Abstract: Aiming at the problem that it is difficult to obtain the spatial information of the top equipment of the fully mechanized mining face with large mining height and the interference collision between the shearer and the hydraulic support, an intelligent detection model for anti-collision between the shearer and the hydraulic support based on YOLOv8 n algorithm is proposed. The image data set of shearer drum-hydraulic support panel is constructed. The image is preprocessed by semi-automatic labeling method and limited contrast adaptive histogram equalization technology. The target detection model is used to segment the image semantic of shearer drum and hydraulic support panel. The model optimization scheme integrates three core improvements: the Selective Kernel Attention mechanism is introduced to enhance the multi-scale feature representation ability through multi-branch convolution kernel adaptive weighted fusion; The C2f_BiFormer backbone module is designed, and the long-range dependency modeling is optimized by combining bidirectional dynamic attention and cross-stage feature fusion. The Lowlevel Feature Alignment feature alignment mechanism is constructed, and the cross-level feature space consistency correction is realized by using deformable convolution and gating unit. The 586 original images collected from the coal mine working face were expanded to 6 788 samples by rotation, brightness transformation, noise injection, etc., and the training set, verification set and test set were divided according to 8∶1∶1. Experiments show that the three optimization modules perform best in similar modules; the ablation experiment verified the contribution of each module: SK Attention increased mAP by 1.1%, C2f_BiFormer increased by 0.9%, LFA increased by 0.8%, and the average accuracy of the BSL-YOLOv8n model reached 93%, which was 2.0% higher than the benchmark YOLOv8n. F1 value is 98.9, the recall rate and the precision rate are both 98.9%, and the reasoning speed is 303 FPS, which meets the real-time detection requirements. The heat map analysis shows that the BSL-YOLOv8 n model can effectively focus the target area and suppress the background interference. Compared with the mainstream model, mAP exceeds YOLOv5n (89.6%), YOLOv6n (86.4%) and Faster R-CNN (84.25%), and FPS is significantly higher than Faster R-CNN (266.96). Deploy an interactive platform based on PyQt5, integrate dual-view comparison and collision warning functions, and support offline multi-source data detection. The model significantly improves the accuracy and robustness of collision detection of underground equipment, and provides reliable technical support for intelligent mining of coal mines.