Intelligent automatic vending system and method integrated with traditional Chinese medicine four-diagnosis detection function

Abstract

The invention relates to the technical field of intelligent medical instruments and health management, and discloses an intelligent automatic vending system integrated with a traditional Chinese medicine four-diagnosis detection function, which comprises an equipment main body adopting a vertical cabinet structure, and the front side of the equipment main body is provided with a high-definition camera, an interaction screen, function keys, a commodity taking port and a telescopic and movable mechanical arm type detection assembly; the traditional Chinese medicine four-diagnosis information collection module is used for sequentially collecting the tongue condition, the face condition, the pulse condition and inquiry information of the user; the tongue image and the face image are shot by the high-definition camera. According to the system, automatic traditional Chinese medicine four-diagnosis information collection, intelligent physique identification and syndrome differentiation based on a knowledge graph, personalized health care product recommendation, automatic vending delivery and user feedback collection are integrated into single equipment, a user can complete the whole process from health state detection to matched health care product acquisition in one-time interaction without manual intervention, and the user experience is improved. The service efficiency and convenience are greatly improved, and the industrial pain point that the inspection is not determined and the determination is not determined is thoroughly solved.

Description

Technical Field

[0001] This invention relates to the field of intelligent medical devices and health management technology, specifically to an intelligent automated vending system and method that integrates the four diagnostic functions of traditional Chinese medicine. Background Technology

[0002] In traditional health service models, TCM constitution testing and health product purchase are two separate steps. Users typically need to undergo TCM four diagnostic methods (inspection, auscultation, inquiry, and palpation) at a professional institution to obtain a constitution report, and then find and purchase corresponding health products based on the report. The whole process is time-consuming and laborious, and there are problems such as inaccurate product matching and inconvenient purchase channels.

[0003] In the existing technology, some devices have emerged that attempt to combine health services with retail. For example, Chinese utility model patent CN208477662U discloses an "automatic medicine vending machine with online consultation capability." This device connects with doctors via video call for online consultation and automatically sells medicines based on the doctor's recommendations. However, this technical solution has the following obvious shortcomings: 1. Reliance on human doctors: Its diagnosis and recommendations rely entirely on the human judgment of backend doctors. Service efficiency is limited by doctor resources and cannot achieve large-scale, real-time self-service.

[0004] 2. Lack of integrated automated TCM testing: It lacks the ability to automatically collect and analyze objective physiological indicators such as tongue appearance, facial appearance, and pulse appearance of users, resulting in highly subjective diagnostic criteria and low standardization.

[0005] 3. Single function and different target products: It was originally designed for the emergency sale of "medicines" rather than personalized recommendations and health management of "health products" based on TCM constitution identification. The two are fundamentally different in terms of applicable population, regulatory requirements and service logic.

[0006] 4. Lack of a closed-loop health record system: The lack of continuous identification of users and accumulation of health data makes it impossible to provide personalized and continuous health management services based on historical data. Summary of the Invention

[0007] The purpose of this invention is to provide an intelligent automated vending system and method that integrates the four diagnostic functions of traditional Chinese medicine, so as to solve the problems mentioned in the background art.

[0008] To achieve the above objectives, the present invention provides the following technical solution: an intelligent automated vending system integrating the four diagnostic methods of traditional Chinese medicine, comprising: The main body of the equipment adopts a vertical cabinet structure, and its front is equipped with a high-definition camera, an interactive screen, function buttons, a product retrieval port, and a retractable and movable robotic arm-type detection component. The TCM four diagnostic information collection module is used to sequentially collect the user's tongue appearance, facial appearance, pulse appearance, and consultation information; the tongue appearance and facial appearance are captured by the high-definition camera, the pulse appearance is automatically located and collected by the array pressure sensor mounted on the robotic arm detection component, and the consultation information is input through the interactive screen; The timing synchronization and acquisition guidance module is used to control the robotic arm detection component to automatically extend and align with the user's wrist to acquire pulse after the tongue and face images are acquired. It also provides real-time graphic guidance through the interactive screen to ensure that the acquisition actions of the four types of data (tongue, face, pulse, and question) are sequentially related in time and coordinated in space. The knowledge graph processing system is used to fuse and analyze collected multi-source heterogeneous data, perform constitution identification and health status assessment, and generate personalized health product recommendations based on the association rules in the knowledge graph. The automated vending module is used to automatically allocate and deliver products according to the recommended plan confirmed by the user; The user identification and health record linkage module is used to identify the user's identity through biometrics or QR code scanning before the user starts the test, and to retrieve the associated historical health records. The shipment verification and feedback collection unit is located in the automatic vending module. It is used to verify whether the product is shipped correctly before and after shipment using a camera or weight sensor, and to collect usage feedback through the interactive interface after the user takes the product.

