A specimen management system and management method for an intelligent storage and retrieval cabinet

Abstract

The present invention discloses a specimen management system and management method of an intelligent storage and retrieval cabinet, which relates to the technical field of specimen management, and includes a sample collection unit, a segmentation processing unit, a feature extraction unit, a layout execution unit, a face recognition unit, an equipment monitoring unit and an auxiliary prompt unit. The present invention further processes the image information after rough segmentation through the segmentation processing unit, so that the target object in the image is completely distinguished from the background, and the recognition of the edge of the target object is more practical. At the same time, the feature extraction unit is used to analyze the image information after segmentation, and the color feature data, shape feature data and texture feature data in the image information are extracted, so that the subsequent model can more accurately determine the category to which the specimen belongs. The layout execution unit is used to repeatedly confirm with the user according to the category information contained in the two-dimensional code and the model prediction result, so as to prevent the two-dimensional code information of the specimen from being inconsistent with the actual information.

Description

Technical Field

[0001] The present invention relates to the technical field of specimen management, and in particular to a specimen management system and a management method for an intelligent storage and retrieval cabinet. Background Art

[0002] The specimen management system can centrally store and manage body fluid specimens, record the specific information of each specimen, track the available storage space in the storage cabinet and monitor the storage capacity, save time, and reduce the risk of misplacing or losing samples. The invention patent with publication number CN112071411A discloses "an intelligent specimen management cabinet, an intelligent specimen management system and a management method. The intelligent specimen management cabinet includes a cabinet, a computer, a micro refrigerator, a micro incubator, a scanner, and an identity recognition device; the computer is installed on the surface of the cabinet, which controls and stores information, connects with the hospital's LIS and HIS systems, and is used by nurses. PDA docking; a micro refrigerator and a micro incubator are arranged in the middle of the cabinet. The micro refrigerator is used to store low-temperature specimens separately to extend the storage time, and the micro incubator is used to store microbiology room specimens to provide a suitable growth environment for pathogens; a scanner and an identity recognition device are arranged on the surface of the cabinet. The scanner is used to recognize the barcode on the test tube, and the identity recognition device is used to identify all personnel operating the cabinet. The present invention enables timely receipt, timely storage, and timely inspection and management of inpatient specimens; compliance with specimen management specifications, and improvement of specimen inspection quality and inspection time; traceability of the specimen delivery process; intelligent transportation".

[0003] The above-mentioned existing technology solves the problems of incorrect and incomplete specimen information filling and improper specimen retention. However, during the use of the system, since the category of the specimen is determined entirely by the information contained in the QR code, it is impossible to avoid the situation where the QR code information of the specimen is inconsistent with the actual information, affecting the subsequent detection of the specimen. At the same time, the person who takes the specimen is not identified, so the specimen may be lost and replaced during storage. The operating status of the equipment is not analyzed. When the equipment malfunctions, the maintenance personnel cannot take timely measures. Summary of the invention

[0004] The object of the present invention is to provide a specimen management system and management method for an intelligent storage and retrieval cabinet to solve the problems raised in the above-mentioned background technology.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solutions: a specimen management system for an intelligent storage and retrieval cabinet, comprising a face recognition unit and an equipment monitoring unit;

[0006] A sample collection unit, after collecting a plurality of different blood, urine and feces samples, uses an industrial camera in the detection area of ​​the intelligent storage and access cabinet to photograph each sample, numbers the obtained sample image information, adds corresponding labels, and stores them into a blood collection, a urine collection, and a feces collection according to categories;

[0007] A segmentation processing unit, wherein the segmentation processing unit receives sample image information and image information corresponding to a sample-free area in the detection area, performs differential processing on the two, and then obtains a rough segmentation result of the sample image information after binarization analysis and connected domain screening, determines the position of the seed point according to the rough segmentation result, calculates the color difference value between the seed points, sets a threshold value according to the obtained color difference value, and uses a difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is less than the threshold value, the current test point is determined to be a seed point, and the operation is repeated until all test points are calculated to obtain the image information after segmentation processing;

[0008] A feature extraction unit, wherein the feature extraction unit receives the segmented image information, determines the color feature data corresponding to the current image information by using the moment value, constructs the corresponding Hessian matrix by using the second-order derivatives of each point of the image information at different directions at the same scale, determines the specific position of the feature point and the corresponding surrounding area according to these data, determines the main direction of the feature point by using the response value and weight of each point in the area, selects a square area according to the main direction of each feature point, analyzes the description domain corresponding to the feature point according to the sum of the response values ​​in the square area, extracts shape feature data in the image information by using all the feature points and the corresponding description domain, counts the distances between different pixel points of the same gray value and the number of rows and columns corresponding to the image information, and then uses these data to analyze the texture feature data corresponding to the image information, inputs the feature data of each sample image information into the SVM model, and updates the model parameters by using the loss function, thereby completing the model training;

[0009] A layout execution unit, wherein the layout execution unit uses a barcode scanner to scan the QR code on the outside of the body fluid sample to be tested, obtains the category corresponding to the body fluid sample to be tested and user personal information, uses an industrial camera to photograph the body fluid sample to be tested, performs segmentation processing on the image information corresponding to the body fluid sample to be tested, determines feature data corresponding to the segmented image information, and uses an SVM model to calculate the predicted category of the body fluid sample to be tested. If the category predicted by the model is inconsistent with the category corresponding to the body fluid sample to be tested, wait for user confirmation. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, a prompt is given to change the QR code information of the body fluid sample to be tested;

[0010] An auxiliary prompt unit, after the auxiliary prompt unit sets the disinfection time length, before performing the sampling operation, the space in the box is automatically disinfected, and the current storage quantity and location of specimens of different categories are output through a visual interface. After the sampling operation, the remaining number of specimens in the current box and the remaining storage time are counted. If the remaining number is lower than the preset value, a prompt is given to replenish the specimens. If the remaining storage time of the specimens in the box is lower than the preset value, the specimen information is output through a visual interface, and a prompt is given to remove the specimens in time, wherein the specimen information includes the number, user personal information, date and specific location. A camera is used to shoot the sampling personnel and the layout personnel, and the obtained image information and time are transmitted to the database for storage. After the hand sanitizer information in the database is extracted, the gravity sensor is used to monitor the weight of the hand sanitizer, and the data are used to determine whether the current hand sanitizer weight is lower than the threshold. If it is lower than the threshold, a prompt is given to replace the hand sanitizer, otherwise no prompt is given.

