Abstract: To address the problems of missed and false detections in vehicle obstacle detection under the complex conditions of severe occlusion, dust interference, and image blurring in open-pit mine operational areas, this study proposed a YOLO-DIS model based on YOLOv11n for vehicle obstacle detection in such complex open-pit mine environments. To enhance feature extraction capability in complex situations, the model introduced an Iterative Attention Feature Fusion (iAFF) mechanism to improve the C3k2 module, strengthening feature extraction through two-stage iterative attention fusion. To effectively compensate for the inaccurate recovery of edge features in occluded targets caused by fixed sampling rules, lightweight dynamic upsampling was adopted to replace the original nearest-neighbor interpolation method. This approach dynamically adjusted sampling positions based on target shape and occlusion by learning sampling point offsets. Furthermore, to solve the problem of imbalanced sample distribution, the SlideLoss function was employed to assign differentiated weights to samples of varying difficulty levels. Experimental results demonstrated that: compared to YOLOv11n, the YOLO-DIS model achieved improvements of 4.4%, 7.3%, and 4.0% in precision, recall, and mAP@0.5, respectively, with only a marginal increase in the number of parameters; compared to mainstream object detection models, the YOLO-DIS model achieved the highest mAP@0.5; the YOLO-DIS model maintained good detection performance on both a custom dataset and the KITTI dataset, indicating strong generalization capability. The YOLO-DIS model provided higher confidence levels for detection bounding boxes in scenarios involving severe occlusion, dust interference, image blurring, small target detection, and background interference, effectively reducing missed detections.