A deep learning-based electronic sphygmomanometer verification recognition method and system

Abstract

The application discloses a kind of based on deep learning's electronic sphygmomanometer verification identification method and system, it is related to artificial intelligence technical field, comprising: S1, utilizes electronic sphygmomanometer image dataset to construct deep learning identification model;S2, based on electronic sphygmomanometer real-time image data utilizes the deep learning identification model to obtain the identification result of electronic sphygmomanometer real-time image data;S3, according to the identification result of the electronic sphygmomanometer real-time image data obtains electronic sphygmomanometer verification identification result.The application replaces the process of artificial reading, counting, editing original data and detection report in the verification process of electronic sphygmomanometer, and solves the problem of low accuracy rate when traditional machine learning method processes different categories of electronic sphygmomanometer readings, improves verification efficiency and identification accuracy.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for calibrating and identifying electronic blood pressure monitors based on deep learning. Background Technology

[0002] Currently, the calibration of traditional electronic blood pressure monitors still relies on manual operation for reading the data. This is not only time-consuming and labor-intensive, but also prone to errors due to operator fatigue or distraction, thus affecting the efficiency and accuracy of the test results. Faced with the ever-increasing calibration demands and the diversification of equipment types, traditional manual reading methods can no longer meet the requirements of efficient and accurate metrological testing.

[0003] Currently, the following main challenges remain when using machine learning methods to solve the problem of electronic blood pressure monitor readings:

[0004] (1) Difficulty in recognizing digital display units: Electronic blood pressure monitors usually use seven-segment digital tubes to display numbers. How to accurately distinguish the seven digital tubes belonging to the same number and correctly identify the number based on their lighting combination is one of the technical bottlenecks in traditional methods. However, the recognition accuracy of existing solutions is low and it is difficult to meet the actual application requirements.

[0005] (2) The diversity of devices leads to insufficient recognition adaptability: There are many brands and models of electronic blood pressure monitors, and the display screen designs vary greatly. Some models may contain a lot of information that is not related to measurement, such as time, patient number, wearing status, etc. These interfering information significantly increase the difficulty of identifying the effective data area, and traditional methods are difficult to adapt to complex and diverse scenarios.

[0006] To address the aforementioned challenges, artificial intelligence technologies, particularly deep learning and machine vision, offer a breakthrough for the intelligent development of electronic blood pressure monitor calibration. Combining the successful experience of machine vision algorithms in these fields, their application to data recognition and recording in electronic blood pressure monitor calibration is inevitable, as is the generation of original records and calibration certificates in metrological testing activities.

[0007] Therefore, there is an urgent need for a deep learning-based method and system for calibrating and identifying electronic blood pressure monitors to address the shortcomings of existing technologies. Summary of the Invention

[0008] The purpose of this invention is to propose a deep learning-based method and system for the verification and identification of electronic blood pressure monitors, replacing the manual operation process in the verification of electronic blood pressure monitors, and solving the problem of low accuracy of traditional machine learning methods when processing readings of different types of electronic blood pressure monitors.

[0009] On the one hand, to achieve the above objectives, the present invention provides a deep learning-based method for calibrating and identifying electronic blood pressure monitors, comprising the following steps:

[0010] S1. Construct a deep learning recognition model using an electronic blood pressure monitor image dataset;

[0011] S2. Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor.

[0012] S3. Obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

[0013] Optionally, the deep learning recognition model is implemented based on the YOLOv11 object detection network.

[0014] Optionally, building a deep learning recognition model using an electronic blood pressure monitor image dataset includes:

[0015] S1-1. Construct an image dataset for electronic blood pressure monitors;

[0016] S1-2. Use the image dataset of the electronic blood pressure monitor to train and construct a deep learning coarse recognition model;

[0017] S1-3. Based on the deep learning coarse recognition model, process the image dataset of the electronic blood pressure monitor to obtain a digit recognition dataset;

[0018] S1-4. Train and construct a deep learning fine recognition model based on the digit recognition dataset;

[0019] S1-5. Obtain the deep learning coarse recognition model and the deep learning fine recognition model as deep learning recognition models;

[0020] The image dataset of the electronic blood pressure monitor contains image data of the region of interest displayed by the electronic blood pressure monitor, which includes the systolic pressure region and the diastolic pressure region.