[0011] Preferably, the segmentation processing unit includes an image optimization module, a connected domain determination module and a seed point detection module. After receiving the sample image information, the image optimization module sets the filter parameters and the window size, and scans each pixel in the image information one by one. If the current pixel is at the center of the window, the spatial domain weight and the value domain weight of the pixels around the pixel are counted, and the value of the current pixel is calculated according to the weight and the surrounding pixel values. After all pixels are scanned, the processed image information is stored in the image set. The connected domain determination module extracts the image information corresponding to the sample-free area in the detection area, performs differential processing on the current image information and the sample-free image information, and then obtains the rough segmentation result of the image information after binarization analysis and connected domain screening. , according to the rough segmentation result, determine the centroid point in the connected domain and the position coordinates of the two endpoints adjacent to the point, mark the three points as seed points, and store them in the seed set. The seed point detection module determines the parameter values ​​of the seed point and each point in the connected domain in the Lab color model, calculates the color difference value between the seed points, sets the threshold value according to the obtained color difference value, and uses the difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is greater than or equal to the threshold, the current test point is determined to be a non-seed point, otherwise it is determined to be a seed point, and it is stored in the set. Repeat the operation until all the test points are calculated, and the image information after segmentation is obtained. The filter is specifically an LC low-pass filter, and the difference analysis algorithm is specifically as follows:

[0012]

[0013] In the formula, Indicates the color difference value, Indicates the L value of the seed point in the Lab color model. Indicates the L value of the growth point in the Lab color model. Indicates the a value of the seed point in the Lab color model. Indicates the a value of the point to be grown in the Lab color model. Indicates the b value of the seed point in the Lab color model. Indicates the b value of the point to be grown in the Lab color model.

[0014] Preferably, the feature extraction unit includes a color feature analysis module, a response value calculation module and a main direction determination module, the color feature analysis module receives the segmented image information, calculates the moment value corresponding to each color channel according to the RGB value corresponding to the pixel point using a color distribution algorithm, and determines the color feature data corresponding to the current image information using the moment value, the response value calculation module determines the grayscale value corresponding to the pixel point in the image information, processes the image information through Gaussian filtering, and calculates the second-order derivatives of each point of the processed image information in different directions, and constructs the corresponding Hessian matrix using the second-order derivatives in different directions of the same scale, and determines the color feature data corresponding to the current image information using the moment value. The response value of each pixel is calculated by the matrix, the main direction determination module sets the size of the spatial area, determines all extreme points contained in the image in the spatial area according to the response value of each pixel at different scales, determines the specific position of the feature point and the surrounding area corresponding to the point according to the extreme point, and then counts the response values ​​of all pixels in the area, sets the response angle range, and calculates the corresponding weight according to the distance between each pixel and the feature point. The response sum at different response angles in the corresponding area is calculated using the response value and weight of each point, and the response angle corresponding to the maximum response sum is taken as the main direction of the feature point. The color distribution algorithm is specifically as follows:

[0015]

[0016] In the formula, Indicates the image The first-order color moment corresponding to the color channel is Represents the total number of pixels. Indicates the image The median value of the color channel is The probability of a pixel appearing is Indicates the image The second-order color moment corresponding to the color channels is Indicates the image The third-order color moment corresponding to the color channel is Indicates a parameter.

[0017] Preferably, the feature extraction unit also includes a shape feature analysis module, a texture feature analysis module and a model training module. The shape feature analysis module selects a corresponding square area according to the main direction of each feature point, divides the area into sixteen sub-areas, selects twenty-five pixels in each sub-area, calculates the corresponding response value sum, analyzes the description domain corresponding to the feature point according to the response value sum in the square area, and uses all feature points and corresponding description domains to extract shape feature data in the image information. After extracting the image information, the texture feature analysis module counts the distances between different pixels of the same gray value and the number of rows and columns corresponding to the image information, and calculates the number of all pixel pairs appearing at the same distance in different directions according to the distance value, and uses the number of pixel pairs corresponding to each distance value to analyze the texture feature data corresponding to the image information. The model training module normalizes the color feature data, shape feature data and texture feature data of each sample image information respectively, and then inputs them into the SVM model, verifies the results generated by the model according to the label information, and uses the loss function to update the model parameters according to the offset value between the two, thereby completing the model training. The description domain of the feature point is specifically the vector information in the area around the feature point.

[0018] Preferably, the layout execution unit includes a QR code scanning module, a specimen shooting module, a specimen category prediction module, a category comparison module and a user confirmation module. The QR code scanning module uses a barcode scanner to scan the QR code on the outside of the body fluid specimen to be tested, obtains the category corresponding to the body fluid specimen to be tested and user personal information, outputs them through a visual interface, and after returning the user confirmation information, transmits these data to a database. The specimen shooting module uses an industrial camera to shoot the body fluid specimen to be tested, and transmits the obtained image information and shooting time to the database. The specimen category prediction module performs segmentation processing on the image information corresponding to the body fluid specimen to be tested, and determines the color feature data corresponding to the segmented image information. , shape feature data and texture feature data, and transmit these feature data to the SVM model to obtain the predicted category of the body fluid sample to be tested. If the category predicted by the model is consistent with the category corresponding to the body fluid sample to be tested, the category comparison module prompts the user to perform a layout operation through a visual interface. Otherwise, the visual interface prompts that the category of the sample is incorrect and waits for user confirmation. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, the user confirmation module prompts to change the QR code information of the body fluid sample to be tested. If the user confirms that the current body fluid sample to be tested is not the category predicted by the model, the user is prompted to perform a layout operation through a visual interface. The industrial camera is specifically a Hikvision industrial camera.

