Deep Learning Fire Detection Patents: Who Leads, Where the Gaps Are 2026
- Every record in scope carries a G06V image-recognition class, yet only 27.3% also carry a G08B alarm-system class — most filings claim the vision model, not the alarm pathway around it.
- Filings peaked in 2025 at 6, up from zero in 2017, with the 2026 count still partial because publication lags filing by roughly 18 months.
- China-origin filings dominate the receiving offices 8 to 2 over WIPO PCT filings, with a single EPO filing — INNOVIRE AG's WO2025219569A1 — carrying by far the most citations in the set.
What this landscape covers
This landscape tracks patent families combining deep learning techniques with fire and smoke detection — models trained to recognise flame or smoke signatures from image or video input, including work on small flame detection, class imbalance in training data, edge deployment of inference models, and reduction of false alarm rates. The scope is narrow by design: it captures records that explicitly combine a fire/smoke detection claim with a named technical challenge in deploying a learned model, rather than every patent that merely mentions fire alongside AI.
Eleven records meet that combined test across the 2015-2026 window, with filing activity concentrated almost entirely in the last three years. That is a small, young corpus rather than a mature field, and the numbers below should be read that way — as an early signal of where claim space is forming, not a settled map of dominant players.
Let an AI agent run this analysis on your own technology
Pick a task. Every answer cites the patents behind it.
Filing trend and technology composition
Two views of the same 11-record corpus: how filing activity has moved year over year, and which IPC subclasses those filings sit in.
Filing trend, 2017-2026
Filings run from zero in 2017 to a peak of 6 in 2025; 2026 shows 2 so far but the year is not complete and publication lag means the true 2025-2026 count will run higher once later filings publish.
IPC subclass composition
G06V (image/video recognition) appears in 100.0% of the 11 records — effectively a precondition for inclusion in this search. G06N (AI computing models) follows at 54.5%, and G08B (signalling and alarm systems) at 27.3%, with G06F, G06T and H04N each appearing once. Because records can carry multiple classes, these shares sum past 100% and should be read against the 11-record total, not against each other.
Shares are the percentage of the 11 records in scope. A patent can carry several IPC classes, so the shares add up to more than 100%.
Go deeper on Deep Learning Models for Fire Detection with Eureka
This page is one run against one query. Ask Eureka your own question about deep learning models for fire detection and every answer comes back with the patent numbers behind it.
Try EurekaMost-cited records in this corpus
WO2025219569A1 — System and method for industrial risk assessment via computer vision
A device, system and method using computer vision for fire prevention, detection and broader risk assessment, combining infrared and visible-light camera data to detect or pre-empt a fire and to flag deviations from an ideal operational state in indoor industrial environments.Filed by INNOVIRE AG, published 2025-10-23; carries 15 citations, the highest in this corpus by a wide margin.
| # | Publication no. | Patent title | Citations |
|---|---|---|---|
| 1 | WO2025219569A1 | System and method for industrial risk assessment via computer vision | 15 |
| 2 | CN119131348A | 模型训练方法及火灾检测方法 | 2 |
| 3 | CN118587650A | 烟火检测方法及系统 | 2 |
| 4 | CN121600652A | 昼夜双模式烟火检测方法及装置 | 1 |
| 5 | CN119723442A | 一种基于轻量级多模态大模型的多阶段野外烟火检测方法 | 1 |
Ranked by citation count within this searched corpus; older records accumulate citations by virtue of age, so treat this as a signal of influence rather than of current technical importance.
Patent titles are shown in the language they were filed in, not translated, so that each record stays verifiable against the original filing — a translated title will not match in Eureka or in any national register. Publication numbers are shown where the record carries one (5 of 5 rows); clicking a row searches Eureka by that number.
Put your own technology through the same analysis
Eureka on the web
When you want the answer in the next five minutes.
The agent works the prompt against patents and technical literature, citing every source.
Run your analysis now →MCP server & REST API
When it has to run inside your own pipeline.
Patent search, landscape analysis and assignee resolution as MCP tools. Drop them into any agent framework, or call REST directly.
Browse MCP servers →What the numbers mean for a filing decision
Three read-outs from the composition and citation data that matter more than the headline record count.
The vision model is the claim, not the alarm system
Every record in scope claims image or video recognition; only 27.3% also claim a G08B signalling/alarm structure. That gap suggests most filers are protecting the detection model itself and leaving the downstream alarm integration, notification routing and false-positive suppression logic comparatively open.
One filing carries disproportionate influence
WO2025219569A1's 15 citations dwarf the next-highest record at 2. A single broadly-drafted computer-vision risk-assessment filing is doing most of the citation work in this corpus, which is worth checking directly rather than inferring from aggregate counts.
Filing activity is concentrated in China, with limited international filing
Eight of the tracked receiving-office filings originate in China against two PCT filings and a single EPO filing. That pattern points to a domestically-driven filing wave that has not yet been broadly internationalised — worth watching for whether Chinese filers pursue PCT or EPO coverage as the field matures.
Eureka can read the same corpus for gaps instead of for coverage: under-claimed branches adjacent to deep learning models for fire detection, with the prior art for and against each one.
Who is filing, and what is still open
The assignee ranking spans 15 companies counted in records, led by a filer with 3 records; by fifth place and tenth place activity has thinned to a single record each — a long tail rather than a concentrated field.
A modest lead, not a dominant one
The top-ranked assignee holds 3 of the 11 records in scope. That is a lead in a young, small corpus rather than evidence of a controlling patent position — there is room for new entrants to establish comparable claim density.