[0009] As a preferred embodiment of the present invention, the TCM four diagnostic methods information collection module further includes: The tongue image acquisition unit is used to capture images of the tongue under standard lighting conditions; Facial image acquisition unit, used to capture facial images; The pulse information acquisition unit uses the array-type pressure sensor to acquire pulse signals. The interactive consultation unit collects user symptom information through structured questionnaires; The multimodal data quality control submodule is used to evaluate image clarity, signal stability, and questionnaire completeness in real time during the data acquisition process. If the quality of any data source is lower than the preset threshold, the re-acquisition process is triggered.

[0010] As a preferred embodiment of the present invention, the knowledge graph processing system includes: The data preprocessing unit is used for image segmentation and feature extraction of tongue and facial images, time-frequency analysis of pulse signals, and entity recognition and standardization of medical records. The graph mapping and reasoning unit is used to map standardized multimodal data to knowledge graph entities and generate constitution types and syndrome elements through the propagation of relationships on the graph structure. The recommendation generation unit generates a recommendation list based on the pre-defined relationships between constitution and product, and syndrome and product in the knowledge graph.

[0011] As a preferred embodiment of the present invention, the automatic vending module includes: Multi-temperature zone storage warehouse, used to store health products with different storage requirements; Mechanical transmission mechanism, used to perform cargo grabbing and pushing operations; The payment system supports multiple payment methods, including QR code scanning and card swiping. The dynamic inventory and recommendation matching submodule is used to monitor the inventory status of each channel in real time. When the recommended product is in short supply, it automatically retrieves alternative products from the knowledge graph and prioritizes them based on the similarity of the user's physical condition and symptoms.

[0012] As a preferred embodiment of the present invention, the system further includes: The cloud-based collaborative knowledge graph service platform is used to receive anonymous data from various terminal devices, perform cross-device health trend analysis, and regularly push knowledge graph update packages to the terminals. The device cluster scheduling and load balancing module is used to dynamically allocate user requests based on geographical location, usage frequency, and device status in multi-device deployment scenarios.

[0013] A smart automated vending method integrating the four diagnostic methods of traditional Chinese medicine includes the following steps: User identification steps: Before a user initiates the detection, the user's identity is identified through biometrics or QR code scanning, and their historical health records are retrieved. The steps for collecting information from the four diagnostic methods in Traditional Chinese Medicine are as follows: Users are guided to complete the collection of tongue, facial, pulse, and consultation information in sequence. Tongue and facial images are captured by a high-definition camera, pulse is automatically located and collected by an array of pressure sensors mounted on a retractable robotic arm, and consultation information is input through an interactive screen. During the collection process, real-time text and image guidance is provided on the screen to ensure the sequential and coordinated nature of the four types of data collection actions. Data preprocessing and feature extraction steps: Preprocess and extract features from the collected multi-source heterogeneous data; Knowledge graph reasoning and health assessment steps: Map feature data to knowledge graph entities, and generate physical constitution type and health status assessment through relation matching and logical reasoning on the graph structure; Personalized recommendation generation steps: Generate personalized health product recommendations based on association rules in the knowledge graph; User confirmation and payment steps: The user confirms the recommendation and completes the payment through the interactive interface; Automatic dispensing and verification steps: The automatic vending module executes the dispensing instruction and verifies whether the product is dispensed correctly before and after dispensing using a camera or weight sensor; Feedback collection and record update steps: After the user picks up the goods, use feedback is collected through the interactive interface, and the test results and feedback data are linked to update the user's health record.

[0014] As a preferred embodiment of the present invention, the data preprocessing and feature extraction steps include: The tongue image is segmented into tongue body and tongue coating, and color, texture and morphological features are extracted respectively; Facial region localization and skin color and gloss analysis were performed on face images. Filtering, segmentation, and time-frequency feature extraction of pulse signals; Keyword extraction and symptom entity recognition are performed on the consultation text.

[0015] As a preferred embodiment of the present invention, the knowledge graph reasoning and recommendation generation steps include: The extracted feature vectors are semantically matched with entity nodes in the knowledge graph. Multi-step relational reasoning is performed using an attention allocation mechanism on a graph structure to generate constitution type and syndrome elements; A recommendation list is generated based on the correlation between body constitution and products, as well as between symptoms and products. A multi-objective optimization strategy is introduced, which simultaneously considers product matching degree, user preferences, inventory status and business rules during the recommendation process.

