A wearable face recognition infusion check method and device based on end-edge-cloud cooperation and local cloud deployment

Abstract

The application belongs to the technical field of medical infusion management, and in particular to a wearable face recognition infusion check method and device based on end-edge-cloud collaboration and local cloud deployment. The device is suitable for fast and accurate checking of the identity of a clinical patient and infusion information, and has the following advantages: the end-side device is designed without a battery, is light in weight, and is comfortable and burden-free to wear; local cloud deployment and in-hospital Wi-Fi encrypted transmission keep medical data in the hospital throughout the process, meet the third level of network security protection and the requirement of medical data localization, and greatly reduce the risk of data leakage; the edge-side device has a simple structure and reduced weight; dual computing units are used for load distribution, and the multi-task concurrent processing capacity is enhanced; audio and visual dual-mode feedback reduces the operation burden of medical staff; multiple terminals are supported, and functions such as drug allergy early warning and infusion speed monitoring can be expanded, thereby meeting the needs of hospitals of different scales.

Description

Technical Field

[0001] This invention belongs to the field of medical infusion management technology, specifically relating to a wearable face recognition infusion verification method and device based on end-edge-cloud collaboration and local cloud deployment. Background Technology

[0002] Intravenous infusion is one of the most widely used treatment methods in clinical practice, playing an irreplaceable role in anti-infection, treatment of electrolyte imbalances, and nutritional support. Verification of patient and medication information is crucial during intravenous infusion. Clinically, nurses typically verify patient and medication information. This verification includes patient name, bed number, hospital number, medication name, dosage, and administration time. Patient identification information, including name, bed number, and hospital number, is also necessary to avoid confusion due to the possibility of duplicate names.

[0003] Because nurses need to administer intravenous infusions to multiple patients, and the pre-prepared medications are stored in a centralized manner, the requirements for nurses to verify information are relatively high. In clinical practice, medical accidents have occurred where patients and medications do not match. If an error occurs where the infusion is not matched with the patient, it may affect the treatment effect and delay the recovery of the condition. In severe cases, it may cause serious allergic reactions, organ damage, or even endanger the patient's life. Summary of the Invention

[0004] The present invention aims to provide a wearable face recognition infusion verification method and device based on end-edge-cloud collaboration and local cloud deployment, so as to realize intelligent recognition, matching and verification of hospital patients during infusion.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: A wearable facial recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment is provided. The device includes: The device is a wearable smart information collection glasses, which includes a first processor, an image acquisition module, a barcode acquisition module, a diffraction grating display module, a first power management chip, and a first communication and power interface. The first communication and power interface is electrically connected to the first power management chip and the first processor. The first processor, the image acquisition module, the barcode acquisition module, and the diffraction grating display module are each electrically connected to the first power management chip, and the image acquisition module, the barcode acquisition module, and the diffraction grating display module are each electrically connected to the first processor. An edge-side device, specifically a portable edge computing terminal, includes a second processor, a memory, a second communication and power interface, a third communication and power interface, a Wi-Fi module, a battery, a speaker module, and a second power management chip. The battery is electrically connected to the second power management chip. The second processor, the second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second power management chip, and the second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second processor. The second communication and power interface is used to electrically connect to the first communication and power interface via a line. The local cloud-side device is an in-hospital server cluster deployed on the hospital's intranet and physically isolated from the public cloud. The edge-side device connects to the local cloud-side device through an in-hospital wireless access point based on Wi-Fi wireless network signals, supports concurrent access from multiple terminals, and enables patient data storage, face model training and distillation, device management, log auditing, and HIS / LIS system integration.

[0006] Preferably, the first processor of the end-side device is an STM32L476RG low-power MCU; the image acquisition module includes an OV2710 sensor and an infrared fill light; the barcode acquisition module is a Honeywell N6603 barcode scanner; the first power management chip is a TI TPS61021 chip; and the first communication and power interface is a TE 1794178-1 industrial-grade Type-C interface.