[0021] Optionally, training a deep learning coarse recognition model using the image dataset from the electronic blood pressure monitor includes:

[0022] S1-2-1. Perform data augmentation processing on the image dataset of the electronic blood pressure monitor to obtain an augmented dataset of the image dataset;

[0023] S1-2-2, Divide the image dataset into augmented datasets to obtain augmented data training set and augmented data validation set;

[0024] S1-2-3. Using the enhanced data training set as input and the corresponding image data of the enhanced data training set as output, train an initial deep learning coarse recognition model based on the YOLOv11 object detection network.

[0025] S1-2-4. Based on the enhanced data validation set, input the initial deep learning coarse recognition model to obtain the image data corresponding to the enhanced data validation set;

[0026] S1-2-5. Determine whether the image data corresponding to the augmented data validation set are all image data of the region of interest. If so, obtain the initial deep learning coarse recognition model as the deep learning coarse recognition model. Otherwise, update the augmented data training set using the augmented data validation set and return to S1-2-3.

[0027] Optionally, obtaining the recognition result of the real-time image data of the electronic blood pressure monitor using the deep learning recognition model based on the real-time image data of the electronic blood pressure monitor includes:

[0028] Acquire real-time image data from an electronic blood pressure monitor;

[0029] Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the region of interest (ROI) digits of the real-time image data of the electronic blood pressure monitor.

[0030] The recognition result of the real-time image data of the electronic blood pressure monitor is obtained by arranging the digital items of the region of interest in the real-time image data.

[0031] Optionally, the region of interest (ROI) digits of the real-time image data of the electronic blood pressure monitor obtained using the deep learning recognition model based on the real-time image data include:

[0032] The coarse recognition result of the real-time image data of the electronic blood pressure monitor is obtained by inputting the deep learning coarse recognition model with the real-time image data of the electronic blood pressure monitor.

[0033] The region of interest in the real-time image data of the electronic blood pressure monitor is obtained by cropping the real-time image data of the electronic blood pressure monitor using the coarse recognition results.

[0034] The region of interest (ROI) of the real-time image data of the electronic blood pressure monitor is input into the deep learning fine recognition model to obtain the ROI digit item of the real-time image data of the electronic blood pressure monitor.

[0035] Optionally, obtaining the electronic blood pressure monitor verification and identification result based on the recognition result of the real-time image data of the electronic blood pressure monitor includes:

[0036] The recognition results of the real-time image data from the electronic blood pressure monitor are processed for display to obtain the display recognition results;

[0037] The identification results are displayed and verified using the blood pressure monitor calibration procedure to obtain the electronic blood pressure monitor calibration identification results.

[0038] On the other hand, to achieve the above objectives, the present invention provides an electronic blood pressure monitor verification and identification system based on deep learning, including: a model building module, a real-time identification module, and a result acquisition module;

[0039] The model building module is used to build a deep learning recognition model using an electronic blood pressure monitor image dataset;

[0040] The real-time recognition module is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor using the deep learning recognition model based on the real-time image data of the electronic blood pressure monitor.

[0041] The result acquisition module is used to obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

[0042] Optionally, the system further includes a model optimization module, which is used to optimize the deep learning recognition model using the electronic blood pressure monitor image dataset and real-time acquired electronic blood pressure monitor image data.

[0043] Compared with the closest existing technology, the present invention has the following advantages:

[0044] This invention utilizes deep learning to process data, adapting to different brands and models of electronic blood pressure monitors. It automates readings, freeing up manual labor and improving verification efficiency. The invention employs a coarse-to-fine recognition process, first identifying the ROI (Region of Interest) for verification, then recognizing the numbers within that region. This effectively eliminates various interfering information in the captured image, improving recognition accuracy. Furthermore, this invention utilizes self-built datasets of electronic blood pressure monitor images and seven-segment digital display images for electronic blood pressure monitor recognition, making the model more adaptable to real-world scenarios. In summary, this invention replaces the manual reading, counting, editing of raw data, and verification certificates in the electronic blood pressure monitor verification process. It also solves the problem of low accuracy when traditional machine learning methods process readings from different types of electronic blood pressure monitors, improving both verification efficiency and recognition accuracy. Attached Figure Description

[0045] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0046] Figure 1 This is a flowchart illustrating a deep learning-based method for calibrating and identifying electronic blood pressure monitors according to an embodiment of the present invention.