[0016] As a preferred embodiment of the present invention, the method further includes: The steps for collecting and analyzing user behavior data include recording the time users spend on the interactive interface, their click behavior, and their purchase decisions. The knowledge graph is dynamically updated by optimizing graph nodes and relationships through incremental learning based on user feedback and new clinical evidence. The abnormal usage behavior detection and intervention steps identify abnormal usage behavior by analyzing the inconsistencies between user operation sequences and physiological data, and trigger secondary verification or manual review processes.

[0017] As a preferred embodiment of the present invention, the method further includes: After shipment, health maintenance suggestions and follow-up testing reminders will be pushed to users through the interactive interface; Knowledge graphs are updated in a distributed manner through a federated learning mechanism: each terminal device uses local data to train the model and only synchronizes parameter increments to the cloud.

[0018] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention creatively integrates automated TCM four diagnostic methods information collection, knowledge graph-based intelligent constitution identification and syndrome differentiation, personalized health product recommendation, automated vending and delivery, and user feedback collection into a single device. Users can complete the entire process from health status testing to obtaining matching health products in a single interaction without human intervention, greatly improving service efficiency and convenience, and completely solving the industry pain point of "testing without decision, and decision without product".

[0019] 2. Hardware collaborative innovation ensures the standardization and efficiency of data acquisition: By introducing a retractable robotic arm detection component and a timing synchronization and acquisition guidance module, the system realizes the sequential, automated, and spatial collaborative control of tongue, facial, and pulse image acquisition. This not only simplifies user operation but also ensures the standardization and consistency of multimodal TCM data acquisition, laying a reliable physical data foundation for subsequent accurate analysis and overcoming the problem of inconsistent data quality caused by the reliance on subjective user cooperation in traditional self-service equipment.

[0020] 3. A dynamic health management model based on continuous identity recognition was constructed: Through the user identity recognition and health record linkage module, the system establishes and maintains a unique time-series health record for each user. Data from each service is recorded and associated, enabling physical constitution identification and recommendations to refer to historical trends. This achieves a leap from single service to continuous and personalized health management, making recommendations more forward-looking and personalized.

[0021] 4. Through a collaborative update mechanism of user feedback data and knowledge graph (such as federated learning), the system can continuously optimize the recommendation algorithm, possessing self-learning and evolution capabilities. The shipment verification and feedback collection unit and abnormal usage behavior detection mechanism effectively ensure transaction accuracy, data quality, and system security, enhancing user trust. The multi-objective optimization recommendation strategy considers physical matching degree while taking into account user preferences, inventory status, and business rules. The generated recommendation list is closer to real consumption scenarios, significantly improving recommendation acceptance rate, user repurchase rate, and customer satisfaction. Attached Figure Description

[0022] Figure 1 This is a schematic diagram illustrating the workflow and collaboration of the TCM four diagnostic methods information collection module in this invention; Figure 2 This is a schematic diagram of the data processing, reasoning, and recommendation generation process of the knowledge graph processing system in this invention; Figure 3 This is a schematic diagram of the entire user service process in an embodiment of the present invention. Detailed Implementation

[0023] The technical solutions of the present invention will be clearly and completely described below with reference to the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0024] Example 1: Verification of System Infrastructure and End-to-End Collaborative Work This embodiment aims to verify the integrity and availability of the basic hardware structure of the system to be protected, the collaborative workflow between core modules, and the basic service functions.

[0025] 1.1 System Hardware Deployment and Module Composition: An intelligent automatic vending system of this invention was deployed in the lobby of a commercial park in Nanshan District, Shenzhen. The main body of the equipment adopts a vertical cabinet structure, and the following hardware units are integrated on its front: High-definition camera: used to capture images of the user's tongue (tongue image) and face (facial image) under standard lighting conditions.

[0026] Interactive screen: Provides full-process graphic guidance, consultation questionnaire input, recommendation result display and payment interface.

[0027] Retractable robotic arm detection component: Located on the right side of the cabinet, this component retracts into the cabinet when not in use. When in use, it is automatically extended under the control of the timing synchronization and acquisition guidance module. An array of pressure sensors is fixed at the end of the robotic arm to automatically acquire the user's radial artery pulse signal under three pulse-taking pressures: floating, middle, and deep.

[0028] Goods pickup point: located at the bottom of the cabinet.

[0029] Multi-temperature zone storage compartment: Located inside the cabinet, it has two independent temperature zones, one for ambient temperature and one for refrigeration (4℃), for storing health products with different requirements for storage temperature.