[0007] Preferably, the second processor of the edge device includes an Intel Core i5-1340P CPU computing unit and an NVIDIA Jetson AGX Orin Nano GPU computing unit; the memory includes LPDDR5 RAM and SSD external storage; the second and third communication and power interfaces are both Type-C interfaces; the Wi-Fi module is an Intel AX210 module; the battery is a 10000mAh removable lithium battery; the speaker module is a KnowlesSPH0641LM4H-1 loudspeaker; and the second power management chip is an ST STUSB4500 PD controller.

[0008] Preferably, the local cloud-side equipment includes a Huawei Kunpeng 920 dual-machine hot standby application server, an NVIDIA DGX Station A100 GPU computing server, a Huawei OceanStor 5500 V5 encrypted storage server, a Huawei S5720 intranet switch, a Huawei USG6300 firewall, and an IBM TS2270 backup device.

[0009] Preferably, the encrypted storage server of the local cloud device adopts RAID 5+1 redundant storage and off-site disaster recovery backup, and the data storage encryption algorithm is AES-256.

[0010] Preferably, the edge device has a built-in patient data cache and lightweight model, which supports independent verification operations when the network is down, and automatically retransmits the data after the network is restored.

[0011] This invention also provides a wearable face recognition infusion verification method based on end-edge-cloud collaboration and local cloud deployment. This method is implemented using the aforementioned device and includes: S100: Preprocessing stage: The local cloud-side device interfaces with the hospital's HIS / LIS system to synchronize patient data and infusion plans, trains a high-precision face model and distills it into a lightweight model, and pushes it to the edge-side device cache; the edge-side device completes the allocation and configuration of computing resources for its CPU computing unit and GPU computing unit, and the edge-side device connects to the hospital's wireless access point through the Wi-Fi module and establishes an encrypted communication link; S200: Real-time verification phase: S210: The second communication and power interface of the edge device is electrically connected to the first communication and power interface of the end device via a line, for supplying power to the wearable smart information collection glasses on the end and establishing a data transmission link. The CPU computing unit of the edge device initializes its speaker module. S220: The end-side device collects the patient's face image, wristband barcode, and infusion label barcode, and uploads them to the edge-side device via the first communication and power interface. The edge-side device completes data preprocessing through its CPU computing unit. S230: The CPU computing unit of the edge device pushes the preprocessed data to its GPU computing unit, which runs the verification algorithm and verifies the barcode consistency in parallel through the CPU computing unit, thus completing the collaborative reasoning of the two computing units. S240: The edge device pushes the verification result to the end device for display, and at the same time its CPU computing unit drives the speaker module to emit corresponding voice prompts to realize visual-audio dual-modal feedback. S250: The edge device incrementally synchronizes and verifies records with the local cloud device through the communication link established by the Wi-Fi module, automatically caches data in weak network environments, and uploads data in batches after the network is restored; S300: Model Iteration and Data Backup Phase: The local cloud-side device iterates and optimizes the model based on the data fed back by the edge-side device and pushes updates. The CPU computing unit of the edge-side device receives updates through the Wi-Fi module during idle periods. At the same time, the local cloud-side device completes local data backup and off-site disaster recovery.

[0012] Preferably, in step S230, the verification algorithm run by the GPU computing unit is YOLOv8-tiny face detection + InsightFace-Buffalo-M feature extraction + PCA-ANN comparison algorithm.

[0013] Compared with existing technologies, the beneficial effects of this invention are as follows: This wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment is suitable for rapid and accurate verification of clinical patient identity and infusion information, and has the following advantages: 1. Ultimate optimized wearable experience: The end-side battery-free design is lightweight and comfortable to wear; the single Type-C cable simplifies operation, and the industrial-grade interface has high resistance to plugging and unplugging cycles, adapting to mobile ward scenarios; 2. Data security and compliance: Local cloud deployment + hospital Wi-Fi 6 encrypted transmission ensures that medical data does not leave the hospital, meeting the requirements of Level 3 Information Security Protection and localization of medical data, significantly reducing the risk of data leakage; 3. Highly efficient and accurate verification: High face recognition rate in obscured scenarios, and low overall verification latency. 4. High-speed Wi-Fi 6 wireless communication (peak rate ≥ 2.4Gbps), automatic caching in weak networks, usability during network outages, and automatic synchronization upon network connection; long battery life on the edge side to meet the needs of all-day clinical work; 5. Improved portability and ease of operation: elimination of gigabit Ethernet interface, simplified edge-side structure, and reduced weight; dual computing unit load balancing, enhanced multi-task concurrent processing capability; audio + visual dual-modal feedback, reducing the operational burden on medical staff; 6. Strong scalability: supports multi-terminal access (≤50 edge terminals + ≤150 glasses), and can be expanded to include functions such as drug allergy warning and infusion rate monitoring, adapting to the needs of hospitals of different sizes. Attached Figure Description