Activity thins out fast past the leader
Fifth place and tenth place in the ranking both sit at a single record. Most of the 15 ranked companies are single-filing entrants, which is typical of a field still forming its claim structure rather than one with entrenched incumbents.
Recent-year momentum has cooled for early movers
Several assignees active in 2025 show zero filings in the latest year and a -100% year-on-year change, while at least one filer shows continued activity into the latest year. Momentum is shifting rather than compounding for any single player so far.
| Assignee | Recent year | YoY |
|---|---|---|
| Yangtze Delta Region Institute of Tsinghua University, Zhejiang | 1 | — |
| INNOVIRE AG | 0 | -100% |
| Guizhou Power Grid Co., Ltd. | 0 | -100% |
| Southwest Jiaotong University Yantai Institute of New Generation Information Technology | 0 | — |
| Aidien (Shandong) Technology Co., Ltd. | 0 | — |
| Tsinghua University Hefei Institute for Public Safety Research | 0 | -100% |
| Shenzhen Haibo Engineering Technology Co., Ltd. | 0 | -100% |
| Shenzhen Xinghai Electromechanical Engineering Co., Ltd. | 0 | -100% |
Where to take this
The corpus is small enough that a single filing or a single competitor move can shift the picture quickly.
Watch the alarm-integration gap
With only 27.3% of records claiming G08B alarm structures against 100.0% claiming G06V recognition, the notification and alarm-routing layer around a detection model remains comparatively open.
Explore this gap in EurekaTrack WO2025219569A1's citation pull
A single filing carries 15 citations against a field where the next-highest sits at 2 — understanding what that filing actually blocks matters more than the aggregate count.
Analyze this filing in EurekaMonitor the China-to-PCT filing gap
Eight China-origin filings against two PCT and one EPO filing suggests international filing has not caught up with domestic activity — a signal worth revisiting as the 2025-2026 cohort publishes.
Set up monitoring in EurekaCommon questions about this landscape
This landscape identifies 11 published records that combine a fire or smoke detection claim with a named deep-learning deployment challenge, such as small flame detection, class imbalance, edge deployment or false alarm reduction, across a 2015-2026 window. That is a narrow, deliberately combined search rather than a count of every patent mentioning fire and AI together, so broader searches will return larger numbers. Filing activity is concentrated in the last three years, with a peak of 6 records in 2025, meaning this is a young and still-forming corpus rather than a mature one.
The assignee ranking covers 15 companies counted in records, and it is a long tail rather than a concentrated field: the leading assignee holds 3 records, while fifth and tenth place both sit at a single record. Most named filers appear only once, which is typical of an early-stage technology area where no single company has yet established a controlling patent position. Readers should treat any current leader as provisional given how thin the ranking gets past the top few positions.
WO2025219569A1, filed by INNOVIRE AG and published 2025-10-23, covers a device, system and method that use computer vision — combining infrared and visible-light camera data — to detect or prevent fires and to assess broader operational risk in indoor industrial environments by flagging deviations from an ideal operational state. It is the most-cited record in this corpus by a wide margin, with 15 citations against a next-highest of 2. Its scope extends beyond fire detection alone into general industrial risk assessment via computer vision, which is worth checking directly against any planned filing that touches camera-based fire monitoring in an industrial setting.
The clearest gap sits between detection and response: every one of the 11 records carries a G06V image-recognition class, but only 27.3% also carry a G08B signalling/alarm class, meaning the alarm-integration and false-alarm-suppression logic around a detection model is comparatively under-claimed. Co-assignee activity is also thin, with only 7 co-assignee pairs across the corpus and none stronger than a single shared filing, suggesting collaborative claim-building has not yet consolidated around any particular technical combination. Edge-deployment model compression and cross-domain generalization across camera hardware are similarly represented in only a handful of the 11 records.
Citation counts inside a searched corpus like this one tend to favour older filings simply because they have had more time to be cited, so a high count signals influence within this dataset rather than current technical importance. With only 11 records total, a single filing such as WO2025219569A1 can carry a citation count many times higher than the rest of the field just by virtue of its broader claim scope and earlier publication relative to the 2025-2026 filing wave. Treat citation rankings here as a starting point for review, not as a verdict on which patents matter most going forward.
Research Deep Learning Models for Fire Detection in depth with Eureka
Go past this page: query the whole deep learning models for fire detection corpus yourself, in your own scope.
Every answer comes back with patent numbers you can open.
Disclaimer. This page is generated from Patsnap Eureka data drawn from a limited snapshot of global patent and scientific-literature records, and is provided for general information and reference only.
Patent data carries inherent limitations: recent filings (typically the most recent 18–24 months) are under-counted due to standard publication lag; counts may be reported at either a patent-family or a patent-record basis and are not always directly comparable; classification, applicant-name, and citation data may contain errors, duplicates, or omissions; and the underlying search query defines and constrains the scope shown. As a result, the analysis may be incomplete or inaccurate and may not reflect the full technology landscape.
Nothing on this page constitutes an exhaustive prior-art, novelty, freedom-to-operate, or validity search, nor does it constitute legal, financial, investment, or professional advice, and it should not be relied upon as such. Any patent, commercial, or strategic decision should be verified independently and reviewed with qualified patent, legal, and domain professionals. Patsnap makes no warranties, express or implied, as to the accuracy, completeness, or fitness for any particular purpose of the information presented.
Machine translation. Assignee and organisation names originally recorded in Chinese, Japanese or Korean have been rendered into English by an AI translation step so that the tables stay readable. These renderings are best-effort and may not match a company’s registered English name; the original name is what the underlying patent record carries, and it is what any Eureka query launched from this page uses.