A daily consumer goods safety grading detection live broadcast traceability early warning method based on microscopic intelligent identification
Patent Information
- Application Number
- CN202611125386.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-07-28
- Publication Date
- 2026-10-09
AI Technical Summary
本发明旨在解决现有消费品显微直播检测人工判读误差大、缺陷无法智能识别、安全风险无分级、溯源数据静态固化、无实时直播预警机制、检测标准不统一的技术空白；构建一套AI微观缺陷智能识别、定量检测、分级判定、动态溯源更新、直播实时预警的全自动智能检测方法，实现消费品安全检测标准化、量化化、智能化、可预警、可追溯，彻底区别于传统人工观测、硬件展示的常规直播检测模式
1）本发明采用AI机器视觉微观智能识别技术，替代传统人工肉眼主观判读，可自动识别多类隐蔽微观安全隐患，实现缺陷定量检测、数据量化输出，彻底解决人工漏检、误检、标准不统一的行业痛点，大幅提升消费品安全检测的精准度与规范性。2）本发明首创四级安全风险分级判定机制，根据量化缺陷数据自动评级并生成处置建议，实现检测结果分层化、标准化、可视化，让普通观众可直观读懂产品安全等级，大幅提升直播科普公信力。3）本发明采用动态可迭代溯源赋码逻辑，区别于传统静态水印绑定模式，溯源码可实时写入检测缺陷数据、风险等级、检测结论，实现一物一码、数据动态更新、全程可追溯、不可篡改，构建标准化质检存证体系。4）本发明具备智能画面适配与分级预警功能，高风险样品自动聚焦缺陷区域、高亮弹窗预警、语音同步播报，解决传统直播重点不突出、隐患不醒目、无风险提示的问题，实现检测直播智能化、人性化、专业化展示。5）本发明形成“智能采集-AI识别定量-分级判定-动态赋码-实时预警-归档溯源-迭代检测”全流程智能闭环，完全区别于实用新型的硬件结构方案，独立形成智能化方法创新，可单独申报，无雷同、无重复授权风险。
Smart Images
Figure FT_1 
Figure FT_2 
Figure FT_3
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of intelligent inspection of consumer products, microscopic machine vision recognition, and live-stream traceability risk warning. Specifically, it relates to an intelligent inspection and display method based on high-definition microscopic imaging, AI-based automatic identification of microscopic defects, safety level classification, dynamic traceability coding and binding, and live-stream real-time risk warning. It is applicable to scenarios such as intelligent screening of microscopic safety hazards, visual live-stream evidence storage, and graded risk disclosure for daily consumer products such as textiles, clothing, baby products, and daily chemical consumables. Background Technology
[0002] Current live-stream microscopic testing of consumer products relies entirely on manual visual observation and subjective judgment, resulting in significant human error and failing to achieve standardized and intelligent testing and assessment. Current live-stream testing models can only capture and display images, lacking the ability to automatically identify microscopic defects, classify hazard types, and categorize safety risks. Hidden safety hazards such as minute fiber damage, harmful chemical residue crystals, coating peeling particles, and microbial impurities are easily missed or misjudged by humans. Furthermore, traditional live-stream testing uses a static data binding model, only allowing for basic sample information entry. It cannot dynamically update traceability files based on defect results, generate corresponding risk level labels, or provide real-time warnings during the live stream. Existing technologies lack a fully automated closed-loop process encompassing "intelligent microscopic image recognition - quantitative defect analysis - safety level determination - dynamic traceability coding - real-time live-stream public warnings." This results in inconsistent testing standards, unquantified results, lack of tiered traceability data, and no risk warnings during live streams, leading to serious deficiencies in testing credibility and compliance. Furthermore, traditional multi-channel live streaming only achieves synchronized playback of images, without intelligently adapting imaging parameters and display priorities according to the inspection scenario. Key areas of microscopic defects cannot be focused and magnified, and ordinary viewers cannot accurately identify the location of safety hazards. This significantly reduces the popular science, authority, and practicality of professional inspection live streams, making it difficult to meet the needs of standardized quality inspection live streams and market supervision evidence preservation. Summary of the Invention
[0003] (1) Technical problems to be solved This invention aims to address the technological gaps in existing live-stream microscopic inspection of consumer products, such as large errors in manual interpretation, inability to intelligently identify defects, lack of tiered safety risks, static and fixed traceability data, absence of real-time live-stream early warning mechanisms, and inconsistent testing standards. It constructs a fully automated intelligent inspection method based on AI, enabling intelligent identification, quantitative detection, tiered judgment, dynamic traceability updates, and real-time live-stream early warnings. This method achieves standardized, quantifiable, intelligent, early warning-enabled, and traceable consumer product safety inspections, completely differentiating itself from the conventional live-stream inspection mode that relies on manual observation and hardware display.
