Abstract: To address the tendency of existing object detection methods to produce false and missed detections due to complex coal-flow background textures and substantial variations in the morphology and scale of typical foreign objects in belt conveyor transport scenarios, a foreign object detection method based on an improved YOLOv11n model was proposed. A Background-Aware Hybrid Feature Extraction Module (BAHFEM) was introduced into the high-level feature extraction stage of the YOLOv11n backbone. By integrating local relational modeling with convolutional representation capabilities, this module enhanced differential responses between foreign object regions and the coal-flow background and suppressed interference from complex textures. A Context-Aware Multi-Scale Feature Fusion Module (CAMFFM) was introduced into the multi-scale feature fusion stage. By jointly modeling contextual semantics, high-frequency details, and low-frequency structural information, this module enhanced information exchange between features at different scales and improved the robustness of recognizing typical foreign objects, such as coal gangue and rock bolts. Experiments were conducted on the CUMT-BelT belt conveyor foreign object dataset. Compared with YOLOv11n, the improved model achieved increases of 1.9%, 1.0%, 1.8%, and 2.9% in precision, recall, mAP@0.5, and mAP@0.5:0.95, respectively. With only slight increases in parameter count and computational cost, it maintained a frame rate of 207 frames/s, achieving a favorable balance between detection accuracy and speed. The improved model accurately detected targets in scenarios with complex textures, partial occlusion, and different target scales, with generally high detection confidence.