[0014] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an architecture diagram of an embodiment of the wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to the present invention.

[0015] Figure 2 This is a schematic diagram of the end-side device in one embodiment of the wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment of the present invention.

[0016] Figure 3 This is a schematic diagram of the edge device in one embodiment of the wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] In one embodiment, a wearable facial recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment is provided, such as... Figure 1 As shown, the wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment includes an end-side device 100, an edge-side device 200, and a local cloud-side device 300. The end-side device 100 is a wearable smart information collection glasses, the edge-side device 200 is a portable edge computing terminal, and the local cloud-side device 300 is an in-hospital server cluster.

[0019] Combination Figure 2 As shown, the wearable smart information collection glasses include a first processor 110, an image acquisition module 120, a barcode acquisition module, a diffraction grating display module 130, a first power management chip 140, and a first communication and power interface. The first processor 110, image acquisition module 120, barcode acquisition module, first power management chip 140, and first communication and power interface are respectively disposed in the frame of the wearable smart information collection glasses. The diffraction grating display module 130 is disposed in the lens. The barcode acquisition module is integrated in the frame area where the image acquisition module 120 is located. The first communication and power interface is electrically connected to... The first power management chip and the first processor are connected to each other to provide power to the first power management chip and the first processor. The first processor, the image acquisition module, the barcode acquisition module, and the diffraction grating display module are electrically connected to the first power management chip, which provides power to the first processor, the image acquisition module, the barcode acquisition module, and the diffraction grating display module through the first power management chip. The image acquisition module, the barcode acquisition module, and the diffraction grating display module are also electrically connected to the first processor, so that the first processor can transmit data or control signals to the image acquisition module, the barcode acquisition module, and the diffraction grating display module, respectively.

[0020] In this embodiment, the first processor of the edge device is a low-power MCU (STM32L476RG); the image acquisition module includes an OV2710 sensor and an infrared fill light; the barcode acquisition module is a Honeywell N6603 barcode scanner; the first power management chip is a TI TPS61021 chip; and the first communication and power interface is a TE 1794178-1 industrial-grade Type-C interface. The edge device 100 has no built-in battery; all operating power is provided by the edge device 200 through the Type-C interface. The edge device 100 only performs data acquisition and result display functions and has no local computing logic, thus meeting the requirements for lightweight design (weight ≤ 45g).

[0021] like Figure 3 As shown, the portable edge computing terminal of the edge device 200 includes a second processor, a memory 230, a second communication and power interface 280, a third communication and power interface 290, a Wi-Fi module 260, a battery 250, a speaker module 270, and a second power management chip 240. The battery 250 is electrically connected to the second power management chip 270. The second processor, the second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second power management chip for power supply. The second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second processor.

[0022] Combination Figure 2 As shown, the second communication and power interface 280 of the edge device 200 is used to be electrically connected to the first communication and power interface via line 150 to realize power supply and communication.

[0023] Combination Figure 3As shown, in this embodiment, the second processor of the edge device 200 includes a CPU computing unit 210 (Intel Core i5-1340P) and a GPU computing unit 220 (NVIDIA Jetson AGX Orin Nano); the memory includes LPDDR5 RAM (8G) and SSD external storage (256G); the second and third communication and power interfaces are both Type-C interfaces; the Wi-Fi module is an Intel AX210 module; the battery is a 10000mAh removable lithium battery; the speaker module is a Knowles SPH0641LM4H-1 loudspeaker; and the second power management chip is an ST STUSB4500 PD controller. In addition, the edge device 200 has a built-in patient data cache and lightweight model to support independent verification operations in the event of a network outage, and automatically re-transmits data after the network is restored.