[0030] Control motherboard: integrates the drive circuits and communication interfaces of each module.

[0031] Identity recognition device: Integrated above the interactive screen, supporting QR code scanning and facial recognition camera.

[0032] 1.2 Software and Data System Initialization: The system has a built-in knowledge graph processing system. This knowledge graph is constructed based on the "Classification and Judgment of Traditional Chinese Medicine Constitution", "Traditional Chinese Materia Medica" and "Formulary". It includes nine basic constitution types (such as balanced constitution, qi deficiency constitution, yang deficiency constitution, etc.), dozens of common syndrome elements (such as "damp heat", "blood stasis", "qi deficiency", etc.) and the relationship of more than 100 registered health food products.

[0033] The system also deploys a backend service for a user identification and health record linkage module to create a unique identifier (User ID) and a blank health record for new users.

[0034] 1.3 Implementation of the entire user service process: We are recruiting 300 white-collar volunteers from the park to participate in a 30-day experience test. The process for a single service session is as follows: Step S101: User identification and file retrieval. When the user clicks "Start Detection" on the screen, the system activates the user identification and health record linkage module. Existing users can quickly log in by scanning their personal QR code or using facial recognition; new users are guided to complete a simple registration. After logging in, the system automatically retrieves the user's historical health records (if they exist) from the cloud or local storage. The records contain the body type, health status assessment results, and purchase records for each test.

[0035] Step S102: Sequential collection of information from the four diagnostic methods of Traditional Chinese Medicine.

[0036] 1. Tongue Image Acquisition: The screen displays a standard tongue image example and the text prompt "Please stick out your tongue naturally and hold for 3 seconds." The timing synchronization and acquisition guidance module controls the high-definition camera to focus on the user's oral cavity area to complete the image capture.

[0037] 2. Facial Image Acquisition: After the tongue image acquisition is completed, the screen prompts "Please look straight ahead and take a picture of your face". The camera automatically adjusts the focus and range to capture an image that includes the entire facial area.

[0038] 3. Pulse Acquisition: After facial image acquisition is completed, the screen displays an image of the wrist placement area and prompts "Please place your wrist flat here". At the same time, the timing synchronization and acquisition guidance module controls the robotic arm detection component to extend smoothly from the cabinet. The end sensor array automatically locates and gently touches the radial artery on the user's wrist. The screen displays a countdown (about 90 seconds) and a "Please remain still" prompt. Under program control, the robotic arm acquires pulse signals in sequence at three pressure levels: floating, medium, and deep.

[0039] 4. Medical Information Collection: After the pulse diagnosis is completed, the robotic arm automatically retracts, and a structured interactive medical questionnaire pops up on the screen, containing about 20 multiple-choice questions about recent symptoms and lifestyle habits.

[0040] Throughout the data collection process, the multimodal data quality control submodule operates in real time. If it detects a blurred tongue image, unstable signal due to wrist movement, or incomplete questionnaire, the system will immediately provide a clear prompt to re-collect the data on the screen.

[0041] Step S103: Data processing, identification and recommendation. The collected data packages (tongue image, face image, pulse signal, and consultation text) are sent to the knowledge graph processing system.

[0042] The data preprocessing unit performs image segmentation and feature extraction, pulse signal filtering and time-frequency feature extraction, and keyword entity recognition on the consultation text.

[0043] The data preprocessing unit performs image segmentation and feature extraction, pulse signal filtering and time-frequency feature extraction, and keyword entity recognition on the consultation text.

[0044] The graph mapping and reasoning unit maps the extracted features to the corresponding entity nodes in the knowledge graph (e.g., "red tongue" maps to the "heat" syndrome node, and "yellow and greasy tongue coating" maps to the "dampness" syndrome node), and propagates through the relational paths on the graph to ultimately infer the user's dominant constitution type and core health problems (syndromes).

[0045] Based on the reasoning results, the recommendation generation unit queries the association edges of "constitution-product" and "syndrome-product" in the knowledge graph to generate a personalized health product recommendation list (usually containing 1-3 products) sorted by matching degree. This list is clearly displayed on the interactive screen and comes with a brief explanation based on the substructure of the knowledge graph, such as "Your current constitution is biased towards 'damp-heat'. Product A is recommended because it contains ingredients such as Poria cocos and Coix seed, which help to strengthen the spleen and remove dampness."

[0046] Step S104: User Confirmation, Payment, and Shipment. After viewing the recommended results, the user can confirm the purchase. The system will redirect to the payment interface, supporting QR code payment. After successful payment, the automatic vending module will start. The dynamic inventory and recommendation adaptation submodule first confirms the inventory of recommended products in the product channels.