[0047] Figure 2This is a flowchart of the model training proposed in an embodiment of the present invention;

[0048] Figure 3 This is a flowchart illustrating the verification and identification process for an electronic blood pressure monitor as proposed in an embodiment of the present invention.

[0049] Figure 4 This is a schematic diagram of the structure of an electronic blood pressure monitor verification and identification system based on deep learning, according to an embodiment of the present invention. Detailed Implementation

[0050] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions in the embodiments of this invention will be clearly and completely described below with reference to specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0051] The terminology used in the embodiments section of this invention is for the purpose of explaining specific embodiments of the invention only, and is not intended to limit the invention.

[0052] This invention aims to provide a deep learning-based method and system for the verification and identification of electronic blood pressure monitors, focusing on solving the core challenges of digital display unit identification and effective data area extraction, thereby achieving an efficient and accurate automated verification process.

[0053] like Figure 1 As shown, this embodiment of the invention provides a deep learning-based method for calibrating and identifying electronic blood pressure monitors, including:

[0054] S1. Construct a deep learning recognition model using an electronic blood pressure monitor image dataset;

[0055] S2. Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor.

[0056] S3. Obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

[0057] Specifically, such as Figure 2As shown, an image dataset of electronic blood pressure monitors is constructed. A deep learning coarse recognition model is trained using this dataset. The model's function is to accurately identify the region of interest (ROI) for electronic blood pressure monitor verification within the image. The trained coarse recognition model is then used to batch process the current electronic blood pressure monitor image dataset, extracting images containing the digits of the seven-segment display to form a digit recognition dataset. A deep learning fine recognition model is trained using this dataset. This model's function is to accurately determine the type, category, and location information of all digits appearing in the image. Images of the target electronic blood pressure monitor are acquired through data collection. The acquired images are sequentially processed by the deep learning coarse and fine recognition models. The recognition system displays the final recognition results on the user interface for verification. After recording all data according to the requirements of the blood pressure monitor verification procedure JJG 692-2010, all recorded data can be imported into the original data record template through the testing system to issue a verification certificate.

[0058] In addition, this embodiment also includes model optimization. During use, the image data collected from each electronic blood pressure monitor is recorded and combined with the existing dataset, enriching the dataset's data volume. Furthermore, the newly collected data is closer to everyday usage scenarios, resulting in a higher accuracy rate for the model optimized with the new dataset in daily use cases.

[0059] There are two scenarios where ROI regions may contain errors: one is an incorrect region, and the other is a correct but incomplete region (especially when the first digit is 1, making it easy to exclude the digit 1). Errors in ROI regions almost always fall into the second category; the first scenario does not occur.

[0060] To address the above issues, the optimizations during training include: (1) improving the quality of the dataset to ensure that the annotations tightly and completely surround all numbers; (2) employing data augmentation, using translation, scaling, and Mosaic data augmentation strategies to simulate different positions, sizes, and distances of the target region in the image; (3) focusing on samples with incorrect identification or low confidence during training and increasing the frequency of these samples in subsequent training cycles; (4) fine-tuning the loss function: considering the actual situation, the bounding box regression loss in the total loss function needs special attention, so after multiple training attempts, the weight of the bounding box regression loss was set to 8.5.

[0061] To address the issue of incomplete regions, the algorithm incorporates additional optimizations. After coarsely identifying the ROI, it expands the original identification result by an additional 10% pixels in both the left and right directions before cropping. This method fills in potentially missing parts without incorporating interference information. Vertical expansion is omitted because there are no missing parts in the vertical direction. The deep learning coarse identification model returns the coordinates of the target box [x1, y1, x2, y2], which can be used to calculate the coordinates of the cropped region. x1 - [1.1*(x2 - x1)] represents the x-coordinate of the left-side point of the cropped box, x2 + [1.1*(x2 - x1)] represents the x-coordinate of the right-side point of the cropped box, and the y-coordinate remains unchanged.