[0024] The edge device 200 has the following features: (1) Dual computing unit collaboration: The CPU is responsible for concurrent tasks such as data preprocessing, device management, Wi-Fi communication, and audio driving, while the GPU focuses on face inference calculation, and the load distribution improves the overall processing efficiency; (2) Dual Type-C interface design: One channel realizes "reverse power supply + USB 3.2 Gen1 data transmission" (dedicated to end-edge), and the other channel is used for self-charging and expansion (supports external display and data storage devices); (3) Wireless communication optimization: The Gigabit Ethernet RJ45 interface is canceled, and the Wi-Fi 6 module is used to realize the internal communication of the hospital. High-speed communication (peak rate ≥ 2.4Gbps), supports automatic reconnection and signal strength detection, and is suitable for complex network environments in wards; (4) Audio feedback: The speaker module supports voice prompts for multiple scenarios (verification pass / fail, power supply abnormality, network interruption), and the volume is adjustable (30-80dB), which is suitable for noisy environments in wards; (5) Local caching and offline work: Built-in localized inference algorithm and patient data caching (≤1000 people), supporting offline independent work; (6) The edge side weighs about 520g and has a volume of about 145mm×95mm×40mm, which is convenient for nursing staff to carry with them.

[0025] The local cloud-side device 300's in-hospital server cluster is deployed within the hospital's intranet, physically isolated from the public cloud. Edge-side devices connect to the local cloud-side device via in-hospital wireless access points based on Wi-Fi wireless network signals, supporting concurrent access from multiple terminals. This enables patient data storage, facial model training and distillation, device management, log auditing, and HIS / LIS system integration. In this embodiment, the local cloud-side device includes a Huawei TaiShan Kunpeng 920 dual-machine hot standby application server, an NVIDIA DGX Station A100 GPU computing server, a Huawei OceanStor 5500 V5 encrypted storage server, a Huawei S5720 intranet switch, a Huawei USG6300 firewall, and an IBM TS2270 backup device. The encrypted storage server of the local cloud-side device adopts RAID 5+1 redundant storage and off-site disaster recovery backup, and the data storage encryption algorithm is AES-256.

[0026] Based on the above embodiments, it can be seen that the wearable infusion verification device based on end-edge-cloud collaboration and local cloud deployment takes the innovation of medical scene face recognition technology as the core and can achieve the following objectives: (1) The end device 100 has no built-in battery and is powered by the edge side through the Type-C interface, which reduces weight, improves wearing comfort, and ensures the convenience and stability of face recognition data collection; (2) The single Type-C interface of the end device 100 realizes power supply and high-speed data transmission at the same time, simplifies wiring, improves the reliability of mobile scenarios, and ensures low latency and integrity of face recognition image data transmission; (3) The local cloud device 300 deploys face recognition model training and data storage locally, and the patient's face biometric data is stored in the hospital network throughout the process, which meets the requirements of biometric data security protection and medical data localization compliance; 4) The edge-cloud collaborative mechanism and medical scene-specific face recognition algorithm can be optimized for low light and mask / glasses occlusion scenarios to improve the face recognition rate under occlusion scenarios, low latency verification, support offline face recognition and verification on the edge side, and ensure the continuity of long-term offline work; (5) The edge side device is equipped with an independent CPU computing unit and speaker module. The CPU is responsible for concurrent tasks such as data preprocessing and device management, while the GPU focuses on face recognition inference calculation. The load distribution improves the face recognition processing efficiency, and the audio feedback reduces the operating burden of medical staff. Combined with the face recognition results, it achieves rapid verification; (6) The edge side device interface design is simplified, the redundant wired network interface is eliminated, and wireless high-speed communication is achieved through Wi-Fi6, taking into account portability and communication stability, and ensuring efficient synchronization of face recognition model iteration data and verification records.

