Abstract: Large coal block accumulation is one of the main causes of blockage at the head transfer point of the scraper conveyor in a fully mechanized mining face. Timely and accurate breaking of large coal blocks is crucial for ensuring smooth coal flow in the fully mechanized mining face. However, short-term occlusion and posture changes of large coal blocks lead to low detection accuracy, which further prevents the crushing robot from accurately breaking them. To address this problem, a tracking and detection model for large coal blocks on a scraper conveyor named DAMP-YOLO11n-BT based on YOLO11n was proposed. The DCSNet module was used to replace the backbone network of the original YOLO11n model, which reduced the floating-point operations of the model while maintaining detection accuracy. The AG-SPPF module was adopted to enhance the model’s attention to the global background information of the coal flow region of the scraper conveyor and the local key information of coal blocks, and to improve its anti-interference capability under uneven illumination and other environmental conditions. Powerful-IoU (PIoU) was introduced to optimize bounding box regression through adaptive penalty and gradient adjustment, strengthen the focus on medium-quality anchor boxes, and enhance the detection capability for large coal blocks in dense coal block scenes. By integrating the DAMP-YOLO11n model with the ByteTrack algorithm, the DAMP-YOLO11n-BT model was proposed to realize the tracking and detection of large coal blocks. Experiments were conducted using a large coal block detection dataset of scraper conveyors collected on site. The results showed that: ① the accuracy, mAP@0.5∶0.95, and recall of the proposed DAMP-YOLO11n model were 86.3%, 77.6%, and 85.5%, respectively, which were improved by 2.4%, 2.4%, and 3.2%, respectively, compared with the original YOLO11n model. The number of parameters, floating-point operations, and model size were 1.95×10⁶, 4.8×10⁹, and 4.09 MiB, which were reduced by 24.4%, 23.8%, and 23.6%, respectively, compared with the original YOLO11n model. The detection speed reached 351 frames/s, which met the real-time detection requirement. ② The multiple object tracking accuracy, multiple object tracking precision, and ID F1 score of DAMP-YOLO11n-BT for large coal block tracking and detection were 76.6%, 74.5%, and 75.2%, respectively, all of which were better than those of YOLO11n-BT. The proposed method solves the problems of missed detection and ID switching of occluded large coal blocks and meets the tracking requirements for precise operation of the crushing robot.