[0062] The core objective of coarse recognition is to ensure that the bounding boxes in the SYS and DIA regions accurately and completely contain all digits. During the data preparation phase, images of various electronic blood pressure monitors are collected (especially those taken in real-world use scenarios to validate model performance).

[0063] In terms of data augmentation strategy, the rotation angle degrees=10, translation amplitude=0.2, and shear intensity=10 are set to help the model adapt to the diversity brought about by shooting angle, target position, and slight deformation. At the same time, the model will benefit from the rich scene combinations brought by Mosaic data augmentation, while this augmentation is turned off in the last 800 rounds of training (parameters obtained after multiple training sessions).

[0064] During the model training phase, the model learns on batches of augmented training data in each round. After forward propagation, the model calculates the loss based on the prediction results and the ground truth annotations. For practical application scenarios, the weight of the bounding box regression loss is increased to 8.5, focusing on solving the problem of whether the recognition box can contain all numbers.

[0065] The validation phase is performed after each training cycle. The focus is on the accuracy of the bounding boxes. Visual inspection confirms that all digits within the SYS and DIA regions are completely included. Simultaneously, mAP50, mAP50-95, precision, recall, and the performance of each loss component on the validation set are recorded. During this process, any incomplete recognition, positional shifts, or misidentifications are carefully recorded, and analysis is conducted to determine which specific brands, lighting conditions, or angles are more prone to errors.

[0066] Finally, there is the adjustment and iteration phase. Based on the feedback from the validation phase, the composition of the dataset is adjusted, and the frequency of these samples is increased in subsequent training cycles.

[0067] This invention provides a deep learning-based method and system for the verification and identification of electronic blood pressure monitors. It utilizes an electronic blood pressure monitor image dataset to construct a deep learning recognition model. The model first coarsely identifies the coordinates of the systolic blood pressure (SYS) and diastolic blood pressure (DIA) regions in real-time images and then expands and extracts the Region of Interest (ROI). Next, it finely identifies the numerical categories and coordinates to combine them into numerical values. Finally, it automatically generates original records and verification certificates that conform to regulations. This achieves full automation from image acquisition to certificate generation, solving the problems of low efficiency and insufficient accuracy of traditional manual readings and algorithms. It improves verification efficiency and accuracy and promotes intelligent metrological testing.

[0068] Furthermore, the deep learning recognition model is implemented based on the YOLOv11 object detection network, specifically:

[0069] Both the deep learning coarse recognition model and the deep learning fine recognition model are implemented based on the YOLOv11 object detection network.

[0070] As one possible implementation, in the above embodiments, step S1 may specifically include the following steps:

[0071] S1-1. Construct an image dataset for electronic blood pressure monitors;

[0072] S1-2. Use the image dataset of the electronic blood pressure monitor to train and construct a deep learning coarse recognition model;

[0073] S1-3. Based on the deep learning coarse recognition model, process the image dataset of the electronic blood pressure monitor to obtain a digit recognition dataset;

[0074] S1-4. Train and construct a deep learning fine recognition model based on the digit recognition dataset;

[0075] S1-5. Obtain the deep learning coarse recognition model and the deep learning fine recognition model as deep learning recognition models.

[0076] In this embodiment, the image dataset of the electronic blood pressure monitor contains images of the ROI region displayed by the electronic blood pressure monitor. During the initial construction, relevant images from the Internet were mainly collected.

[0077] As one possible implementation, in the above embodiments, step S1-2 may specifically include the following steps:

[0078] S1-2-1. Perform data augmentation processing on the image dataset of the electronic blood pressure monitor to obtain an augmented dataset of the image dataset;

[0079] S1-2-2, Divide the image dataset into augmented datasets to obtain augmented data training set and augmented data validation set;

[0080] S1-2-3. Using the enhanced data training set as input and the corresponding image data of the enhanced data training set as output, train an initial deep learning coarse recognition model based on the YOLOv11 object detection network.

[0081] S1-2-4. Based on the enhanced data validation set, input the initial deep learning coarse recognition model to obtain the image data corresponding to the enhanced data validation set;

[0082] S1-2-5. Determine whether the image data corresponding to the augmented data validation set are all image data of the region of interest. If so, obtain the initial deep learning coarse recognition model as the deep learning coarse recognition model. Otherwise, update the augmented data training set using the augmented data validation set and return to S1-2-3.

