Research map / Label-efficient and robust learning / Recognition and long-tail
AI-Based Energy Transportation Safety: Pipeline Radial Threat Estimation Using Intelligent Sensing System
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a radial threat estimation method for energy pipelines using distributed optical fiber (DAS) signals, fusing multi-view, multi-domain space-time-frequency features across adjacent defense zones and classifying threats with a transfer-learned EfficientNetV2.
Problem
Pipeline safety early warning with distributed acoustic sensing struggles to determine the radial distance of external threat events from the pipeline, leading to high false alarm rates.
Approach
Builds a space-time-frequency fused feature map by stitching time-frequency maps of a defense zone and its two adjacent zones, then classifies three radial threat areas with an EfficientNetV2 initialized from a pre-trained model.
Key results
- The proposed method reaches 97.82% average precision, 97.67% average recall and 97.69% average F1 on the three-class radial threat task, versus 88.02% precision for STFF without augmentation.
- False alarm rate drops to 0.99%, compared with 2.05% for EfficientNetV1 and 9.22% for MobileNetV2.
- With pre-training, about 40 epochs match the performance that takes 200 epochs without pre-training.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add space-time-frequency features from continuous defense zones instead of time-frequency features only (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset | F1_ave: 63.33 -> 87.48 | Helped |
| Add time-frequency features instead of using time-domain features only (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset | F1_ave: 53.59 -> 63.33 | Helped |
| Add space-time-frequency feature processing instead of using raw data (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset | F1_ave: 51.50 -> 87.48 | Helped |
| Add data augmentation to STFF (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset | F1_ave: 87.48 -> 97.69 | Helped |
| Add data augmentation to STFF (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset | FAR: 5.81 -> 0.99 | Helped |
| Use a pre-trained model (ablation) | EfficientNetV2 pipeline threat classifier · Pipeline radial threat dataset training | Training epochs to reach reported effect: 200 -> 40 | Helped |
| Use the proposed model instead of EfficientNetV1 | Pipeline threat recognition model · Pipeline radial threat dataset | F1_ave: 95.04 -> 97.69 | Helped |
| Use the proposed model instead of EfficientNetV1 | Pipeline threat recognition model · Pipeline radial threat dataset | FAR: 2.05% -> 0.99% | Helped |
Limitations
- Dataset is small and collected at a single site with a self-developed DAS system, limiting generality.
- Misclassified samples fall mostly into the tracking area, which is close to the alarm area in distance.
- Only three radial threat classes and a limited set of baselines are evaluated.
Experimental setup
Datasets: self-collected DAS field recordings of mechanical excavation at 5m, 20m, 40m radial distances in a Zhejiang natural gas
Benchmarks: self-collected DAS pipeline excavation dataset
Metrics: precision recall F1-score false alarm rate parameters testing time
Baselines: Raw data time-domain features time-frequency domain features space-time-frequency domain features without augmentation MobileNetV2 EfficientNetV1 Swin Transformer-B
Builds on: EfficientNetV2 squeeze-and-excitation variational mode decomposition
Closest papers by meaning
- Defining Operational Conditions for Safety-Critical AI-Based Systems from Data (NeurIPS 2026 · similarity 0.78)
- Position: Require Frontier AI Labs To Release Small "Analog" Models (NeurIPS 2025 · similarity 0.78)
- SafetyAnalyst: Interpretable, Transparent, and Steerable Safety Moderation for AI Behavior (ICML 2025 · similarity 0.77)
- Quantitative Predictive Monitoring and Control for Safe Human-Machine Interaction (AAAI 2025 · similarity 0.77)
- Towards Visual Discrimination and Reasoning of Real-World Physical Dynamics: Physics-Grounded Anomaly Detection (CVPR 2025 · similarity 0.77)
- Data Disparity and Temporal Unavailability Aware Asynchronous Federated Learning for Predictive Maintenance on Transportation Fleets (AAAI 2024 · similarity 0.77)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.