[0019] Preferably, the face recognition unit includes a feature map output module, a pooling processing module, a feature refinement module, a similarity analysis module and a volume autonomous setting module. The feature map output module uses a camera to shoot the user's face, and the obtained image information is transmitted to the face recognition model after denoising. The convolution kernel of the size performs feature transformation on the current image information and outputs the corresponding feature map. The pooling processing module uses two The feature map is downsampled by the pooling kernels of different sizes respectively, one of the pooling kernels is used to slide on the feature map, and the maximum value in each framed area is output. The feature map after maximum value pooling is obtained based on these data, and the other pooling kernel is used to slide on the feature map to output the average value in each framed area. The feature map after average value pooling is obtained based on these data. The feature refinement module calculates the feature map after maximum value pooling and the feature map after average value pooling through the feature summation algorithm to obtain an aggregated feature map, and outputs the refined facial feature information after the convolution operation. The similarity analysis module extracts the facial feature information stored in the database, and compares the refined facial feature information with all the stored facial feature information in turn to obtain to multiple similarity values, if the highest similarity value is greater than the threshold, it is determined that the facial information of the current person already exists and the sampling operation is allowed, otherwise it is determined that the facial information of the current person does not exist and the sampling operation is not allowed. If the facial information of the current person already exists, the volume autonomous setting module retrieves the personal information associated with the facial feature information, and determines the age data of the current person according to the personal information. If the facial information of the current person does not exist, the age recognition model constructed by deep learning technology is used to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data. The camera is specifically a Hikvision camera, and the face recognition model is specifically a face feature point recognition model based on the ResNet network.

[0020] Preferably, the equipment monitoring unit includes a state division module, a data prediction module and a weight update module. The state division module divides the equipment state into five types, namely normal operation, lock control damage, refrigeration abnormality, network offline and disinfection switch abnormality, extracts real-time monitoring data of the equipment, wherein the equipment is an intelligent storage and retrieval cabinet, and the monitoring data includes actual voltage of the equipment, maximum power of the equipment, electromagnetic sensor data, network connection status, temperature and humidity sensor data and sound sensor data, which are stored in a database. The data prediction module inputs the real-time monitoring data of the equipment into the ARIMA model, XGBoost model and LSTM model for analysis, determines the prediction data generated by each model, sets the current weight corresponding to each model, calculates the final prediction data of the equipment at the next time point according to the current weight and the prediction data of each model, uses these data to determine the current state of the equipment, and outputs it through a visual interface. The weight update module uses the prediction data of the equipment and the actual monitoring data of the next time period to calculate the key coefficient, analyzes these data according to the weight update algorithm, and obtains the current weight of each model. After that, the prediction data of the equipment at the next time point is calculated according to the current weight.

[0021] The specimen management method of the intelligent storage and retrieval cabinet comprises the following steps:

[0022] S1. Collect sample image information: Use the industrial camera in the detection area of ​​the smart storage and access cabinet to take pictures of each body fluid sample, number the obtained sample image information, add corresponding labels, and store them into blood collection, urine collection and feces collection according to categories;

[0023] S2, image information segmentation processing: receiving the rough segmentation result of the image information, determining the position of the seed point, calculating the color difference value between the seed points, setting a threshold value according to the obtained color difference value, and using the difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is less than the threshold, the current test point is determined to be a seed point. Repeat the operation until all the test points are calculated, and obtain the image information after segmentation processing;

[0024] S3, extracting sample image features: after receiving the segmented image information, analyzing the corresponding color feature data, shape feature data and texture feature data according to the moment value and gray value of the image information, inputting these data into the sample classification model, updating the model parameters, and completing the model training;

[0025] S4. Determine the category of the test specimen: Use a barcode scanner to scan the QR code on the outside of the body fluid specimen to be tested, obtain the category corresponding to the body fluid specimen to be tested and the user's personal information, use the SVM model to calculate the predicted category of the body fluid specimen to be tested, if the category predicted by the model is inconsistent with the category corresponding to the body fluid specimen to be tested, wait for user confirmation, if the user confirms that the current category of the body fluid specimen to be tested is the category predicted by the model, then prompt to change the QR code information of the body fluid specimen to be tested;

[0026] S5. Identify facial information of a person: Use a camera to shoot the user's face and transmit it to the face recognition model. After the convolution operation, the refined facial feature information is output. The refined facial feature information is compared with all the stored facial feature information in sequence. If the comparison passes, it is determined that the sampling operation is allowed, otherwise it is determined that the sampling operation is not allowed;

[0027] S6. Adjust the volume of the device: Use the age recognition model built by deep learning technology to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data;

[0028] S7, equipment status prediction: The real-time monitoring data of the smart locker is input into the ARIMA model, XGBoost model and LSTM model for analysis respectively, and the final prediction data of the locker at the next time point is calculated according to the current weight and the prediction data of each model, and the status of the locker is analyzed using these data;

[0029] S8. Auxiliary prompt: After extracting the hand sanitizer information from the database, use the gravity sensor to monitor the weight of the hand sanitizer. Based on these data, determine whether the current weight of the hand sanitizer is lower than the threshold. If it is lower than the threshold, a prompt will be given to replace the hand sanitizer, otherwise no prompt will be given.

[0030] Compared with the prior art, the present invention has the following beneficial effects:

[0031] 1. The present invention further processes the image information after rough segmentation through the segmentation processing unit, so that the target object in the image is completely distinguished from the background, and the recognition of the edge of the target object is more practical. At the same time, the feature extraction unit is used to analyze the image information after segmentation, and the color feature data, shape feature data and texture feature data in the image information are extracted, so that the subsequent model can more accurately determine the category to which the specimen belongs. Since a large amount of sample image information of different categories is collected to train the SVM model and the parameters are continuously updated, the prediction result of the model is guaranteed to be more accurate and reliable. The lofting execution unit is used to repeatedly confirm with the user according to the category information contained in the two-dimensional code and the model prediction result, so as to prevent the situation that the two-dimensional code information of the specimen is inconsistent with the actual information;

[0032] 2. The present invention compares the facial feature information of the sampling personnel through the face recognition unit, so that the personnel whose facial feature information is retained in the database are allowed to perform the operation, thereby preventing the specimen from being replaced and taken out by other personnel during the storage process, protecting the privacy of the user, and at the same time analyzing the corresponding age of the current person according to the facial information, and automatically adjusting the broadcast volume of the equipment according to the age, so that older people can also clearly hear a series of prompt information issued by the equipment, and the equipment monitoring unit will analyze the current operating status of the equipment according to the real-time monitoring data. Once the equipment is abnormal, the maintenance personnel can take intervention measures in time, and the auxiliary prompt unit is to facilitate users to better use the equipment and ensure the safety of the specimen. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] Figure 1 Provides a schematic diagram of the overall system flow for an embodiment of the present invention;

[0034] Figure 2 A block diagram of the internal modules of a segmentation processing unit provided in an embodiment of the present invention;

[0035] Figure 3 An internal module block diagram of a feature extraction unit provided by an embodiment of the present invention;

[0036] Figure 4 An internal module block diagram of a lofting execution unit provided in an embodiment of the present invention;

[0037] Figure 5 A block diagram of the internal modules of a face recognition unit provided by an embodiment of the present invention;

[0038] Figure 6 This is a block diagram of the internal modules of the equipment monitoring unit provided in an embodiment of the present invention.

