<p>UAV Intelligent Inspection Systems (UAV-IIS) are vital for improving safety and reducing economic losses in mining operations through proactive hazard detection. In UAV imagery, targets are often extremely small, heavily occluded by rugged terrain, and densely distributed, such as sheep, cattle, and personnel, which leads to frequent false positives and false negatives when using conventional small object detection models. Therefore, this paper proposes <b>MIT-YOLO</b>, an intrusion target detection model for mining areas. Built upon YOLOv8, it integrates SMFANet, FcaNet, and a Coordinate Attention (CA) module to mitigate feature ambiguity caused by extremely small targets, fully leverage semantic and contextual information, and enhance spatial perception, thereby improving detection performance. For performance validation, comprehensive experiments were conducted using both our proprietary Mining-area Intrusion Target (MIT) dataset and the widely adopted VisDrone2019 benchmark. Quantitative results confirm that MIT-YOLO outperforms the original YOLOv8 model, exhibiting a 2.1% increase in mAP@0.5 and a 4.7% gain in mAP@0.5:0.95 specifically on the MIT test set. Cross-dataset validation on VisDrone2019 further confirms MIT-YOLO’s robustness, with mAP@0.5 and mAP@0.5:0.95 increasing by 4.5% and 3% respectively. This consistent enhancement across distinct datasets validates the model’s capability to accurately detect small-scale intrusion targets in complex mining environments, offering a reliable technical framework for autonomous inspection.</p>