• A bio-inspired FPE-YOLO framework for small object detection in pipeline intrusion monitoring. • Progressive three-stage enhancement: attention → cross-scale fusion → feature enhancement. • FoveaAttention improves localization via multi‑scale center‑region features. • E‑DCFP embeds foveal attention in bidirectional pyramids for adaptive scales. • 99.14% mAP@0.5, 97.26% mAP@0.5:0.95 at 65.4 FPS; +4.91/+4.59 vs YOLOv8n. As oil and gas pipeline networks expand, third-party intrusions such as mechanical excavation, illegal construction, and unauthorized presence pose serious threats to pipeline integrity, affecting safety and causing unexpected outages. We introduce FPE‑YOLO, an enhanced YOLOv8-based system designed for real-time detection of intrusions, especially under challenging conditions involving small, sparse objects and significant background interference along rights‑of‑way. Inspired by foveal vision, our approach employs a progressive method: bio-inspired attention, cross-level fusion, and single-level enhancement, comprising three modules: FoveaAttention, E-DCFP, and PSEM. Fovea-guided adaptive fusion and small-object enhancement work together to detect and represent subtle intrusions accurately. On the FPE dataset, which contains 6,750 images with 75.8% small objects, FPE‑YOLO achieves a mAP@0.5 of 99.14% and a mAP@0.5:0.95 of 97.26%, outperforming YOLOv8n by +4.91 and +4.59 percentage points, respectively. It operates in real time at 65.4 FPS, with only a 19.5% increase in computational complexity. These results demonstrate the framework’s effectiveness and suitability for continuous pipeline monitoring, enabling earlier detection and reducing operational risks in energy infrastructure.
Dong et al. (Sun,) studied this question.