Abstract: To address the problems of false detections and missed detections caused by complex coal-flow background textures, large differences in typical foreign-object shapes, and significant scale variations in belt conveyor transportation scenarios, a foreign-object detection algorithm based on background awareness and multi-scale context is proposed. Taking YOLOv11n as the baseline model, a Background-Aware Hybrid Feature Extraction Module (BAHFEM) is introduced into the high-level feature extraction stage of the backbone network to enhance the differential response between foreign-object regions and the coal-flow background by integrating local relation modeling with convolutional representation, thereby suppressing complex texture interference. In the multi-scale feature fusion stage of the detection head, a Context-Aware Multi-Scale Feature Fusion Module (CAMFFM) is incorporated to jointly model contextual semantic information, high-frequency detail information, and low-frequency structural information, so as to strengthen information interaction among features at different scales and improve robust recognition performance for typical foreign objects such as large blocks and anchor rods. Experiments conducted on the CUMT-BelT belt conveyor foreign-object dataset show that, compared with the YOLOv11n baseline model, the improved model increases precision from 81.7% to 83.6%, recall from 77.5% to 78.5%, mAP@0.5 from 84.4% to 86.2%, and mAP@0.5:0.95 from 43.7% to 46.6%, while still maintaining a detection speed of 207 FPS with only a slight increase in the number of parameters and computational complexity. Comparative experiments further demonstrate that the proposed algorithm achieves a favorable balance among accuracy, real-time performance, and model complexity, and can satisfy the real-time foreign-object detection requirements of belt conveyors. Keywords: Belt conveyor; Foreign object detec