[0027] In one embodiment, a wearable face recognition infusion verification method based on end-edge-cloud collaboration and local cloud deployment is provided. This method is implemented using the device described in the preceding embodiment and includes: S100: Preprocessing stage: The local cloud-side device interfaces with the hospital's HIS / LIS system to synchronize patient data and infusion plans, trains a high-precision face model and distills it into a lightweight model, and pushes it to the edge-side device for caching; the edge-side device completes the allocation and configuration of computing resources for its CPU computing unit and GPU computing unit, and the edge-side device connects to the hospital's wireless access point through the Wi-Fi module and establishes an encrypted communication link.

[0028] In this preprocessing stage, the local cloud-side device interfaces with the hospital's HIS / LIS system via the HL7FHIR protocol to synchronize patient identity information (name, hospital number, facial image) and infusion plans (drug name, dosage, administration time), and stores them in an encrypted database (AES-256 encryption). The local cloud-side device trains a high-precision facial model based on the ResNet50 backbone network, generates a lightweight model through knowledge distillation technology, pushes it to all edge-side devices, and caches it. The edge-side device receives and caches the current department's patient data and the lightweight model, completes the allocation and configuration of CPU and GPU computing resources (CPU priority: data transmission > device management > audio driver > Wi-Fi communication; GPU priority: facial inference), and completes pairing with the Type-C interface of the smart glasses on the edge side (automatically negotiates PD power supply protocol and USB3.2 Gen1 data transmission protocol). The Wi-Fi 6 module of the edge-side device automatically connects to the designated wireless access point in the hospital, establishes an encrypted communication link (WPA3-Enterprise encryption), and enables the signal strength monitoring function (triggers reconnection when the signal strength is below -70dBm).

[0029] S200: Real-time verification phase: S210: The second communication and power interface of the edge device is electrically connected to the first communication and power interface of the end device via a line, for supplying power to the wearable smart information acquisition glasses on the end and establishing a data transmission link. The CPU computing unit of the edge device initializes its speaker module.

[0030] Step S210 is used to establish power supply and communication connections. The edge device outputs a stable 5V / 3A power supply to the smart glasses via a Type-C cable, and at the same time establishes a bidirectional data transmission link. The CPU initializes the speaker module, with the default volume set to 50dB. The Wi-Fi 6 module maintains the communication link with the local cloud side and provides real-time feedback on the network status.

[0031] S220: The end-side device collects the patient's face image, wristband barcode, and infusion label barcode, and uploads them to the edge-side device via the first communication and power interface. The edge-side device completes data preprocessing through its CPU computing unit.

[0032] Step S220 is used for data acquisition. Medical staff operate the glasses on the end side to trigger the image acquisition module to acquire the patient's face image (supports low light / occlusion scenarios), and the barcode acquisition module to scan the patient's wristband barcode and infusion label barcode. The acquired data is uploaded to the edge side in real time via the Type-C interface, and the CPU completes the data preprocessing (format conversion and noise reduction).

[0033] S230: The CPU computing unit of the edge device pushes the preprocessed data to its GPU computing unit, which runs the verification algorithm and verifies the barcode consistency in parallel through the CPU computing unit, thus completing the collaborative inference between the two computing units. In this step, the verification algorithm run by the GPU computing unit is YOLOv8-tiny face detection + InsightFace-Buffalo-M feature extraction + PCA-ANN comparison algorithm.

[0034] Step S230 is used for collaborative inference between the two computing units on the edge side. The CPU pushes the preprocessed data to the GPU, and the GPU runs the collaborative algorithm (YOLOv8-tiny face detection + InsightFace-Buffalo-M feature extraction + PCA-ANN alignment). At the same time, the CPU verifies the consistency between the wristband barcode, the infusion label barcode and the patient's hospital number in parallel.

[0035] S240: The edge device pushes the verification result to the end device for display, and at the same time, its CPU computing unit drives the speaker module to issue corresponding voice prompts, realizing visual-audio dual-modal feedback.

