Task-oriented edge AI improves region fire detection with low data needs
TERRA: Task-Oriented Edge AI for Regional Fire Detection with Identity-Free Reporting
Machine LearningArtificial Intelligence
Summary
Monitoring large areas for fire often involves many sensors sending data, but the goal is to know if a region has fire, not details from each sensor. The authors propose TERRA, a system where sensors do local analysis and send simple summaries without needing to reveal their identity. This approach allows combining reports efficiently to detect fires at a regional level. Their tests show accurate fire detection using only small amounts of data sent from each sensor, and the system is robust even if some messages are missing or incorrect.
What this means in practice
- •For fire safety engineers: Enable building fire alarm systems to identify fire presence in regions with minimal sensor communication and no device identity tracking.
- •For industrial iot developers: Develop efficient edge AI modules for distributed sensor networks that aggregate regional event data while saving bandwidth and preserving privacy.
Tested on one dataset.
Authors
Mohammad Kazemi, Julie A. McCann
Abstract
Distributed fire-monitoring systems may involve many sensing devices reporting over a shared uplink, even though the operational objective is often to identify regions exhibiting evidence of fire rather than to recover device-specific measurements. We introduce TERRA (Task-Oriented Edge Representation and Regional Aggregation), an edge-AI framework that combines task-oriented local inference, identity-free region-tagged reporting, and permutation-invariant regional evidence aggregation. TERRA supports two reporting modes: TERRA-Score, in which each device communicates a quantized local fire probability, and TERRA-Feature, in which it communicates a compact learned representation for receiver-side aggregation using a permutation-invariant DeepSets model. Using all 15 numerical sensing modalities of the distributed indoor EN54 fire-alarm dataset as a proof-of-concept testbed, we evaluate regional inference for an episode-level fire-scenario target under repetition-disjoint three-fold cross-validation. We study representation size, task-payload quantization, report availability, and robustness to missed and false recovered messages. With nine reports and perfect message recovery, TERRA-Feature achieves mean AUPRC 0.990 using a 24-bit task payload, while TERRA-Score achieves 0.987 using only 6 bits; both attain an MDR of 1.81% at FAR $\leq$1%. TERRA-Feature achieves slightly higher AUPRC under reliable recovery, whereas the more compact TERRA-Score is less sensitive to increasing false-message contamination.