[0039] In the figure: 1. Sample acquisition unit; 2. Segmentation processing unit; 201. Image optimization module; 202. Connected domain determination module; 203. Seed point detection module; 3. Feature extraction unit; 301. Color feature analysis module; 302. Response value calculation module; 303. Main direction determination module; 304. Shape feature analysis module; 305. Texture feature analysis module; 306. Model training module; 4. Stakeout execution unit; 401. QR code scanning module; 402. Specimen shooting module; 403. Specimen category prediction module; 404. Category comparison module; 405. User confirmation module; 5. Face recognition unit; 501. Feature map output module; 502. Pooling processing module; 503. Feature refinement module; 504. Similarity analysis module; 505. Volume autonomous setting module; 6. Equipment monitoring unit; 601. State division module; 602. Data prediction module; 603. Weight update module; 7. Auxiliary prompt unit. DETAILED DESCRIPTION

[0040] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0041] See also Figure 1-Figure 6 ,The present invention provides a technical solution: a specimen management system for an intelligent storage and retrieval cabinet, comprising a face recognition unit 5 and an equipment monitoring unit 6;

[0042] The sample collection unit 1 collects a plurality of different blood, urine and feces samples, and then uses an industrial camera in the detection area of ​​the intelligent storage and access cabinet to photograph each sample, numbers the obtained sample image information, adds corresponding labels, and stores them into a blood collection, a urine collection and a feces collection according to categories;

[0043] Segmentation processing unit 2, segmentation processing unit 2 receives sample image information and image information corresponding to the sample-free area, performs differential processing on the two, and then obtains a rough segmentation result of the sample image information after binarization analysis and connected domain screening, determines the seed point position according to the rough segmentation result, calculates the color difference value between the seed points, sets a threshold value according to the obtained color difference value, and uses a difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is less than the threshold value, the current test point is determined to be a seed point, and the operation is repeated until all test points are calculated to obtain the image information after segmentation processing;

[0044] Feature extraction unit 3, feature extraction unit 3 receives the segmented image information, uses moment values ​​to determine color feature data corresponding to the current image information, uses the second-order derivatives of each point of the image information at the same scale and in different directions to construct a corresponding Hessian matrix, determines the specific position of the feature point and the corresponding surrounding area based on these data, uses the response value and weight of each point in the area to determine the main direction of the feature point, selects a square area based on the main direction of each feature point, analyzes the description domain corresponding to the feature point based on the sum of the response values ​​in the square area, extracts shape feature data from the image information using all feature points and the corresponding description domain, counts the distances between different pixel points of the same gray value and the number of rows and columns corresponding to the image information, uses these data to analyze the texture feature data corresponding to the image information, inputs the feature data of each sample image information into the SVM model, and uses the loss function to update the model parameters, thereby completing the model training;

[0045] The layout execution unit 4 uses a barcode scanner to scan the QR code on the outside of the body fluid sample to be tested, obtains the category corresponding to the body fluid sample to be tested and the user's personal information, uses an industrial camera to shoot the body fluid sample to be tested, performs segmentation processing on the image information corresponding to the body fluid sample to be tested, determines the feature data corresponding to the segmented image information, and uses the SVM model to calculate the predicted category of the body fluid sample to be tested. If the category predicted by the model is inconsistent with the category corresponding to the body fluid sample to be tested, wait for user confirmation. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, it prompts to change the QR code information of the body fluid sample to be tested;

[0046] Auxiliary prompt unit 7. After the auxiliary prompt unit 7 sets the disinfection time length, before performing the sampling operation, the space in the box is automatically disinfected, and the storage quantity and location of the current specimens of different categories are output through the visual interface. After the sampling operation, the remaining number of specimens in the current box and the remaining storage time are counted. If the remaining number is lower than the preset value, a prompt is given to replenish the specimens. If the remaining storage time of the specimens in the box is lower than the preset value, the specimen information is output through the visual interface, and a prompt is given to remove the specimens in time. The specimen information includes the number, user personal information, date and specific location. The sampling personnel and the layout personnel are photographed by a camera, and the obtained image information and time are transmitted to the database for storage. After the hand sanitizer information in the database is extracted, the gravity sensor is used to monitor the weight of the hand sanitizer. Based on these data, it is determined whether the current weight of the hand sanitizer is lower than the threshold. If it is lower than the threshold, a prompt is given to replace the hand sanitizer, otherwise no prompt is given.

[0047] The segmentation processing unit 2 includes an image optimization module 201, a connected domain determination module 202 and a seed point detection module 203. After receiving the sample image information, the image optimization module 201 sets the filter parameters and the window size, and scans each pixel in the image information one by one. If the current pixel is at the center of the window, the spatial domain weight and the value domain weight of the pixels around the pixel are counted, and the value of the current pixel is calculated according to the weight and the surrounding pixel values. After all the pixels are scanned, the processed image information is stored in the image set. The connected domain determination module 202 extracts the image information corresponding to the sample-free area in the detection area, performs differential processing on the current image information and the sample-free image information, and then obtains the rough image information after binarization analysis and connected domain screening. Segmentation result, determine the centroid point in the connected domain and the position coordinates of the two endpoints adjacent to the point according to the rough segmentation result, mark the three points as seed points, and store them in the seed set. The seed point detection module 203 determines the parameter values ​​of the seed point and each point in the connected domain in the Lab color model, calculates the color difference value between the seed points, sets the threshold value according to the obtained color difference value, and uses the difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is greater than or equal to the threshold, the current test point is determined to be a non-seed point, otherwise it is determined to be a seed point, and it is stored in the set. Repeat the operation until all test points are calculated, and the image information after segmentation processing is obtained. The filter is specifically an LC low-pass filter, and the difference analysis algorithm is specifically as follows:

[0048]

[0049] In the formula, Indicates the color difference value, Indicates the L value of the seed point in the Lab color model. Indicates the L value of the growth point in the Lab color model. Indicates the a value of the seed point in the Lab color model. Indicates the a value of the point to be grown in the Lab color model. Indicates the b value of the seed point in the Lab color model. Indicates the b value of the point to be grown in the Lab color model;

[0050] The feature extraction unit 3 includes a color feature analysis module 301, a response value calculation module 302 and a main direction determination module 303. The color feature analysis module 301 receives the segmented image information, calculates the moment value corresponding to each color channel according to the RGB value corresponding to the pixel point using a color distribution algorithm, and determines the color feature data corresponding to the current image information using the moment value. The response value calculation module 302 determines the grayscale value corresponding to the pixel point in the image information, processes the image information through Gaussian filtering, and calculates the second-order derivatives of each point of the processed image information in different directions, and constructs the corresponding Hessian matrix using the second-order derivatives in different directions of the same scale. The response value of each pixel is calculated according to the Hessian matrix. The main direction determination module 303 sets the size of the spatial region. According to the response value of each pixel at different scales, all extreme points contained in the image in the spatial region are determined. After the specific position of the feature point and the surrounding area corresponding to the point are determined according to the extreme point, the response values ​​of all pixels in the area are counted, and the response angle range is set. The corresponding weight is calculated according to the distance between each pixel and the feature point. The response sum at different response angles in the corresponding area is calculated using the response value and weight of each point. The response angle corresponding to the maximum response sum is taken as the main direction of the feature point. The color distribution algorithm is specifically as follows:

[0051]

[0052] In the formula, Indicates the image The first-order color moment corresponding to the color channel is Represents the total number of pixels. Indicates the image The median value of the color channel is The probability of a pixel appearing is Indicates the image The second-order color moment corresponding to the color channels is Indicates the image The third-order color moment corresponding to the color channel is Represents parameters;

[0053] The feature extraction unit 3 also includes a shape feature analysis module 304, a texture feature analysis module 305 and a model training module 306. The shape feature analysis module 304 selects a corresponding square area according to the main direction of each feature point, divides the area into sixteen sub-areas, selects twenty-five pixels in each sub-area, calculates the corresponding response value sum, analyzes the description domain corresponding to the feature point according to the response value sum in the square area, and extracts shape feature data in the image information using all feature points and the corresponding description domain. After the texture feature analysis module 305 extracts the image information, it counts the distances between different pixels of the same gray value and the number of rows and columns corresponding to the image information, and then calculates the number of all pixel pairs appearing at the same distance in different directions according to the distance value, and analyzes the texture feature data corresponding to the image information using the number of pixel pairs corresponding to each distance value. The model training module 306 normalizes the color feature data, shape feature data and texture feature data of each sample image information respectively, and then inputs them into the SVM model, verifies the results generated by the model according to the label information, and uses the loss function to update the model parameters according to the offset value between the two, thereby completing the model training;

[0054] The layout execution unit 4 includes a two-dimensional code scanning module 401, a specimen shooting module 402, a specimen category prediction module 403, a category comparison module 404 and a user confirmation module 405. The two-dimensional code scanning module 401 uses a barcode scanner to scan the two-dimensional code on the outside of the body fluid specimen to be tested, obtains the category and user personal information corresponding to the body fluid specimen to be tested, outputs it through a visual interface, and after returning the user confirmation information, transmits these data to the database. The specimen shooting module 402 uses an industrial camera to shoot the body fluid specimen to be tested, and transmits the obtained image information and shooting time to the database. The specimen category prediction module 403 performs segmentation processing on the image information corresponding to the body fluid specimen to be tested, and determines the segmented image information corresponding to the user information. The color feature data, shape feature data and texture feature data are transmitted to the SVM model to obtain the predicted category of the body fluid sample to be tested. If the category predicted by the model is consistent with the category corresponding to the body fluid sample to be tested, the category comparison module 404 prompts the user to perform a layout operation through a visual interface. Otherwise, the visual interface prompts that the category of the sample is incorrect and waits for user confirmation. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, the user confirmation module 405 prompts to change the QR code information of the body fluid sample to be tested. If the user confirms that the current body fluid sample to be tested is not the category predicted by the model, the user is prompted to perform a layout operation through a visual interface. The industrial camera is specifically a Hikvision industrial camera;

[0055] The face recognition unit 5 includes a feature map output module 501, a pooling processing module 502, a feature refinement module 503, a similarity analysis module 504 and a volume autonomous setting module 505. The feature map output module 501 uses a camera to shoot the user's face, and the obtained image information is transmitted to the face recognition model after denoising. The convolution kernel of the size performs feature transformation on the current image information and outputs the corresponding feature map. The pooling processing module 502 uses two The feature map is downsampled by the pooling kernels of different sizes respectively, one of the pooling kernels is used to slide on the feature map, and the maximum value in each frame selection area is output. The feature map after maximum value pooling is obtained based on these data, and the average value in each frame selection area is output by sliding on the feature map through another pooling kernel. The feature map after average value pooling is obtained based on these data. The feature refinement module 503 calculates the feature map after maximum value pooling and the feature map after average value pooling through a feature summation algorithm to obtain an aggregated feature map, and outputs the refined face feature information after convolution operation. The similarity analysis module 504 extracts the face feature information stored in the database, and compares the refined face feature information with all the stored face feature information in turn to obtain multiple Similarity value, if the highest similarity value is greater than the threshold, it is determined that the facial information of the current person already exists and the sampling operation is allowed, otherwise it is determined that the facial information of the current person does not exist and the sampling operation is not allowed. If the facial information of the current person already exists, the volume self-setting module 505 retrieves the personal information associated with the facial feature information, and determines the age data of the current person according to the personal information. If the facial information of the current person does not exist, the age recognition model constructed by deep learning technology is used to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data. The camera is specifically a Hikvision camera, the face recognition model is specifically a face feature point recognition model based on the ResNet network, and the feature summation algorithm is specifically:

[0056]

[0057] In the formula, represents the aggregate feature map, represents the convolution operation, Represents the feature map after maximum pooling, represents the feature map after average pooling, Represents a compression processing function;

[0058] The equipment monitoring unit 6 includes a state division module 601, a data prediction module 602 and a weight update module 603. The state division module 601 divides the equipment state into five types, namely normal operation, lock control damage, refrigeration abnormality, network offline and disinfection switch abnormality, extracts the real-time monitoring data of the equipment, wherein the equipment is an intelligent storage cabinet, and the monitoring data includes the actual voltage of the equipment, the maximum power of the equipment, the electromagnetic sensor data, the network connection status, the temperature and humidity sensor data and the sound sensor data, and stores them in the database. The data prediction module 602 inputs the real-time monitoring data of the equipment into the ARIMA model, the XGBoost model and the LSTM model for analysis, determines the prediction data generated by each model, sets the current weight corresponding to each model, calculates the final prediction data of the equipment at the next time point according to the current weight and the prediction data of each model, uses these data to determine the current state of the equipment, and outputs it through a visual interface. The weight update module 603 uses the prediction data of the equipment and the actual monitoring data of the next time period to calculate the key coefficient, analyzes these data according to the weight update algorithm, and after obtaining the current weight of each model, calculates the prediction data of the equipment at the next time point according to the current weight. The weight update algorithm is specifically as follows:

[0059]

[0060]

[0061] In the formula, Indicates The combined weight of the sub-models, Indicates The key coefficients of the sub-models are Indicates The transformed key coefficients in each sub-model are represents the total number of models, Represents parameters;

[0062] A specimen management method for an intelligent storage and retrieval cabinet comprises the following steps:

[0063] S1. Collect sample image information: Use the industrial camera in the detection area of ​​the smart storage and access cabinet to take pictures of each body fluid sample, number the obtained sample image information, add corresponding labels, and store them into blood collection, urine collection and feces collection according to categories;

[0064] S2, image information segmentation processing: receiving the rough segmentation result of the image information, determining the position of the seed point, calculating the color difference value between the seed points, setting a threshold value according to the obtained color difference value, and using the difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is less than the threshold, the current test point is determined to be a seed point. Repeat the operation until all the test points are calculated, and obtain the image information after segmentation processing;

[0065] S3, extracting sample image features: after receiving the segmented image information, analyzing the corresponding color feature data, shape feature data and texture feature data according to the moment value and gray value of the image information, inputting these data into the sample classification model, updating the model parameters, and completing the model training;

[0066] S4. Determine the category of the test specimen: Use a barcode scanner to scan the QR code on the outside of the body fluid specimen to be tested, obtain the category corresponding to the body fluid specimen to be tested and the user's personal information, use the SVM model to calculate the predicted category of the body fluid specimen to be tested, if the category predicted by the model is inconsistent with the category corresponding to the body fluid specimen to be tested, wait for user confirmation, if the user confirms that the current category of the body fluid specimen to be tested is the category predicted by the model, then prompt to change the QR code information of the body fluid specimen to be tested;

[0067] S5. Identify facial information of a person: Use a camera to shoot the user's face and transmit it to the face recognition model. After the convolution operation, the refined facial feature information is output. The refined facial feature information is compared with all the stored facial feature information in sequence. If the comparison passes, it is determined that the sampling operation is allowed, otherwise it is determined that the sampling operation is not allowed;

[0068] S6. Adjust the volume of the device: Use the age recognition model built by deep learning technology to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data;

[0069] S7, equipment status prediction: The real-time monitoring data of the smart locker is input into the ARIMA model, XGBoost model and LSTM model for analysis respectively, and the final prediction data of the locker at the next time point is calculated according to the current weight and the prediction data of each model, and the status of the locker is analyzed using these data;

[0070] S8. Auxiliary prompt: After extracting the hand sanitizer information from the database, use the gravity sensor to monitor the weight of the hand sanitizer. Based on these data, determine whether the current weight of the hand sanitizer is lower than the threshold. If it is lower than the threshold, a prompt will be given to replace the hand sanitizer, otherwise no prompt will be given.

[0071] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0072] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A specimen management system for an intelligent storage cabinet, comprising a face recognition unit (5) and an equipment monitoring unit (6), characterized in that: A sample collection unit (1), wherein the sample collection unit (1) collects a plurality of different blood, urine and feces samples, and then uses an industrial camera in a detection area of ​​the intelligent storage and access cabinet to photograph each sample, numbers the obtained sample image information, adds corresponding labels, and then stores the information according to categories into a blood collection, a urine collection and a feces collection; A segmentation processing unit (2), the segmentation processing unit (2) receiving sample image information and image information corresponding to a sample-free area in a detection region, performing differential processing on the two, and then performing binarization analysis and connected domain screening to obtain a rough segmentation result of the sample image information, determining the position of a seed point according to the rough segmentation result, calculating the color difference value between the seed points, setting a threshold value according to the obtained color difference value, and calculating the color difference value between the test point and the adjacent seed point in the connected domain using a difference analysis algorithm, if the value is less than the threshold value, determining that the current test point is a seed point, repeating the operation until all test points are calculated, and obtaining the image information after segmentation processing; A feature extraction unit (3), wherein the feature extraction unit (3) receives the segmented image information, determines the color feature data corresponding to the current image information using the moment value, constructs the corresponding Hessian matrix using the second-order derivatives of each point of the image information at the same scale and in different directions, determines the specific position of the feature point and the corresponding surrounding area based on these data, determines the main direction of the feature point using the response value and weight of each point in the area, selects a square area based on the main direction of each feature point, analyzes the description domain corresponding to the feature point based on the sum of the response values ​​in the square area, extracts shape feature data from the image information using all the feature points and the corresponding description domain, counts the distances between different pixel points of the same gray value and the number of rows and columns corresponding to the image information, and uses these data to analyze the texture feature data corresponding to the image information, inputs the feature data of each sample image information into the SVM model, and updates the model parameters using the loss function, thereby completing the model training; A layout execution unit (4), wherein the layout execution unit (4) uses a barcode scanner to scan the QR code on the outside of the body fluid sample to be tested, obtains the category corresponding to the body fluid sample to be tested and user personal information, uses an industrial camera to photograph the body fluid sample to be tested, performs segmentation processing on the image information corresponding to the body fluid sample to be tested, determines feature data corresponding to the segmented image information, and uses an SVM model to calculate the predicted category of the body fluid sample to be tested. If the category predicted by the model is inconsistent with the category corresponding to the body fluid sample to be tested, then wait for user confirmation. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, then prompts the user to change the QR code information of the body fluid sample to be tested; An auxiliary prompt unit (7), wherein after the auxiliary prompt unit (7) sets the disinfection time length, before the sampling operation is performed, the space in the box is automatically disinfected, and the storage quantity and location of the current specimens of different categories are output through a visual interface. After the sampling operation, the remaining number of specimens in the current box and the remaining storage time are counted. If the remaining number is lower than a preset value, a prompt is given to replenish the specimens. If the remaining storage time of the specimens in the box is lower than a preset value, the specimen information is output through a visual interface, and a prompt is given to remove the specimens in time, wherein the specimen information includes a number, user personal information, date and specific location. A camera is used to shoot the sampling personnel and the sample placement personnel, and the obtained image information and time are transmitted to a database for storage. After the hand sanitizer information in the database is extracted, the weight of the hand sanitizer is monitored using a gravity sensor. Based on these data, it is determined whether the current weight of the hand sanitizer is lower than a threshold value. If it is lower than the threshold value, a prompt is given to replace the hand sanitizer, otherwise no prompt is given.