[0036] Step S240 is used to provide result feedback (visual + audio dual-modality). The edge side pushes the verification result (successful / failed match + discrepancies) to the end-side glasses via the Type-C interface. The name and verification result (green "pass" / red "mismatch") are displayed in the diffraction grating display module. The CPU drives the megaphone module to issue corresponding voice prompts: "Verification passed" (0.5 seconds) for successful matching, "Information mismatch, please verify" (1 second) for failed matching, "Power failure, please check connection" for power supply abnormality, and "Offline mode, data will be synchronized later" for network interruption. S250: Edge devices incrementally synchronize and verify records with local cloud devices through a communication link established by the Wi-Fi module. Data is automatically cached in weak network environments and uploaded in batches after the network is restored.

[0037] Step S250 is used for data synchronization. The CPU incrementally synchronizes and verifies records with the local cloud via the Wi-Fi 6 link (only transmitting newly added / changed data), with a synchronization latency of ≤50ms to avoid bandwidth occupation. When the network signal is weak, the synchronized data is automatically cached and uploaded in batches after the signal is restored.

[0038] S300: Model Iteration and Data Backup Phase: The local cloud-side device iterates and optimizes the model based on the data feedback from the edge-side device and pushes updates. The CPU computing unit of the edge-side device receives updates through the Wi-Fi module during idle periods. At the same time, the local cloud-side device completes local data backup and off-site disaster recovery.

[0039] The local cloud side regularly (quarterly) collects misidentified cases and verification data uploaded by the edge side, updates the training dataset, retrains the high-precision face model and distills it into a lightweight version, and pushes it to the edge terminal (the CPU receives and updates the model during idle periods without affecting GPU inference and audio feedback); the local cloud side automatically backs up data daily (RAID 5+1 redundant storage) and completes off-site disaster recovery backup weekly to ensure no data loss; the edge CPU monitors GPU load, loudspeaker working status and Wi-Fi 6 network quality in real time, and alarms are simultaneously triggered through the display module and loudspeaker when anomalies occur.

[0040] The method has the following characteristics: (1) A collaborative reasoning algorithm optimized for medical scenarios, combined with YOLOv8-tiny and the quantized InsightFace model, optimized for mask occlusion and low light environment, with high recognition rate in occlusion scenarios and low latency in end-edge-cloud full-process verification; (2) Multi-level offline redundancy mechanism: The edge side caches ≥1000 patient data and lightweight models, can work independently for ≥15 days in the state of network disconnection, and automatically retransmits data after network recovery to ensure clinical continuity; (3) Edge side dual computing unit collaborative architecture: Independent CPU and GPU realize load balancing, the CPU is responsible for concurrent task processing, and the GPU focuses on inference calculation, which improves processing efficiency in multi-task scenarios; the redundant wired interface is eliminated, simplifying the design and improving portability; (4) Visual-audio dual-modal feedback: The loudspeaker module provides multi-scenario voice prompts, reducing the dependence of medical staff on the display interface, and improving both operation convenience and verification accuracy.

[0041] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0042] 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. A wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment, characterized in that, The device includes: The device is a wearable smart information collection glasses, which includes a first processor, an image acquisition module, a barcode acquisition module, a diffraction grating display module, a first power management chip, and a first communication and power interface. The first communication and power interface is electrically connected to the first power management chip and the first processor. The first processor, the image acquisition module, the barcode acquisition module, and the diffraction grating display module are each electrically connected to the first power management chip, and the image acquisition module, the barcode acquisition module, and the diffraction grating display module are each electrically connected to the first processor. An edge-side device, specifically a portable edge computing terminal, includes a second processor, a memory, a second communication and power interface, a third communication and power interface, a Wi-Fi module, a battery, a speaker module, and a second power management chip. The battery is electrically connected to the second power management chip. The second processor, the second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second power management chip, and the second communication and power interface, the third communication and power interface, the memory, the Wi-Fi module, and the speaker module are all electrically connected to the second processor. The second communication and power interface is used to electrically connect to the first communication and power interface via a line. The local cloud-side device is an in-hospital server cluster deployed on the hospital's intranet and physically isolated from the public cloud. The edge-side device connects to the local cloud-side device through an in-hospital wireless access point based on Wi-Fi wireless network signals, supports concurrent access from multiple terminals, and enables patient data storage, face model training and distillation, device management, log auditing, and HIS / LIS system integration.

