Intelligent data management system and method applied to intelligent attendance checking

Abstract

The invention discloses an intelligent data management system and method applied to intelligent attendance, and relates to the technical field of artificial intelligence, the system comprises an image processing module, a standard determination module, a sample matching module and a standard updating module; the image processing module is used for collecting attendance checking images of attendance checking personnel and performing face recognition to obtain face images; the standard determination module is used for extracting face image features and forming feature vector standards; the sample matching module is used for performing face recognition and feature vector matching on the newly collected attendance checking images of the attendance checking personnel; and the standard updating module is used for periodically updating the characteristic vector standard according to the change condition of the face image characteristic vector of the card punching personnel at different time. The invention also provides a method for realizing the system. According to the invention, the condition of lack of a standard updating mechanism in an identification process in the prior art can be effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an intelligent data management system and method for smart attendance. Background Art

[0002] As the scale of enterprises expands and the demand for refined management increases, the disadvantages of traditional attendance methods become more prominent. Manual statistics are prone to errors, low efficiency, and data is easily lost, which makes it difficult to meet the complex attendance scenarios of modern enterprises. At the same time, technological development has promoted intelligent management to become a trend, and technologies such as big data, cloud computing, and mobile Internet are becoming increasingly mature. Against this background, smart attendance intelligent data management came into being. It can realize the automatic collection, real-time analysis and accurate statistics of attendance data, facilitate employee operations through multiple channels such as mobile terminals, provide enterprises with comprehensive and accurate attendance information, improve management efficiency, reduce management costs, and meet the urgent needs of enterprises for efficient, convenient, and scientific attendance management.

[0003] In the existing smart attendance technology, some simple classification models may not be able to handle complex facial feature vector classification problems well, resulting in low differentiation of feature vectors of different clocking-in personnel and prone to misjudgment. In addition, over time, facial features of personnel may change, such as changes in hairstyle, makeup, wearing glasses, etc. If the traditional attendance system does not have a corresponding update mechanism, it may fail to correctly identify employees. Summary of the invention

[0004] The purpose of the present invention is to provide an intelligent data management system and method for smart attendance to solve the problems raised in the prior art.

[0005] To achieve the above object, the present invention provides the following technical solutions: an intelligent data management system applied to smart attendance, the system comprising: an image processing module, a standard determination module, a sample matching module and a standard update module;

[0006] The image processing module is used to collect attendance images of clock-in personnel and perform face recognition to obtain face images; the standard determination module is used to extract features of face images and form feature vector standards; the sample matching module is used to perform face recognition and feature vector matching on newly collected attendance images of clock-in personnel; the standard update module is used to periodically update the feature vector standards according to changes in feature vectors of face images of clock-in personnel at different times;

[0007] The output end of the image processing module is connected to the input end of the standard determination module; the output end of the standard determination module is connected to the input end of the sample matching module; the output end of the sample matching module is connected to the input end of the standard updating module.

[0008] The image processing module includes an image acquisition unit, a face recognition unit and an image processing unit;

[0009] The image acquisition unit is used to collect the attendance image of the clock-in personnel; the face recognition unit is used to identify the attendance image using a face recognition algorithm; the image processing unit is used to crop and standardize the attendance image according to the recognition situation to obtain a face image;

[0010] The output end of the image acquisition unit is connected to the input end of the face recognition unit; the output end of the face recognition unit is connected to the input end of the image processing unit; the output end of the image processing unit is connected to the input end of the standard determination module.

[0011] The standard determination module includes a feature extraction unit, a hyperplane construction unit, a distance calculation unit and a standard determination unit;

[0012] The feature extraction unit is used to use a pre-trained model to extract features from a face image and form a feature vector; the hyperplane construction unit is used to construct hyperplanes for different punching personnel to classify the feature vectors of the face images of different punching personnel; the distance calculation unit is used to calculate the distance between all feature vectors corresponding to the face image of the same punching personnel and the hyperplane corresponding to the punching personnel; the standard determination unit is used to select the feature vector corresponding to the maximum distance as the feature vector standard;

[0013] The output end of the feature extraction unit is connected to the input end of the hyperplane construction unit; the output end of the hyperplane construction unit is connected to the input end of the distance calculation unit; the output end of the distance calculation unit is connected to the input end of the standard determination unit; the output end of the standard determination unit is connected to the input end of the sample matching module.

[0014] The sample matching module includes a new sample recognition unit, a similarity calculation unit and a face matching unit;

[0015] The new sample recognition unit is used to perform face recognition on the newly collected attendance image of the punching person to obtain the feature vector of the newly collected sample; the similarity calculation unit is used to calculate the cosine similarity between the new sample and all feature vector standards; the face matching unit is used to set a threshold and select the punching person corresponding to the feature vector standard with the highest cosine similarity value and meeting the threshold range as the face matching result;

[0016] The output end of the new sample recognition unit is connected to the input end of the similarity calculation unit; the output end of the similarity calculation unit is connected to the input end of the face matching unit; the output end of the face matching unit is connected to the input end of the standard update module.