2. The specimen management system of the intelligent storage and retrieval cabinet according to claim 1, characterized in that: The segmentation processing unit (2) comprises an image optimization module (201), a connected domain determination module (202) and a seed point detection module (203). After receiving the sample image information, the image optimization module (201) sets the filter parameters and the window size, scans each pixel point in the image information one by one, and if the current pixel point is at the center point of the window, counts the spatial domain weight and the value domain weight of the pixels around the pixel point, calculates the value of the current pixel point according to the weight and the surrounding pixel values, and stores the processed image information in the image set until all pixel points are scanned. The connected domain determination module (202) extracts the image information corresponding to the sample-free area in the detection area, performs differential processing on the current image information and the sample-free image information, and then performs binarization. After analysis and connected domain screening, a rough segmentation result of the image information is obtained. The centroid point in the connected domain and the position coordinates of the two endpoints adjacent to the point are determined according to the rough segmentation result. The three points are marked as seed points and stored in a seed set. The seed point detection module (203) determines the parameter values ​​of the seed point and each point in the connected domain in the Lab color model, calculates the color difference value between the seed points, sets a threshold value according to the obtained color difference value, and uses a difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is greater than or equal to the threshold, the current test point is determined to be a non-seed point. Otherwise, the current test point is determined to be a seed point and stored in the set. The operation is repeated until all test points are calculated, and the image information after segmentation is obtained.

3. The specimen management system of the intelligent storage and retrieval cabinet according to claim 1, characterized in that: The feature extraction unit (3) comprises a color feature analysis module (301), a response value calculation module (302) and a main direction determination module (303). The color feature analysis module (301) receives the segmented image information, calculates the moment value corresponding to each color channel according to the RGB value corresponding to the pixel point using a color distribution algorithm, and determines the color feature data corresponding to the current image information using the moment value. The response value calculation module (302) determines the grayscale value corresponding to the pixel point in the image information, processes the image information through Gaussian filtering, and calculates the second-order derivatives of each point of the processed image information in different directions, and uses the second-order derivatives in different directions of the same scale to obtain the color feature data corresponding to the current image information. A corresponding Hessian matrix is ​​constructed, and a response value of each pixel is calculated according to the Hessian matrix. The main direction determination module (303) sets the size of the spatial region, determines all extreme value points contained in the image in the spatial region according to the response value of each pixel at different scales, determines the specific position of the feature point and the surrounding area corresponding to the point according to the extreme value point, and then counts the response values ​​of all pixels in the area, sets the response angle range, calculates the corresponding weight according to the distance between each pixel and the feature point, calculates the response sum at different response angles in the corresponding area using the response value and weight of each point, and takes the response angle corresponding to the maximum response sum as the main direction of the feature point.

4. The specimen management system of the intelligent storage cabinet according to claim 3 is characterized in that: The feature extraction unit (3) further comprises a shape feature analysis module (304), a texture feature analysis module (305) and a model training module (306), wherein the shape feature analysis module (304) selects a corresponding square area according to the main direction of each feature point, divides the area into sixteen sub-areas, selects twenty-five pixel points in each sub-area, calculates the corresponding sum of response values, analyzes the description domain corresponding to the feature point according to the sum of response values ​​in the square area, extracts shape feature data in the image information using all feature points and the corresponding description domain, and extracts image information by using the texture feature analysis module (305). After that, the distances between different pixels of the same grayscale value and the number of rows and columns corresponding to the image information are counted, and then the number of all pixel pairs appearing at the same distance in different directions is calculated according to the distance value, and the texture feature data corresponding to the image information is analyzed using the number of pixel pairs corresponding to each distance value. The model training module (306) normalizes the color feature data, shape feature data and texture feature data of each sample image information respectively, and then inputs them into the SVM model, verifies the results generated by the model according to the label information, and uses the loss function to update the model parameters according to the offset value between the two, thereby completing the model training.

5. The specimen management system of the intelligent storage cabinet according to claim 1, characterized in that: The sample execution unit (4) comprises a two-dimensional code scanning module (401), a specimen photographing module (402), a specimen category prediction module (403), a category comparison module (404) and a user confirmation module (405). The two-dimensional code scanning module (401) uses a barcode scanner to scan the two-dimensional code on the outside of the body fluid specimen to be tested, obtains the category corresponding to the body fluid specimen to be tested and user personal information, outputs the information through a visual interface, and after returning user confirmation information, transmits the data to a database. The specimen photographing module (402) uses an industrial camera to photograph the body fluid specimen to be tested, and transmits the obtained image information and the photographing time to the database. The specimen category prediction module (403) classifies the image information corresponding to the body fluid specimen to be tested. The segmentation process is performed to determine the color feature data, shape feature data and texture feature data corresponding to the segmented image information, and these feature data are transmitted to the SVM model to obtain the predicted category of the body fluid sample to be tested. If the category predicted by the model is consistent with the category corresponding to the body fluid sample to be tested, the category comparison module (404) prompts the user to perform a layout operation through a visual interface. Otherwise, the visual interface prompts the user to indicate that the category of the sample is incorrect and waits for the user to confirm. If the user confirms that the current category of the body fluid sample to be tested is the category predicted by the model, the user confirmation module (405) prompts the user to change the QR code information of the body fluid sample to be tested. If the user confirms that the current category of the body fluid sample to be tested is not the category predicted by the model, the user is prompted to perform a layout operation through a visual interface.