[0004] (2) Technical solution A live-stream traceability and early warning method for safety grading testing of daily consumer goods based on microscopic intelligent recognition is characterized by the following steps: S1. Sample standardization and dynamic traceability code initialization: For the daily consumer goods to be tested, the sample category, production batch, sampling source, and compliance standard parameters are entered to establish a standardized sample testing file; a unique intelligent traceability code with dynamic data update function is generated, and the traceability code reserves a port for writing defect data, risk level, and test conclusions to achieve dynamic iterative binding of data during the testing process. S2. Multi-camera frame-level synchronous intelligent adaptation and acquisition: Based on the PTP clock synchronization protocol, microsecond-level synchronous acquisition of three channels of images—microscopic imaging, product panorama, and operation demonstration—is completed; the light source color temperature, exposure parameters, and sampling frame rate are automatically adapted according to the sample material type; the imaging parameter mode is adaptively switched for flexible fabrics, rigid daily chemical consumables, and porous baby products to obtain a standardized original microscopic image sequence without reflection, noise, or distortion. S3. AI-powered Intelligent Defect Recognition and Quantitative Analysis of Microscopic Images: Constructs a microscopic defect recognition model for consumer products, extracts pixel-level features from standardized microscopic images, and automatically identifies five types of microscopic safety hazards: fiber breakage, coating impurities, chemical residue crystallization, pore defects, and foreign matter contamination. It also quantitatively calculates defect area, number, and distribution density, generating quantitative raw test data. S4. Consumer Product Safety Level Classification: Presets safety testing and grading threshold standards for daily consumer products. Based on AI quantitative test data, samples are automatically classified into four risk levels: Grade A (compliant and high-quality), Grade B (minor defects), Grade C (safety hazard), and Grade D (serious non-compliance). Corresponding standardized test conclusions and risk management suggestions are generated. S5. Intelligent Layered Synthesis and Dynamic Traceability Watermark Overlay for Live Streaming: Adaptively switches between split-screen layout and focus mode in live streaming based on the test risk level. High-risk samples automatically trigger local magnification and highlighting of microscopic defect areas. Risk level labels, quantitative test data, and standardized conclusions are dynamically embedded into the traceability watermark layer, enabling real-time updates of traceability code information with test results, unlike static fixed watermarks. S6. Real-time Risk Warning and Dual-Terminal Synchronous Public Display: For C-level and D-level risk samples, a highlighted risk warning pop-up appears on the live stream screen, with simultaneous voice broadcast of the hazard level and risk warning; the synthesized standardized live stream video stream is pushed to the cloud live streaming platform with low latency through layered encoding, and simultaneously output to the local display screen, achieving real-time public display on both online and offline platforms. S7. Hierarchical Data Archiving and Iterative Updates for Traceability: AI-identified defect data, risk level, test conclusions, full live stream recordings, and defect focus screenshots are bound in real-time to dynamic intelligent inspection traceability codes and archived in the traceability database; users can scan the code to query complete sample test data, risk rating, and the entire live stream process, while also supporting online submission requests, automatically generating a new round of testing traceability tasks, forming an intelligent testing live stream traceability closed loop.Preferably, in step S3, the microscopic defect recognition model is a machine vision model specifically trained for daily consumer goods. Through extensive training and optimization using numerous microscopic samples of textiles, baby products, and daily chemical products, it can accurately distinguish between normal textures and abnormal defects, eliminating false or missed identifications. Preferably, in step S4, the four risk levels correspond to preset national and industry standard detection thresholds, achieving standardized and quantifiable detection judgments and avoiding errors from subjective human judgment. Preferably, in step S5, the dynamic watermark uses an encrypted embedded overlay method, ensuring that data is not lost or tampered with during video streaming, transcoding, and playback, guaranteeing the authenticity and uniqueness of the detection and traceability data. Preferably, in step S6, the risk warning intensity adaptively adjusts according to the risk level; a Level D severely non-compliant sample triggers a red highest-level warning and forces a highlighted defect display at the top.