[0083] As one possible implementation, in the above embodiments, step S2 may specifically include the following steps:

[0084] S2-1. Acquire real-time image data from the electronic blood pressure monitor;

[0085] S2-2. Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the region of interest (ROI) digital items of the real-time image data of the electronic blood pressure monitor.

[0086] S2-3. Arrange the digital items of the region of interest in the real-time image data of the electronic blood pressure monitor to obtain the recognition result of the real-time image data of the electronic blood pressure monitor.

[0087] like Figure 3 As shown, the display of the electronic blood pressure monitor is acquired using a computer camera or other image acquisition module and transmitted to the system; a deep learning coarse recognition model is used to identify the acquired image data, that is, according to the verification procedure JJG 692-2010, the location and coordinate information of the ROI region used for verification are identified; the acquired image data is processed according to the coordinate information, and each ROI region is extracted from the original image data; a deep learning fine recognition model is used to identify the category and coordinate information of the digits in each ROI region; and the identified digits are restored to specific numbers according to the coordinate information of the identified digits.

[0088] For electronic blood pressure monitor measurements, there are two test items. The data recorded for both items correspond to the SYS and DIA positions displayed on the electronic blood pressure monitor. Therefore, the coarsely identified ROI region is the area displayed as SYS and DIA on the electronic blood pressure monitor. SYS and DIA are the classifications output by the coarsely identified object detection network.

[0089] The predicted return includes the bounding box coordinates of the target, in the form [x1, y1, x2, y2], where (x1, y1) is the coordinate of the top-left corner of the bounding box, and (x2, y2) is the coordinate of the bottom-right corner. Based on these coordinates, a coarsely identified Region of Interest (ROI) is extracted from the original image and used as input to the deep learning fine-grained recognition model. Similarly, in the deep learning fine-grained recognition model, the center coordinates of each identified digit are calculated based on [x1, y1, x2, y2] for digit combination. For example, if the deep learning fine-grained recognition model identifies a 1 and a 2, the result needs to be synthesized based on the center coordinates to produce 21 instead of 12.

[0090] As one possible implementation, in the above embodiments, step S2-2 may specifically include the following steps:

[0091] S2-2-1. Input the real-time image data of the electronic blood pressure monitor into the deep learning coarse recognition model to obtain the coarse recognition result of the real-time image data of the electronic blood pressure monitor;

[0092] S2-2-2: Using the coarse recognition results of the real-time image data of the electronic blood pressure monitor, the region of interest of the real-time image data of the electronic blood pressure monitor is obtained by cropping the real-time image data of the electronic blood pressure monitor.

[0093] The results of the extracted content vary depending on the scenario. There are two test items for testing electronic blood pressure monitors: "static pressure indication error" and "blood pressure indication repeatability verification". The display area for recording the numbers for both different items is the area on the electronic blood pressure monitor that displays SYS and DIA.

[0094] For the "test of repeatability of blood pressure readings", the results to be identified are the numbers within the SYS and DIA regions. The content extracted at this time is the ROI region in the tested sample, which will vary depending on the shooting angle, distance, and model of the tested sample.

[0095] However, for the "static pressure indication error" item, different blood pressure monitors have different display methods, with three scenarios: the first is displaying it in the SYS area, the second is displaying it in the DIA area, and the third is displaying it in both the SYS and DIA areas. For this item, the benchmark for cropping is determined based on the confidence level of the deep learning coarse recognition model's results. If multiple targets are identified in a single coarse recognition, the area with the highest confidence level will be cropped, discarding the other targets. In this item, only one image is cropped from the coarse recognition.

[0096] S2-2-3. Based on the region of interest (ROI) of the real-time image data of the electronic blood pressure monitor, input the deep learning fine recognition model to obtain the ROI digit item of the real-time image data of the electronic blood pressure monitor.

[0097] As one possible implementation, in the above embodiments, step S3 may specifically include the following steps:

[0098] S3-1. Display the recognition results of the real-time image data of the electronic blood pressure monitor to obtain the display recognition results;

[0099] The digital recognition results within each ROI region are displayed on the user interface of the deep learning-based intelligent recognition system for electronic blood pressure monitor calibration.

