Using Multimodal Large Language Models for False Alarm Reduction in Image-based Fire Detection

Using Multimodal Large Language Models for False Alarm Reduction in Image-based Fire Detection | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Using Multimodal Large Language Models for False Alarm Reduction in Image-based Fire Detection Qie Gao, Haihui Wang, Zhenhai Qin, Linhao Fan, Kang Li, Chong Wang, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8847038/v1 This work is licensed under a CC BY 4.0 License Status: Under Review Version 1 posted 9 You are reading this latest preprint version Abstract Existing vision-based methods suffer from high false alarm rates in urban flame detection. Applying Multimodal Large Language Models (MLLMs) for secondary filtering shows great potential in reducing false alarms, yet they have high inference latency and are prone to reasoning collapse on negative samples without explicit Chain-of-Thought (CoT) guidance. To overcome these challenges, this study proposed Flash-Cascade, the first sub-second MLLM-based firewall to leverage CoT to efficiently filter false alarms. We deconstructed the flame detection process into four logical stages (planning, observation, analysis, and judgment), which informed the design of three switchable reasoning modes (Detailed, Quick, and Rapid) to achieve inference acceleration via CoT compression. We fine-tuned Qwen2-VL-7B-Instruct on a multi-grained instruction dataset via Low-Rank Adaptation. This process internalizes explicit reasoning logic into implicit parameter representations, enabling the model to maintain robust reasoning capability even without explicit CoT guidance. On our newly constructed benchmark incorporating real-world hard negatives, Flash-Cascade achieves an accuracy of 97.79% and an F1-score of 0.9767 in Rapid mode, outperforming the baseline by 61.63 percentage points (pp) and 0.5152, respectively. Furthermore, it outperforms the state-of-the-art object detector DEIMv2 by 14.64 pp in accuracy. The method exhibits exceptional sample efficiency, converging with only 600 samples and 2 epochs, and improves inference speed by 810% over standard CoT. This study will open a door for robust and efficient flame detection in high-interference scenarios. Flame Detection Multimodal Large Language Models Chain-of-Thought Compression Few-shot Learning False Alarm Suppression Full Text Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Review Version 1 posted Reviews received at journal 13 Mar, 2026 Reviewers agreed at journal 08 Mar, 2026 Reviews received at journal 08 Mar, 2026 Reviewers agreed at journal 25 Feb, 2026 Reviewers agreed at journal 21 Feb, 2026 Reviewers invited by journal 19 Feb, 2026 Editor assigned by journal 14 Feb, 2026 Submission checks completed at journal 14 Feb, 2026 First submitted to journal 10 Feb, 2026 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8847038","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":595015004,"identity":"7fce2274-d871-4ba5-9f21-3935a1f6e97a","order_by":0,"name":"Qie Gao","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Qie","middleName":"","lastName":"Gao","suffix":""},{"id":595015005,"identity":"0dba38f8-df57-422f-b317-a8263449633b","order_by":1,"name":"Haihui Wang","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Haihui","middleName":"","lastName":"Wang","suffix":""},{"id":595015006,"identity":"92f4b1f5-b581-4387-bb82-8c375c736e31","order_by":2,"name":"Zhenhai Qin","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Zhenhai","middleName":"","lastName":"Qin","suffix":""},{"id":595015008,"identity":"982e3869-8d5f-4a0d-a1ff-1ab8f4dfb648","order_by":3,"name":"Linhao Fan","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Linhao","middleName":"","lastName":"Fan","suffix":""},{"id":595015009,"identity":"4daa31eb-554d-4653-9baa-e83ed84a442f","order_by":4,"name":"Kang Li","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Kang","middleName":"","lastName":"Li","suffix":""},{"id":595015010,"identity":"a2e6deae-25d9-45dd-9af5-ef0ed88c38fb","order_by":5,"name":"Chong Wang","email":"","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":false,"prefix":"","firstName":"Chong","middleName":"","lastName":"Wang","suffix":""},{"id":595015011,"identity":"6197a295-59c5-427b-88ef-ba460c649257","order_by":6,"name":"Qixing Zhang","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAzklEQVRIiWNgGAWjYDACZgaGAwwMNlAeG/Fa0qCqidICAYdJ0MJ3nMfwcMGv84nz5zc/YPhQdpiBf3YDfi2Sh9kSDs/su5244RibAeOMc4cZJO4cwK/F4DDzgcO8PUAtbDwMzLxthxkMJBIIaWFsAGo5lzi/DajlL3FagLbw/DiQ2HAMqIWRGC1gv/A2JBtvOJZmcLDnXDqPxA0CWvjOnzH+zPPHTnZ+8+GHD36UWcvxzyCgBRSPDIxtCDYPAfVQZQx/CKsbBaNgFIyCEQwA/5BF/Z6n4mwAAAAASUVORK5CYII=","orcid":"","institution":"University of Science and Technology of China","correspondingAuthor":true,"prefix":"","firstName":"Qixing","middleName":"","lastName":"Zhang","suffix":""}],"badges":[],"createdAt":"2026-02-11 04:38:27","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8847038/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8847038/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":103505958,"identity":"c78bc149-2bcb-4a96-baec-dd404082a3c6","added_by":"auto","created_at":"2026-02-26 13:33:39","extension":"pdf","order_by":1,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":5454328,"visible":true,"origin":"","legend":"","description":"","filename":"UsingMultimodalLargeLanguageModelsforfalsealarmreductioninimagebasedfiredetection.