[0047] The mechanical transmission mechanism moves to the corresponding cargo channel according to the instructions, grabs the product and transports it to the picking port.

[0048] The shipment verification and feedback collection unit is activated: Before shipment, a miniature camera inside the cargo channel takes an image of the empty exit; after the product slides out, it takes another image and compares the images. At the same time, the weight sensor at the bottom of the cargo channel records the weight change. After the double verification is passed, the screen displays "Successful pickup".

[0049] Step S105: Feedback collection and file update. After the user picks up the goods, a short feedback questionnaire pops up on the screen, such as "Are you satisfied with the matching degree of the recommended products? (1-5 points)". After the feedback is submitted, the user identification and health record linkage module encrypts the complete service data (collected data, identification results, recommended products, payment records, feedback scores) and updates it synchronously to the user's historical health record, forming a new time-series record.

[0050] 1.4 Data Collection and Analysis Collect the following key metrics: Average total time per service: from identity verification to successful pickup.

[0051] Accuracy of constitution identification: The results of independent offline four diagnostic methods for the same user by two associate chief physicians were compared as the "gold standard".

[0052] Product recommendation acceptance rate: The percentage of users who confirm a purchase.

[0053] User satisfaction: Scores (1-5 points) on the operation process and product matching degree are collected through questionnaires.

[0054] Table 1: Core Performance and User Experience Data of Example 1 Evaluation indicators result Average total time per service 5.8 ± 0.9 minutes Accuracy of body constitution identification 91.3％（274 / 300） Product recommendation acceptance rate 78.0％（234 / 300） Operation process guides satisfaction 4.6 ± 0.5 points Recommended product matching satisfaction 4.4 ± 0.6 points 1.5 Analysis and Explanation: This embodiment verifies the effective integration and collaboration of the system's core hardware (robotic arm, camera, screen) and core modules (four diagnostic methods data collection, timing control, knowledge graph, automatic vending, and identity linkage). Under the control of the timing synchronization and data collection guidance modules, the multimodal data collection process is smooth, with an average time consumption of less than 6 minutes, resulting in a good user experience and a constitution identification accuracy rate as high as 91.3%. This demonstrates the effectiveness of the knowledge graph processing system in fusing and reasoning multi-source heterogeneous TCM data, which is also the basis for the high recommendation acceptance rate (78%). The user identity recognition and health record linkage module ensures the continuity of service and provides a data foundation for subsequent personalized optimization. The shipment verification and feedback collection unit constitutes a closed-loop guarantee for service quality.

[0055] Example 2: Verification of Advanced Functions and System Dynamic Optimization Performance This embodiment focuses on verifying the system's self-learning evolution capability and intelligent decision optimization function, especially the recommendation strategy under multi-objective constraints.

[0056] 2.1 System Deployment and Optimization Mechanism Enabled This system was deployed in three chain pharmacies in Haidian District, Beijing, and the following advanced functions were enabled: User behavior feedback and knowledge graph collaborative update mechanism: The system records user feedback ratings and subsequent repurchase behavior.

[0057] Federated learning mechanism: Every two weeks, each terminal device uses locally anonymized user feedback data to train a lightweight model locally, which is used to fine-tune the weights of the "syndrome-product" association edges in the knowledge graph (i.e., the product efficacy confidence). Only the incremental model parameters are encrypted and uploaded to the cloud collaborative knowledge graph service platform. The cloud platform aggregates the parameters of each terminal, generates an updated knowledge graph model, and then distributes it to each terminal.

[0058] Multi-objective optimization strategy: In the recommendation generation stage, not only the matching degree between products and physical condition is considered, but also real-time inventory, users' historical purchase preferences, and business rules (such as new product promotion and profit weighting) are considered simultaneously.

[0059] 2.2 Implementation process and control settings: The 60-day testing period is divided into two phases: Phase 1 (Days 1-30): All devices use an initial, unoptimized knowledge graph for recommendations, and the recommendations are based solely on body type compatibility.

[0060] Phase Two (Days 31-60): All devices use the knowledge graph optimized through federated learning based on user feedback data from Phase One. During this phase, users are randomly divided into two groups: Group A: Use a multi-objective optimization strategy for recommendations.

[0061] Group B: Recommendations will still be made based solely on baseline physical fitness matching.