[0100] S3-2. Based on the identification results, the identification results are verified using the blood pressure monitor verification procedure to obtain the electronic blood pressure monitor verification results;

[0101] According to the operational requirements of the blood pressure monitor verification procedure JJG 692-2010, after completing all verification items, the read data results are uniformly imported into the technical institution's original record template, automatically generating the original record. The metrological verification results are judged according to the requirements of verification procedure JJG692-2010, and the final verification certificate is generated, which can be directly uploaded to the technical institution's electronic certificate system. This embodiment, through linkage with the technical institution's electronic certificate system, can directly generate the original record used for issuing the verification certificate.

[0102] Considering the characteristics of metrology work and practical usage scenarios, it is impossible to achieve result recognition of electronic blood pressure monitors using traditional computer vision algorithms. This embodiment essentially completes result recognition in a two-stage process, from coarse to fine.

[0103] For electronic blood pressure monitors used in metrology, result recognition only requires identifying the numbers displayed in the SYS and DIA areas (heart rate, time, etc., are entirely useless data and interference). Even if traditional algorithms (including deep learning models with direct digit recognition capabilities) can accurately identify each number, subsequent digit filtering still lacks an efficient and accurate algorithm. The main reason is that the relative positions of non-ROIs and ROIs are not fixed, and there is no absolute discrimination logic.

[0104] Therefore, the aforementioned algorithmic challenges are circumvented through a two-stage approach: employing two completely different deep learning models to progressively complete the task of recognizing measurement results. First, the regions of SYS and DIA are located, and the coordinate information returned by the recognition is used to calculate the position of the entire recognition box, extracting the recognition result from the original image. Then, the extracted result is used in a deep learning fine-grained recognition model to identify the types of digits. Finally, based on the coordinate information of the digits in the fine-grained recognition, the center coordinates of each digit are calculated, and the result is determined based on the center coordinates. Furthermore, the coarse recognition can also help eliminate easily misleading icons on the display screen.

[0105] like Figure 4 As shown, this embodiment of the invention provides an electronic blood pressure monitor verification and identification system based on deep learning, including: a model building module, a real-time identification module, and a result acquisition module;

[0106] The model building module is used to build a deep learning recognition model using an electronic blood pressure monitor image dataset;

[0107] The model building module collects an image dataset of electronic blood pressure monitors, combines data augmentation strategies, trains a deep learning recognition model based on the YOLOv11 object detection network, and sets the bounding box regression loss weight to 8.5 to build a model that can accurately locate the SYS and DIA regions.

[0108] The real-time recognition module is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor using the deep learning recognition model based on the real-time image data of the electronic blood pressure monitor.

[0109] The real-time recognition module first acquires real-time image data of the electronic blood pressure monitor through the image acquisition unit, and then the model identifies the bounding box coordinates [x1, y1, x2, y2] of the SYS and DIA regions. It then expands the ROI region by 10% pixels in both the left and right directions. Next, it identifies the category and coordinates of the numbers in the ROI image, and calculates the specific values ​​by combining the coordinates of the center point of the numbers, thereby obtaining the recognition result of the real-time image data.

[0110] The result acquisition module is used to obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

[0111] The results acquisition module displays the identified blood pressure values ​​on the operation interface. In accordance with the requirements of JJG 692-2010 verification procedure, it automatically imports the data into the original record template to generate the original record. After completing the determination of the metrological verification results, it generates a verification certificate, which can be directly uploaded to the electronic certificate system of the technical institution.

[0112] Furthermore, the system also includes a model optimization module, which is used to optimize the deep learning recognition model using the electronic blood pressure monitor image dataset and real-time acquired electronic blood pressure monitor image data.

[0113] In addition, the system also has a model optimization module, which records the electronic blood pressure monitor image data collected by the real-time recognition module during use, combines it with the original dataset to form a new dataset, optimizes and trains the deep learning recognition model, and continuously improves the recognition accuracy of the model in daily scenarios.