2. The wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to claim 1, characterized in that: The first processor of the end-side device is an STM32L476RG low-power MCU; the image acquisition module includes an OV2710 sensor and an infrared fill light; the barcode acquisition module is a Honeywell N6603 barcode scanner; the first power management chip is a TI TPS61021 chip; and the first communication and power interface is a TE1794178-1 industrial-grade Type-C interface.

3. The wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to claim 2, characterized in that: The second processor of the edge device includes an Intel Core i5-1340P CPU computing unit and an NVIDIA Jetson AGX Orin Nano GPU computing unit; the memory includes LPDDR5 RAM and SSD external storage; the second and third communication and power interfaces are both Type-C interfaces; the Wi-Fi module is an Intel AX210 module; the battery is a 10000mAh removable lithium battery; the speaker module is a Knowles SPH0641LM4H-1 loudspeaker; and the second power management chip is an ST STUSB4500 PD controller.

4. The wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to claim 3, characterized in that: The local cloud-side equipment includes Huawei TaiShan Kunpeng 920 dual-machine hot standby application server, NVIDIA DGX Station A100 GPU computing server, Huawei OceanStor 5500 V5 encrypted storage server, Huawei S5720 intranet switch, Huawei USG6300 firewall, and IBM TS2270 backup device.

5. The wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to claim 4, characterized in that: The encrypted storage server of the local cloud-side device adopts RAID 5+1 redundant storage and off-site disaster recovery backup, and the data storage encryption algorithm is AES-256.

6. The wearable face recognition infusion verification device based on end-edge-cloud collaboration and local cloud deployment according to claim 1, characterized in that: The edge device has a built-in patient data cache and lightweight model, which supports independent verification operations when the network is down, and automatically retransmits the data after the network is restored.

7. A wearable face recognition infusion verification method based on end-edge-cloud collaboration and local cloud deployment, characterized in that, This method is implemented based on the apparatus of claim 4, and the method includes: S100: Preprocessing stage: The local cloud-side device interfaces with the hospital's HIS / LIS system to synchronize patient data and infusion plans, trains a high-precision face model and distills it into a lightweight model, and pushes it to the edge-side device cache; the edge-side device completes the allocation and configuration of computing resources for its CPU computing unit and GPU computing unit, and the edge-side device connects to the hospital's wireless access point through the Wi-Fi module and establishes an encrypted communication link; S200: Real-time verification phase: S210: The second communication and power interface of the edge device is electrically connected to the first communication and power interface of the end device via a line, for supplying power to the wearable smart information collection glasses on the end and establishing a data transmission link. The CPU computing unit of the edge device initializes its speaker module. S220: The end-side device collects the patient's face image, wristband barcode, and infusion label barcode, and uploads them to the edge-side device via the first communication and power interface. The edge-side device completes data preprocessing through its CPU computing unit. S230: The CPU computing unit of the edge device pushes the preprocessed data to its GPU computing unit, which runs the verification algorithm and verifies the barcode consistency in parallel through the CPU computing unit, thus completing the collaborative reasoning of the two computing units. S240: The edge device pushes the verification result to the end device for display, and at the same time its CPU computing unit drives the speaker module to emit corresponding voice prompts to realize visual-audio dual-modal feedback. S250: The edge device incrementally synchronizes and verifies records with the local cloud device through the communication link established by the Wi-Fi module, automatically caches data in weak network environments, and uploads data in batches after the network is restored; S300: Model Iteration and Data Backup Phase: The local cloud-side device iterates and optimizes the model based on the data fed back by the edge-side device and pushes updates. The CPU computing unit of the edge-side device receives updates through the Wi-Fi module during idle periods. At the same time, the local cloud-side device completes local data backup and off-site disaster recovery.

8. The wearable face recognition infusion verification method based on end-edge-cloud collaboration and local cloud deployment according to claim 7, characterized in that: In step S230, the verification algorithm run by the GPU computing unit is YOLOv8-tiny face detection + InsightFace-Buffalo-M feature extraction + PCA-ANN comparison algorithm.

Images