[0005] (3) Beneficial effects 1) This invention employs AI machine vision microscopic intelligent recognition technology to replace traditional manual subjective judgment. It can automatically identify various hidden microscopic safety hazards, achieve quantitative defect detection and data output, and completely solve the industry pain points of missed detections, false detections, and inconsistent standards caused by manual inspections, significantly improving the accuracy and standardization of consumer product safety testing. 2) This invention pioneers a four-level safety risk grading and judgment mechanism. Based on quantified defect data, it automatically rates the product and generates handling suggestions, achieving hierarchical, standardized, and visualized test results. This allows ordinary viewers to intuitively understand the product's safety level, significantly enhancing the credibility of live-streamed science popularization. 3) This invention adopts a dynamic and iterative traceability coding logic, different from the traditional static watermark binding mode. The traceability code can be written in real time with the detected defect data, risk level, and test conclusions, achieving one code per item, dynamic data updates, full traceability, and tamper-proof functionality, constructing a standardized quality inspection and evidence storage system. 4) This invention features intelligent image adaptation and tiered early warning functions. High-risk samples automatically focus on defect areas, receive high-brightness pop-up warnings, and receive synchronized voice broadcasts, solving the problems of traditional live broadcasts lacking focus, visibility of potential hazards, and risk warnings. This achieves intelligent, user-friendly, and professional live broadcasting of inspections. 5) This invention forms a fully intelligent closed loop encompassing "intelligent acquisition - AI-based quantitative identification - tiered judgment - dynamic coding - real-time early warning - archiving and traceability - iterative detection." Completely different from the hardware structure solutions of utility models, this invention independently forms an innovative intelligent method, which can be applied for separately without the risk of plagiarism or duplicate authorization. Attached Figure Description
[0006] Figure 1 This is a block diagram of the overall hardware signal connection of the present invention; Figure 2 This is a three-dimensional assembly structure diagram of the microscopic imaging acquisition unit of the present invention; Figure 3 This is a schematic diagram of the hardware connection of the multi-camera synchronous acquisition module of the present invention; Figure 4 This is a schematic diagram illustrating the live stream split-screen synthesis and traceability code watermark overlay display of the present invention; Figure 5This is a schematic diagram of the spatial installation layout of the complete laboratory testing live broadcast workstation of this invention; wherein, 1-microscopic imaging acquisition unit, 2-multi-camera synchronous acquisition unit, 3-real-time image processing and synthesis terminal, 4-streaming media push and interaction unit, 5-traceability coding data linkage industrial control unit, 6-local synchronous display large screen, 7-cloud live broadcast platform server, 8-sample intelligent inspection code traceability database; 10-stereoscopic digital microscope host, 11-4K industrial imaging camera, 12-ring adjustable coaxial light source, 13-sample shock absorption and anti-static stage, 14-stage anti-slip pad, 15-light source color temperature adjustment controller, 16-microscope lifting and adjusting bracket; 20-PTP clock synchronization controller, 21-multi-channel HDMI high-definition acquisition card, 22-elimination 23-Auxiliary camera for consumer goods panoramic view; 24-HDMI input interface for microscope camera; 25-Synchronization signal trigger harness; 26-USB data output terminal for acquisition card; 30-Microscopic image display zone; 31-Panoramic appearance zone for consumer goods; 32-Auxiliary screen zone for operator demonstration; 33-Smart inspection code traceability QR code watermark layer; 34-Sample batch text label layer; 35-Institutional qualification mark layer; 40-Microscopic imaging inspection operator table; 41-1.2m fume hood; 42-PP anti-corrosion exhaust duct and activated carbon purification box; 43-Top-mounted fresh air filter purification box; 44-Ceiling synchronous display projection screen; 45-Image processing and traceability industrial control host; 46-Top shock-absorbing hanging bracket. Detailed Implementation
[0007] The technical solution of the present invention will now be described in a clear and complete manner with reference to the accompanying drawings.
[0008] This invention discloses a live-stream traceability and early warning method for safety classification and testing of daily consumer products based on microscopic intelligent recognition. Specifically, it includes seven core steps: standardized sample filing and dynamic traceability code initialization, multi-camera frame-level synchronous intelligent adaptation acquisition, AI-powered intelligent defect identification and quantitative analysis of microscopic images, safety level classification of consumer products, intelligent layered synthesis of live-stream images and dynamic traceability watermark overlay, real-time risk warning and dual-end synchronous publicity, and graded data archiving and iterative update of traceability.