[0062] 2.3 Data Collection and Analysis: Monitor and compare the following metrics between the two groups of users in the two phases and within Phase Two: Recommendation acceptance rate; User repurchase rate (users use the device again and make a purchase within 30 days); Click-through rate of recommended lists (the percentage of products in the recommended list that are ultimately clicked by users to view details, reflecting the attractiveness of the list).

[0063] Table 2: Comparison of System Optimization Before and After Implementation and the Effects of Different Recommendation Strategies in Example 2 Group / Phase Recommendation acceptance rate User repurchase rate Recommended list click-through rate Phase 1 (All, Initial Map) 75.5％ 15.2％ not applicable Phase Two - Group A (Optimization of the Atlas + Multi-Objective Strategy) 83.7％ 22.5％ 68％ Phase Two - Group B (Optimized Graph + Basic Strategy) 79.1％ 18.8％ 52％ 2.4 Analysis and Explanation: This embodiment profoundly reveals the system's intelligent evolution and decision optimization capabilities.

[0064] First, comparing the overall data from Phase 1 and Phase 2, both the recommendation acceptance rate and repurchase rate have significantly improved. This proves the effectiveness of the user behavior feedback and knowledge graph collaborative update module (achieved through federated learning). The system is not a static tool; its recommendation accuracy can continuously evolve with the accumulation of actual usage data.

[0065] Secondly, within Phase Two, Group A (which enabled a multi-objective optimization strategy) significantly outperformed Group B across all metrics, particularly in the click-through rate of the recommendation list (68% vs 52%). This indicates that Group A's recommendation list was more attractive and valuable for users to explore. This is because the multi-objective optimization strategy not only recommends "the right products" but also integrates "products that users might like," "products in stock," and "products that meet operational goals," thereby generating a more pragmatic, intelligent, and user-friendly recommendation solution. This represents a leap from "algorithmic precision" to "business intelligence."

[0066] Example 3: Verification of System Robustness, Anomaly Handling, and Security Assurance This embodiment aims to simulate extreme or abnormal usage scenarios to verify the quality control, security, and fault tolerance mechanisms involved in the system.

[0067] 3.1 Test Scenario: Simulate the following scenario in a controlled laboratory environment: Scenario A (Poor Data Acquisition Quality): Invite 10 volunteers to rapidly extend and retract their tongues during tongue image acquisition (simulating image blurring) and to slightly and regularly rotate their wrists during pulse image acquisition (simulating signal interference).

[0068] Scenario B (Abnormal / Malicious Use): Simulate the same user account initiating 3 detection requests in a row within 10 minutes, and the system detects that the physiological data (such as average heart rate) of the three times has medically unreasonable drastic fluctuations (e.g., from 60 beats / min to 120 beats / min).

[0069] Scenario C (Shipping Process Abnormality): A slight obstruction in the cargo channel is artificially created to simulate a shipping failure.

[0070] 3.2 Implementation process and system response observation: For scenario A: Observe the response of the multimodal data quality control submodule. It is expected that the image blur and signal-to-noise ratio should be identified in real time as being below the threshold, and a prompt should be immediately displayed on the interactive screen: "Tongue image acquisition is unclear, please stick out your tongue again to keep it stable" or "Pulse signal is unstable, please keep your wrist still" and trigger the reacquisition process.

[0071] For scenario B: Observe whether the abnormal usage behavior detection and intervention steps are triggered. The system should determine the abnormal behavior by analyzing the inherent contradiction between the user's operation sequence (high-frequency detection in a short period of time) and physiological data.

[0072] For scenario C: Observe the operation of the shipment verification and feedback collection unit. After the mechanical transmission mechanism executes the shipment command, the verification unit should be able to detect through image comparison or weight sensing that the product has not fallen into the picking port normally.

[0073] 3.3 Data Collection and Analysis: Record the success rate of the system's defense and correction mechanisms in various scenarios.

[0074] Table 3: System Robustness and Anomaly Handling Test Results in Example 3 Test Scenario Number of tests System Response and Results Scenario A (Poor Data Quality) 50 times 98% (49 / 50) of the times triggered a re-collection prompt, and 96% (47 / 49) of those times the data quality met the standards after re-collection. Scenario B (Abnormal usage behavior) 20 times 100% (20 / 20) were identified as abnormal by the system. Among them, 85% (17 / 20) triggered secondary verification (by answering a random security question on an interactive screen), and 15% (3 / 20) were directly transferred to the manual review queue due to obvious data contradictions. Scenario C (Shipping process abnormal) 30 times If 100% (30 / 30) of the shipments are identified by the shipment verification unit, the system will automatically retry the shipment once. If it fails again, the transaction will be terminated, a refund will be issued via the original payment method, and a fault alarm log will be generated in the background. 3.4 Analysis and Explanation: This embodiment fully verifies the system's stability and security when faced with non-ideal inputs or malicious operations. The extremely high data quality resampling rate ensures the reliability of input data, which is the prerequisite for all subsequent accurate services. The 100% identification rate of abnormal usage behavior and the tiered intervention (from secondary verification to manual review) effectively prevent resource abuse or fraudulent behavior, protecting the interests of the system and legitimate users. The shipment verification mechanism constitutes the last line of insurance in the transaction process, avoiding economic losses and user complaints caused by mechanical failures, and greatly enhancing the system's commercial reliability and user trust.

