AuxTransUNet: Enhancing Remote Sensing Image Segmentation of Open-Pit Mining Areas in Qinghai–Tibet Plateau

Accurate and efficient extraction of mining areas from remote sensing imagery is essential for resource investigation, environmental assessment, and ecological management. This task is particularly crucial for the intelligent analysis of large-scale mining landscapes such as those found on the Qinghai–Tibet Plateau. However, existing approaches face notable limitations. Many deep learning models, especially those based on convolutional neural networks (CNNs), struggle to capture the complex and heterogeneous morphological features of mining areas in diverse geographic settings. To address these challenges, we propose AuxTransUNet, a hybrid deep learning framework that integrates CNNs with Transformers to enhance both local detail extraction and global contextual understanding. The architecture incorporates a frequency-aware and boundary-guided feature fusion strategy, which improves segmentation accuracy by suppressing misclassification noise and refining boundary delineation. In addition, an auxiliary classification branch provides patch-level supervision, further strengthening semantic consistency. Extensive experiments conducted on a mining segmentation dataset covering the Qinghai–Tibet Plateau demonstrate that AuxTransUNet achieves superior performance compared to strong baseline models in terms of both segmentation accuracy and computational efficiency. The results highlight its potential as a robust and scalable solution for large-area mining monitoring using remote sensing data.