In this dataset, it is a patent family that claims both a distributed fiber-optic sensing method — such as distributed acoustic sensing, distributed vibration sensing, or phase-sensitive OTDR — and a machine learning or pattern-recognition component used for tasks like event classification or signal denoising. The search combined core sensing IPC codes (G01D5/353, G01H9) with AI computing codes (G06N3), so pure optical-hardware patents without any learning claim are excluded, as are software-only AI patents with no fiber-sensing hardware. This is why only 40 of the 161 tracked families fall under G06N even though all of them are fiber-sensing related.

Recent-year momentum data shows Eagle Technology as the only assignee with active filings in the most recent tracked year, while several other named assignees — including NEC Laboratories America, Dow Global Technologies, Halliburton, and King Abdullah University of Science and Technology — show zero filings in that same year. This does not mean those organisations have exited the space; publication lag of roughly 18 months means recent activity is undercounted. Citation data separately points to earlier influential filers behind the top-cited GAN-based and autocalibration patents.

Filing rose from zero in 2017 to a peak of 38 in 2024, but the midpoint year of 2022 already reached 29 filings, close to the eventual peak. That pattern suggests the growth curve flattened well before 2024 rather than accelerating toward it, and it is consistent with an area where the core claim space around DAS hardware and basic classification has become more occupied over time.

The clearest gap is the intersection of AI-specific claim language with the sensing hardware: only 40 of 161 families sit in G06N despite 153 sitting in G01H, meaning most inventions still treat the learning model as incidental. Sub-areas worth checking specifically include multi-sensor fusion with DAS, edge-deployed inference for phi-OTDR systems, transfer learning across different fiber types, few-shot event classification, and explainable AI for DAS alarm systems — all technically plausible extensions with comparatively thin claim density in this corpus.

US12560475B2, assigned to Eagle Technology and published 2026-02-24, claims a DAS system that trains multiple machine learning networks on phase-sensitive OTDR data using different optimizers, then selects among them using a game-theoretic model before generating an acoustic event report. The novelty sits specifically in the selection mechanism — using game theory to pick among competing trained models — rather than in the underlying DAS hardware or in machine learning classification generally. Systems that use a single fixed model, or that select among models by simple accuracy comparison rather than a game-theoretic criterion, are the more obvious design-around routes, though any freedom-to-operate conclusion should be checked against the full claim set, not the abstract alone.