To address the challenges of object detection in high-resolution industrial CAD images—such as large variations in equipment size, dense object distribution, and GPU memory constraints—this paper proposes BBOD-CAD (Block Based Object Detection in CAD Images), an enhanced YOLOv8 framework tailored for intelligent detection of characteristic equipment in engineering drawings. BBOD-CAD adopts a sliding window partitioning strategy to efficiently process ultra-large images, coupled with bounding box reconstruction and boundary-aware fusion techniques to maintain object integrity across image blocks. This approach reduces memory overhead while preserving detection accuracy. To alleviate class imbalance and improve recognition of underrepresented equipment types, the method incorporates transfer learning and adaptive data augmentation, significantly enhancing robustness in limited-data scenarios. Experiments conducted on a proprietary dataset of 208 annotated CAD drawings, encompassing 36 types of industrial equipment, show that the baseline YOLOv8 achieves only 0.274 mAP@0.5, while BBOD-CAD reaches 0.821 mAP@0.5, with 0.832 precision and 0.744 recall. These results demonstrate that BBOD-CAD offers an effective and scalable solution for automated annotation, quality inspection, and equipment recognition in intelligent manufacturing environments.