[0114] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0115] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0116] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0117] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0118] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A deep learning-based method for calibrating and identifying electronic blood pressure monitors, characterized in that, include: S1. Construct a deep learning recognition model using an electronic blood pressure monitor image dataset; The deep learning recognition model is implemented based on the YOLOv11 object detection network. S1-1. Construct an image dataset for electronic blood pressure monitors; S1-2. Use the image dataset of the electronic blood pressure monitor to train and construct a deep learning coarse recognition model; S1-3. Based on the deep learning coarse recognition model, process the image dataset of the electronic blood pressure monitor to obtain a digit recognition dataset; S1-4. Train and construct a deep learning fine recognition model based on the digit recognition dataset; S1-5. Obtain the deep learning coarse recognition model and the deep learning fine recognition model as deep learning recognition models; The image dataset of the electronic blood pressure monitor contains image data of the region of interest displayed by the electronic blood pressure monitor, which includes the systolic blood pressure region and the diastolic blood pressure region; S2. Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor. S2-1. Acquire real-time image data from the electronic blood pressure monitor; S2-2. Based on the real-time image data of the electronic blood pressure monitor, the deep learning recognition model is used to obtain the region of interest (ROI) digital items of the real-time image data of the electronic blood pressure monitor. S2-2-1. Input the real-time image data of the electronic blood pressure monitor into the deep learning coarse recognition model to obtain the coarse recognition result of the real-time image data of the electronic blood pressure monitor; The region of interest in the real-time image data of the electronic blood pressure monitor is obtained by cropping the real-time image data of the electronic blood pressure monitor using the coarse recognition results. Based on the region of interest (ROI) of the real-time image data of the electronic blood pressure monitor, the deep learning fine recognition model is used to obtain the ROI digital item of the real-time image data of the electronic blood pressure monitor. S2-3. Arrange the digital items of the region of interest in the real-time image data of the electronic blood pressure monitor to obtain the recognition result of the real-time image data of the electronic blood pressure monitor. S3. Obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

2. The method for calibrating and identifying an electronic blood pressure monitor based on deep learning as described in claim 1, characterized in that, The deep learning coarse recognition model is trained using the image dataset from the electronic blood pressure monitor, including: S1-2-1. Perform data augmentation processing on the image dataset of the electronic blood pressure monitor to obtain an augmented dataset of the image dataset; S1-2-2, Divide the image dataset into augmented datasets to obtain augmented data training set and augmented data validation set; S1-2-3. Using the enhanced data training set as input and the corresponding image data of the enhanced data training set as output, train an initial deep learning coarse recognition model based on the YOLOv11 object detection network. S1-2-4. Based on the enhanced data validation set, input the initial deep learning coarse recognition model to obtain the image data corresponding to the enhanced data validation set; S1-2-5. Determine whether the image data corresponding to the augmented data validation set are all image data of the region of interest. If so, obtain the initial deep learning coarse recognition model as the deep learning coarse recognition model. Otherwise, update the augmented data training set using the augmented data validation set and return to S1-2-3.

3. The method for calibrating and identifying an electronic blood pressure monitor based on deep learning according to claim 2, characterized in that, The electronic blood pressure monitor calibration and identification results are obtained based on the recognition results of the real-time image data of the electronic blood pressure monitor, including: The recognition results of the real-time image data from the electronic blood pressure monitor are processed for display to obtain the display recognition results; The identification results are displayed and verified using the blood pressure monitor calibration procedure to obtain the electronic blood pressure monitor calibration identification results.

4. A deep learning-based electronic blood pressure monitor verification and identification system, implementing the method as described in any one of claims 1-3, characterized in that, include: Model building module, real-time recognition module, and result acquisition module; The model building module is used to build a deep learning recognition model using an electronic blood pressure monitor image dataset; The real-time recognition module is used to obtain the recognition result of the real-time image data of the electronic blood pressure monitor using the deep learning recognition model based on the real-time image data of the electronic blood pressure monitor. The result acquisition module is used to obtain the electronic blood pressure monitor verification and identification results based on the recognition results of the real-time image data of the electronic blood pressure monitor.

5. The system for the verification and identification method of an electronic blood pressure monitor based on deep learning according to claim 4, characterized in that, The system also includes a model optimization module, which is used to optimize the deep learning recognition model using the electronic blood pressure monitor image dataset and real-time acquired electronic blood pressure monitor image data.

Images