[0075] Comparison with Example 1: Compared with the traditional split service model To highlight the advantages of the "integrated and closed-loop" service of this invention, a traditional comparison mode is set up: In the same business park, a traditional, single-function "AI TCM Four Diagnostic Instrument" (similar to CN106580257A) is set up. After the user completes the test, he / she obtains a paper constitution report. The user needs to take the report to a nearby pharmacy and, with the help of the pharmacy staff (non-professional TCM doctors), try to find health products related to the "suggested conditioning direction" on the report.

[0076] Results: The average time to complete the process from "testing" to "obtaining the product" was approximately 25 minutes. The average satisfaction score of users on the "final product purchased" and "match with their own physical condition" was only 3.1 points. More than 40% of users indicated that they could not find the specific health products indirectly mentioned in the report or the ones they were looking for in their understanding at the pharmacy.

[0077] Analysis: This comparative example vividly highlights the immense value of this invention. Through hardware integration and software scheduling, this invention seamlessly connects multiple stages such as detection, analysis, decision-making, sales, and feedback into a single physical entity and a single user interaction, greatly improving service efficiency (reducing it from 25 minutes to approximately 6 minutes) and user experience (increasing the matching satisfaction score from 3.1 to over 4.4). It completely solves the core pain points in health services, such as "detection without decision, decision without substance, and substance not matching the symptoms."

[0078] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. An intelligent automated vending system integrating the four diagnostic methods of traditional Chinese medicine, characterized in that, include: The main body of the equipment adopts a vertical cabinet structure, and its front is equipped with a high-definition camera, an interactive screen, function buttons, a product retrieval port, and a retractable and movable robotic arm-type detection component. The TCM four diagnostic information collection module is used to sequentially collect the user's tongue appearance, facial appearance, pulse appearance, and consultation information; the tongue appearance and facial appearance are captured by the high-definition camera, the pulse appearance is automatically located and collected by the array pressure sensor mounted on the robotic arm detection component, and the consultation information is input through the interactive screen; The timing synchronization and acquisition guidance module is used to control the robotic arm detection component to automatically extend and align with the user's wrist to acquire pulse after the tongue and face images are acquired. It also provides real-time graphic guidance through the interactive screen to ensure that the acquisition actions of the four types of data (tongue, face, pulse, and question) are sequentially related in time and coordinated in space. The knowledge graph processing system is used to fuse and analyze collected multi-source heterogeneous data, perform constitution identification and health status assessment, and generate personalized health product recommendations based on the association rules in the knowledge graph. The automated vending module is used to automatically allocate and deliver products according to the recommended plan confirmed by the user; The user identification and health record linkage module is used to identify the user's identity through biometrics or QR code scanning before the user starts the test, and to retrieve the associated historical health records. The shipment verification and feedback collection unit is located in the automatic vending module. It is used to verify whether the product is shipped correctly before and after shipment using a camera or weight sensor, and to collect usage feedback through the interactive interface after the user takes the product.

2. The intelligent automatic vending system integrating the four diagnostic functions of traditional Chinese medicine as described in claim 1, characterized in that, The TCM four diagnostic methods information collection module also includes: The tongue image acquisition unit is used to capture images of the tongue under standard lighting conditions; Facial image acquisition unit, used to capture facial images; The pulse information acquisition unit uses the array-type pressure sensor to acquire pulse signals. The interactive consultation unit collects user symptom information through structured questionnaires; The multimodal data quality control submodule is used to evaluate image clarity, signal stability, and questionnaire completeness in real time during the data acquisition process. If the quality of any data source is lower than the preset threshold, the re-acquisition process is triggered.

3. The intelligent automatic vending system integrating the four diagnostic functions of traditional Chinese medicine as described in claim 1, characterized in that, The knowledge graph processing system includes: The data preprocessing unit is used for image segmentation and feature extraction of tongue and facial images, time-frequency analysis of pulse signals, and entity recognition and standardization of medical records. The graph mapping and reasoning unit is used to map standardized multimodal data to knowledge graph entities and generate constitution types and syndrome elements through the propagation of relationships on the graph structure. The recommendation generation unit generates a recommendation list based on the pre-defined relationships between constitution and product, and syndrome and product in the knowledge graph.