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8847038/v1_covered_cb98e163-7114-4d90-bc35-b17f75e4fd2b.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Using Multimodal Large Language Models for False Alarm Reduction in Image-based Fire Detection","fulltext":[],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":false,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":true,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":true,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"fire-technology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fire","sideBox":"Learn more about [Fire Technology](http://link.springer.com/journal/10694)","snPcode":"10694","submissionUrl":"https://submission.springernature.com/new-submission/10694/3","title":"Fire Technology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false},"keywords":"Flame Detection, Multimodal Large Language Models, Chain-of-Thought Compression, Few-shot Learning, False Alarm Suppression","lastPublishedDoi":"10.21203/rs.3.rs-8847038/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8847038/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003eExisting vision-based methods suffer from high false alarm rates in urban flame detection. Applying Multimodal Large Language Models (MLLMs) for secondary filtering shows great potential in reducing false alarms, yet they have high inference latency and are prone to reasoning collapse on negative samples without explicit Chain-of-Thought (CoT) guidance. To overcome these challenges, this study proposed Flash-Cascade, the first sub-second MLLM-based firewall to leverage CoT to efficiently filter false alarms. We deconstructed the flame detection process into four logical stages (planning, observation, analysis, and judgment), which informed the design of three switchable reasoning modes (Detailed, Quick, and Rapid) to achieve inference acceleration via CoT compression. We fine-tuned Qwen2-VL-7B-Instruct on a multi-grained instruction dataset via Low-Rank Adaptation. This process internalizes explicit reasoning logic into implicit parameter representations, enabling the model to maintain robust reasoning capability even without explicit CoT guidance. On our newly constructed benchmark incorporating real-world hard negatives, Flash-Cascade achieves an accuracy of 97.79% and an F1-score of 0.9767 in Rapid mode, outperforming the baseline by 61.63 percentage points (pp) and 0.5152, respectively. Furthermore, it outperforms the state-of-the-art object detector DEIMv2 by 14.64 pp in accuracy. The method exhibits exceptional sample efficiency, converging with only 600 samples and 2 epochs, and improves inference speed by 810% over standard CoT. This study will open a door for robust and efficient flame detection in high-interference scenarios.\u003c/p\u003e","manuscriptTitle":"Using Multimodal Large Language Models for False Alarm Reduction in Image-based Fire Detection","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-02-24 06:29:10","doi":"10.21203/rs.3.rs-8847038/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"editorInvitedReview","content":"","date":"2026-03-13T07:40:59+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"335810156941754347395574862684039760258","date":"2026-03-08T10:37:39+00:00","index":"hide","fulltext":""},{"type":"editorInvitedReview","content":"","date":"2026-03-08T09:05:34+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"138755386239964749250777624833059093477","date":"2026-02-25T12:25:10+00:00","index":"hide","fulltext":""},{"type":"reviewerAgreed","content":"156673764810658405534457433308208089797","date":"2026-02-22T01:58:35+00:00","index":"hide","fulltext":""},{"type":"reviewersInvited","content":"","date":"2026-02-20T01:30:51+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2026-02-14T23:12:35+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2026-02-14T07:14:12+00:00","index":"","fulltext":""},{"type":"submitted","content":"Fire Technology","date":"2026-02-11T04:31:26+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"fire-technology","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"fire","sideBox":"Learn more about [Fire Technology](http://link.springer.com/journal/10694)","snPcode":"10694","submissionUrl":"https://submission.springernature.com/new-submission/10694/3","title":"Fire Technology","twitterHandle":"","acdcEnabled":true,"dfaEnabled":true,"editorialSystem":"stoa","reportingPortfolio":"Springer Hybrid","inReviewEnabled":true,"inReviewRevisionsEnabled":false}}],"origin":"","ownerIdentity":"31a4cd3e-faa2-4c68-83fc-b54f6182a9e2","owner":[],"postedDate":"February 24th, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"under-review","subjectAreas":[],"tags":[],"updatedAt":"2026-02-24T06:29:10+00:00","versionOfRecord":[],"versionCreatedAt":"2026-02-24 06:29:10","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8847038","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8847038","identity":"rs-8847038","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}