6. The specimen management system of the intelligent storage cabinet according to claim 1, characterized in that: The face recognition unit (5) comprises a feature map output module (501), a pooling processing module (502), a feature refinement module (503), a similarity analysis module (504) and a volume autonomous setting module (505). The feature map output module (501) uses a camera to shoot the user's face, and after the obtained image information is subjected to denoising processing, it is transmitted to the face recognition model. The convolution kernel of the size performs feature transformation on the current image information and outputs the corresponding feature map. The pooling processing module (502) performs feature transformation on the current image information and outputs the corresponding feature map. The feature map is downsampled by using the pooling kernels of different sizes, one of the pooling kernels is used to slide on the feature map, and the maximum value in each framed area is outputted, and a feature map after maximum value pooling is obtained based on these data, and another pooling kernel is used to slide on the feature map, and the average value in each framed area is outputted, and a feature map after average value pooling is obtained based on these data, and the feature refinement module (503) calculates the feature map after maximum value pooling and the feature map after average value pooling through a feature summation algorithm to obtain an aggregated feature map, and outputs refined facial feature information after convolution operation, and the similarity analysis module (504) extracts the facial feature information stored in the database, and outputs the refined facial feature information. The facial feature information is compared with all stored facial feature information in turn to obtain multiple similarity values. If the highest similarity value is greater than a threshold, it is determined that the facial information of the current person already exists and the sampling operation is allowed. Otherwise, it is determined that the facial information of the current person does not exist and the sampling operation is not allowed. If the facial information of the current person already exists, the volume autonomous setting module (505) retrieves personal information associated with the facial feature information and determines the age data of the current person based on the personal information. If the facial information of the current person does not exist, the age recognition model constructed using deep learning technology is used to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data.

7. The specimen management system of the intelligent storage cabinet according to claim 1, characterized in that: The equipment monitoring unit (6) comprises a state classification module (601), a data prediction module (602) and a weight updating module (603), wherein the state classification module (601) divides the equipment state into five types, namely normal operation, lock control damage, refrigeration abnormality, network offline and disinfection switch abnormality, extracts real-time monitoring data of the equipment, wherein the equipment is a smart storage cabinet, and the monitoring data comprises actual voltage of the equipment, maximum power of the equipment, electromagnetic sensor data, network connection status, temperature and humidity sensor data and sound sensor data, and stores the data in a database, and the data prediction module (602) inputs the real-time monitoring data of the equipment into A The RIMA model, the XGBoost model and the LSTM model are analyzed to determine the prediction data generated by each model, and the current weight corresponding to each model is set. The final prediction data of the device at the next time point is calculated based on the current weight and the prediction data of each model. The current state of the device is determined using these data and outputted through a visual interface. The weight update module (603) calculates the key coefficients using the prediction data of the device and the actual monitoring data of the next time period. The data are analyzed according to the weight update algorithm. After the current weight of each model is obtained, the prediction data of the device at the next time point is calculated based on the current weight.

8. A specimen management method for an intelligent storage and retrieval cabinet, characterized in that: The specimen management method is applicable to a specimen management system of a smart storage and retrieval cabinet according to any one of claims 1 to 7, and comprises the following steps: S1. Collect sample image information: Use the industrial camera in the detection area of ​​the smart storage and access cabinet to take pictures of each body fluid sample, number the obtained sample image information, add corresponding labels, and store them into blood collection, urine collection and feces collection according to categories; S2, image information segmentation processing: receiving the rough segmentation result of the image information, determining the position of the seed point, calculating the color difference value between the seed points, setting a threshold value according to the obtained color difference value, and using the difference analysis algorithm to calculate the color difference value between the test point and the adjacent seed point in the connected domain. If the value is less than the threshold, the current test point is determined to be a seed point. Repeat the operation until all the test points are calculated, and obtain the image information after segmentation processing; S3, extracting sample image features: after receiving the segmented image information, analyzing the corresponding color feature data, shape feature data and texture feature data according to the moment value and gray value of the image information, inputting these data into the SVM model, updating the model parameters, and completing the model training; S4. Determine the category of the test specimen: Use a barcode scanner to scan the QR code on the outside of the body fluid specimen to be tested, obtain the category corresponding to the body fluid specimen to be tested and the user's personal information, use the SVM model to calculate the predicted category of the body fluid specimen to be tested, if the category predicted by the model is inconsistent with the category corresponding to the body fluid specimen to be tested, wait for user confirmation, if the user confirms that the current category of the body fluid specimen to be tested is the category predicted by the model, then prompt to change the QR code information of the body fluid specimen to be tested; S5. Identify facial information of a person: Use a camera to shoot the user's face and transmit it to the face recognition model. After the convolution operation, the refined facial feature information is output. The refined facial feature information is compared with all the stored facial feature information in sequence. If the comparison passes, it is determined that the sampling operation is allowed, otherwise it is determined that the sampling operation is not allowed; S6. Adjust the volume of the device: Use the age recognition model built by deep learning technology to analyze the facial feature information, output the predicted age data of the current person, and adjust the corresponding device playback volume according to the age data; S7, equipment status prediction: The real-time monitoring data of the smart locker is input into the ARIMA model, XGBoost model and LSTM model for analysis respectively, and the final prediction data of the locker at the next time point is calculated according to the current weight and the prediction data of each model, and the status of the locker is analyzed using these data; S8. Auxiliary prompt: After extracting the hand sanitizer information from the database, use the gravity sensor to monitor the weight of the hand sanitizer. Based on these data, determine whether the current weight of the hand sanitizer is lower than the threshold. If it is lower than the threshold, a prompt will be given to replace the hand sanitizer, otherwise no prompt will be given.

Images