[0009] First, the sample standardization and dynamic traceability code initialization steps are performed: staff enter parameters such as the category, production batch, sampling channel, and corresponding national testing standard threshold of the textiles and apparel, baby products, and daily chemical consumables to be tested. The system establishes an independent standardized sample testing file; a unique intelligent inspection traceability code with dynamic data writing function is generated, and a data storage port for defect type, defect quantification data, risk level, and test conclusion is reserved in advance to provide a unique identifier for subsequent dynamic binding and iteration of test data.
[0010] Subsequently, multi-camera frame-level synchronous intelligent adaptation acquisition is performed: the system uses the PTP clock synchronization protocol to perform microsecond-level timing calibration on the microscopic imaging acquisition path, the product panoramic acquisition path, and the operation table demonstration acquisition path to ensure synchronous exposure and sampling of the three images; at the same time, the imaging parameters are automatically adapted according to the material properties of the sample, the anti-reflective imaging mode is automatically activated for flexible textile fabrics, the imaging clarity is improved for rigid daily chemical consumables, and the noise reduction parameters are optimized for porous maternal and infant consumables, outputting a standardized, distortion-free microscopic image sequence that can be used for precise AI recognition.
[0011] Next, the AI-powered intelligent defect identification and quantitative analysis of microscopic images is performed: a dedicated microscopic defect identification model for consumer products is invoked to perform pixel-level feature comparison and extraction on the collected standardized microscopic images, automatically distinguishing between normal textures and abnormal defects in the samples, and accurately identifying five types of safety hazards: fiber breakage, coating impurities, chemical residue crystallization, pore defects, and foreign matter contamination; at the same time, the area, quantity, and distribution density of each type of defect are quantitatively calculated, and accurate digital detection data is output, completely eliminating the subjective judgment mode.
[0012] Next, the safety level classification of consumer products is determined: the system pre-stores the national and industry safety testing classification thresholds for daily consumer products, compares the quantitative defect data output by AI with the preset thresholds, and automatically classifies the samples into four risk levels: A-level compliant and high-quality, B-level minor defects, C-level safety hazards, and D-level serious non-compliance. The system also outputs standardized test conclusions, risk descriptions, and compliance handling suggestions accordingly, so as to achieve standardization, classification, and quantification of test results.
[0013] The system then performs intelligent layered synthesis of the live stream footage and dynamic traceability watermark overlay: The system adaptively adjusts the live stream display strategy according to the sample risk level. Samples of normal risk level adopt the normal split-screen display mode, while high-risk samples of level C and D automatically trigger local magnification of defect areas and highlighting of outlines and focusing. The sample risk level label, quantitative defect data, and standardized test conclusions are written into the traceability code watermark layer in real time, and the video watermark information is dynamically updated. This is different from the traditional fixed static watermark and realizes real-time synchronization of test results and traceability data.
[0014] The following steps were then implemented: real-time risk warning and simultaneous public display on both online and offline platforms. For samples identified as having a Class C safety hazard or a Class D serious non-compliance, a bright red warning pop-up window appeared at the front of the live stream screen, indicating the risk level and hazard type, accompanied by real-time voice risk warnings. The synthesized complete live stream video was compressed and adapted for network use through layered encoding technology and pushed to major cloud live streaming platforms with a low-latency protocol. At the same time, it was simultaneously output to a local large display screen, enabling online viewers and offline on-site personnel to view the risk warning information in real time.
[0015] Finally, the system completes hierarchical data archiving and iterative updates for traceability: AI defect identification data, risk rating results, test conclusions, full live stream recordings, and key screenshots focusing on defects are all bound to dynamic intelligent inspection traceability codes and archived in the sample intelligent inspection code traceability database. End users can access the complete set of sample test data, risk rating reports, and full live stream evidence videos with a single click by scanning the traceability QR code on the live stream screen. Simultaneously, the system receives online user submission requests in real time, automatically generating new testing tasks and unique dynamic traceability codes, iteratively completing a new round of intelligent testing, live streaming, and traceability closed-loop processes.
[0016] This invention employs intelligent algorithm judgment, dynamic data binding, and hierarchical risk warning logic throughout the entire process, eliminating the need for human experience-based judgment. It achieves standardization, intelligentization, traceability, and early warning upgrades for live microscopic testing of daily consumer products, and is applicable to various scenarios such as compliance quality inspection live broadcasts, science popularization live broadcasts, and evidence submission for consumer products.