4. The intelligent automatic vending system integrating the four diagnostic functions of traditional Chinese medicine as described in claim 1, characterized in that, The automated vending module includes: Multi-temperature zone storage warehouse, used to store health products with different storage requirements; Mechanical transmission mechanism, used to perform cargo grabbing and pushing operations; The payment system supports multiple payment methods, including QR code scanning and card swiping. The dynamic inventory and recommendation matching submodule is used to monitor the inventory status of each channel in real time. When the recommended product is in short supply, it automatically retrieves alternative products from the knowledge graph and prioritizes them based on the similarity of the user's physical condition and symptoms.

5. The intelligent automatic vending system integrating the four diagnostic functions of traditional Chinese medicine as described in claim 1, characterized in that, The system also includes: The cloud-based collaborative knowledge graph service platform is used to receive anonymous data from various terminal devices, perform cross-device health trend analysis, and regularly push knowledge graph update packages to the terminals. The device cluster scheduling and load balancing module is used to dynamically allocate user requests based on geographical location, usage frequency, and device status in multi-device deployment scenarios.

6. An intelligent automated vending method integrating the four diagnostic functions of traditional Chinese medicine, characterized in that, Includes the following steps: User identification steps: Before a user initiates the detection, the user's identity is identified through biometrics or QR code scanning, and their historical health records are retrieved. The steps for collecting information from the four diagnostic methods in Traditional Chinese Medicine are as follows: Users are guided to complete the collection of tongue, facial, pulse, and consultation information in sequence. Tongue and facial images are captured by a high-definition camera, pulse is automatically located and collected by an array of pressure sensors mounted on a retractable robotic arm, and consultation information is input through an interactive screen. During the collection process, real-time text and image guidance is provided on the screen to ensure the sequential and coordinated nature of the four types of data collection actions. Data preprocessing and feature extraction steps: Preprocess and extract features from the collected multi-source heterogeneous data; Knowledge graph reasoning and health assessment steps: Map feature data to knowledge graph entities, and generate physical constitution type and health status assessment through relation matching and logical reasoning on the graph structure; Personalized recommendation generation steps: Generate personalized health product recommendations based on association rules in the knowledge graph; User confirmation and payment steps: The user confirms the recommendation and completes the payment through the interactive interface; Automatic dispensing and verification steps: The automatic vending module executes the dispensing instruction and verifies whether the product is dispensed correctly before and after dispensing using a camera or weight sensor; Feedback collection and record update steps: After the user picks up the goods, use feedback is collected through the interactive interface, and the test results and feedback data are linked to update the user's health record.

7. The method according to claim 6, characterized in that, The data preprocessing and feature extraction steps include: The tongue image is segmented into tongue body and tongue coating, and color, texture and morphological features are extracted respectively; Facial region localization and skin color and gloss analysis were performed on face images. Filtering, segmentation, and time-frequency feature extraction of pulse signals; Keyword extraction and symptom entity recognition are performed on the consultation text.

8. The method according to claim 6, characterized in that, The knowledge graph reasoning and recommendation generation steps include: The extracted feature vectors are semantically matched with entity nodes in the knowledge graph. Multi-step relational reasoning is performed using an attention allocation mechanism on a graph structure to generate constitution type and syndrome elements; A recommendation list is generated based on the correlation between body constitution and products, as well as between symptoms and products. A multi-objective optimization strategy is introduced, which simultaneously considers product matching degree, user preferences, inventory status and business rules during the recommendation process.

9. The method according to claim 6, characterized in that, The method further includes: The steps for collecting and analyzing user behavior data include recording the time users spend on the interactive interface, their click behavior, and their purchase decisions. The knowledge graph is dynamically updated by optimizing graph nodes and relationships through incremental learning based on user feedback and new clinical evidence. The abnormal usage behavior detection and intervention steps identify abnormal usage behavior by analyzing the inconsistencies between user operation sequences and physiological data, and trigger secondary verification or manual review processes.

10. The method according to claim 6, characterized in that, The method further includes: After shipment, health maintenance suggestions and follow-up testing reminders will be pushed to users through the interactive interface; Knowledge graphs are updated in a distributed manner through a federated learning mechanism: each terminal device uses local data to train the model and only synchronizes parameter increments to the cloud.

Images