteristics improve the discrimination accuracy of similar rock types such as granite; ③After feature dimension reduction, the model parameters were significantly reduced and the classification time was greatly shortened.

Conclusions This model effectively breaks through the bottleneck of single spectral classification through multi−feature fusion, providing a new lithology identification scheme with high precision and high efficiency for the bedrock area. It can be adapted to data such as WorldView−3 and extended to similar bedrock exposure areas.