We present UPE-YOLOv7, a novel UAV-based framework for real-time tailings pollution monitoring, combining an enhanced YOLOv7 detector with multi-modal image enhancement. Targeting challenges like dust, low-light, and small lowsaliency targets in mining areas, our method introduces: (1) a multi-scale pyramid network for noise-resistant feature extraction, (2) a hybrid edge-texture enhancer to sharpen contamination boundaries, (3) a context-aware semantic module for background suppression, and (4) a self-adaptive denoiser for degraded images. Experiments on a custom UAV dataset show UPE-YOLOv7 achieves 82.3% mAP (vs YOLOv7’s 75.6%), with stable performance across target sizes and noise levels. This demonstrates its potential as a scalable solution for automated environmental monitoring in challenging mining scenarios.