Claims
1. A live-stream traceability and early warning method for safety grading and testing of daily consumer goods based on microscopic intelligent recognition, characterized in that, Includes the following steps: S1. Sample standardization and dynamic traceability code initialization: Enter the category, production batch, sampling source, and compliance standard parameters of the daily consumer goods to be tested, establish a standardized sample testing file, and generate a unique intelligent inspection traceability code with dynamic data update function. The traceability code reserves a port for writing defect data, risk level, and test conclusion. S2, Multi-camera frame-level synchronous intelligent adaptive acquisition: Based on the PTP clock synchronization protocol, it completes microsecond-level synchronous acquisition of three-channel images: microscopic imaging, product panorama, and operation demonstration. It automatically adapts the light source color temperature, exposure parameters, and sampling frame rate according to the sample material type. It adaptively switches imaging modes for flexible fabrics, rigid consumables, and porous materials to obtain standardized and distortion-free microscopic image sequences. S3. Microscopic Image AI Intelligent Defect Recognition and Quantitative Analysis: By using a preset consumer product microscopic defect recognition model, pixel-level features are extracted from microscopic images. Five types of microscopic safety hazards are automatically identified: fiber breakage, coating impurities, chemical residue crystallization, hole defects, and foreign matter contamination. The defect area, defect quantity, and defect distribution density are quantitatively calculated to generate quantitative detection data. S4. Consumer Product Safety Level Classification: The system presets safety testing and classification threshold standards for daily consumer products. Based on AI quantitative testing data, the system automatically classifies samples into four risk levels: Grade A (compliant and high-quality), Grade B (minor defects), Grade C (safety hazard), and Grade D (serious non-compliance). The system then generates standardized testing conclusions and risk management recommendations accordingly. S5. Intelligent layered synthesis and dynamic traceability watermark overlay of live broadcast: Adaptively switch between live broadcast split-screen layout and defect focus mode according to the detection risk level, and highlight and magnify the micro defect areas of high-risk samples; dynamically embed the risk level label, quantitative detection data and detection conclusion into the traceability watermark layer to realize real-time iterative update of traceability data; S6. Live Streaming Real-Time Risk Warning and Dual-End Synchronous Publicity: For C-level and D-level risk samples, trigger the live stream screen to highlight the pop-up warning and the voice risk broadcast. The synthesized video stream is layered and encoded to push it to the cloud live streaming platform with low latency. At the same time, it is simultaneously output to the local display screen to complete the dual-end real-time publicity. S7. Hierarchical Data Archiving and Iterative Updates for Traceability: Quantitative defect data, risk levels, test results, live recordings, defect screenshots, and dynamic intelligent inspection traceability codes are bound and archived to the traceability database. Users can scan the code to query complete traceability test information. At the same time, new test traceability tasks are automatically generated iteratively based on online test submission requirements.
2. The method for live-stream traceability and early warning of safety classification testing of daily consumer goods based on microscopic intelligent recognition as described in claim 1, characterized in that: In step S3, the micro-defect recognition model is a machine vision model specifically trained for textile fabrics, baby products, and daily chemical consumables. It is used to distinguish between normal textures and abnormal defects in samples, and to avoid misidentification and missed identification.
3. The method for live-stream traceability and early warning of safety classification testing of daily consumer goods based on microscopic intelligent recognition according to claim 1, characterized in that: In step S4, the grading thresholds for the four risk levels correspond to national and industry consumer product safety testing standards, enabling quantitative, standardized, and error-free testing and judgment.
4. The method for live-stream traceability and early warning of safety classification and testing of daily consumer goods based on microscopic intelligent recognition as described in claim 1, characterized in that: In step S5, the dynamic traceability watermark adopts an encrypted embedded overlay method, ensuring that the data is not lost or tampered with during video streaming, transcoding, and playback, thus guaranteeing the authenticity and uniqueness of the traceability evidence.
5. The method for live-stream traceability and early warning of safety classification and testing of daily consumer goods based on microscopic intelligent recognition according to claim 1, characterized in that: In step S6, the risk warning level is linked to the sample risk level. A D-level seriously non-conforming sample triggers the highest level red warning and forces the enlarged defect image to be displayed at the top.