[0017] The standard update module includes an image preprocessing unit, a timing model unit and a standard update unit;

[0018] The image preprocessing unit is used to obtain feature vectors and perform standardization processing; the time series model unit is used to build a time series model to analyze the trend of feature vector data changing over time; the standard updating unit is used to update the feature vector standard;

[0019] The output end of the image preprocessing unit is connected to the input end of the timing model unit; the output end of the timing model unit is connected to the input end of the standard updating unit.

[0020] An intelligent data management method for smart attendance, the method comprising the following steps:

[0021] Step 1: Collect attendance images of clock-in personnel and perform face recognition to obtain face images;

[0022] Step 2, extract the features of the face image and form a feature vector standard;

[0023] Step 3: Perform face recognition and feature vector matching on the newly collected attendance images of clock-in personnel;

[0024] Step 4: Update the feature vector standard according to the changes in the facial image feature vectors of the people who punch in at different times.

[0025] In step 1, the clock-in personnel are marked as: [D1, D2, …, D m ]; For clock-in personnel D i , collect its attendance images and mark them as: [K1,K2,…,K H(Di) ];

[0026] Among them, m is a positive integer, which indicates the number of clock-in personnel, D m represents the mth collector; i∈{1,2,…,m}, represents the sequence of clock-in personnel, D i represents the i-th clock-in person; H(Di) represents the number of attendance images of the i-th clock-in person; K H(Di) represents the H(Di)th attendance image of the i-th clock-in person;

[0027] For the attendance image K of the clock-in personnel Di j, perform face recognition on it: use the face detection algorithm HaarCascades to identify the face area in the attendance image; according to the face detection result, crop the face area from the attendance image; adjust and re-crop the cropped face area, standardize and normalize the face area, and obtain a face image P with a fixed size and a fixed pixel range. j ;

[0028] Where j is a positive integer, j∈{1,2,…,H(Di)}, representing the i-th collected personnel attendance image sequence, K j represents the jth attendance image of the i-th collected person; P j Represents the attendance image K j The corresponding face image;

[0029] Perform face recognition on all attendance images to obtain face images.

[0030] In step 2, for the clock-in person D i The face image P j , use the pre-trained deep learning model VGGFace to extract features from face images and obtain a feature vector of fixed length: F i,j =[F1,F2,…,F n ]; Standardize the feature vector;

[0031] Among them, F i,j Indicates the clock-in person D i The face image P j Corresponding feature vector; n is a positive integer, indicating the number of features, F n represents the nth feature;

[0032] Perform feature extraction on each face image to obtain the corresponding feature vector;

[0033] For clock-in personnel D i H(Di) groups of feature vectors corresponding to the face image are used to construct the hyperplane segmentation D i The facial image feature vector of and the facial image feature vectors of other clock-in personnel are:

[0034] The punch-in person D i The feature vector of is marked as positive, and the feature vectors of other clock-in personnel are marked as negative;

[0035] The SVM classification model is used to construct a segmentation hyperplane; the hyperplane is represented as: ω T x+b=0;

[0036] Among them, ω represents the normal vector, b is the bias, and x represents the input feature vector;

[0037] The model is trained based on the collected feature vectors, and the parameters ω and b are adjusted by minimizing the classification error and maximizing the margin between the two classes;

[0038] Calculate the eigenvector F i,j The distance from the hyperplane: d(F i,j )=|ω T ·F i,j +b| / ||ω||;

[0039] Among them, |ω T ·F i,j +b| is the eigenvector F i,j The algebraic distance to the hyperplane represents the projection of the eigenvector in the direction of the hyperplane normal vector; ||ω|| is the modulus of the normal vector, representing the length of the normal vector of the hyperplane;

[0040] Calculate the number of clock-in personnel D i The distance between the eigenvectors corresponding to all face images and the hyperplane, the eigenvector corresponding to the maximum distance is selected as the eigenvector standard;

[0041] Selecting the feature vector with the largest distance as the standard can ensure that the selected feature vector best represents the characteristics of the person punching the card. This feature vector has the largest separation from the hyperplane in the feature space and can better reflect the identity characteristics of the person punching the card.

[0042] Construct hyperplanes for the facial image feature vectors of all clock-in personnel, and calculate the feature vector standards corresponding to all clock-in personnel.

[0043] In step 3, face recognition is performed on the newly collected attendance image of the clock-in personnel to obtain the feature vector of the newly collected sample; a threshold A is set, and the cosine similarity between the feature vector of the newly collected sample and the standard feature vector of each clock-in personnel is calculated respectively: Sim = (F new ·F std ) / (||F new ||||F std ||); select the check-in personnel with the highest cosine similarity value and the feature vector standard that meets the threshold range as the face matching result; when the similarity does not meet the threshold, it is indicated as a matching anomaly;

[0044] Among them, F new Represents the feature vector of the newly collected sample; F std Denotes the eigenvector norm.

[0045] In step 4, for the clock-in person D i , record its t1~t rAttendance image data at each moment; face recognition is performed to obtain face image data, and feature extraction is performed to obtain feature vectors; the feature vectors are standardized to obtain feature vectors [F i (t1),F i (t2),...,F i (t r )];

[0046] Among them, r is a positive integer, which indicates the number of attendance times of clock-in personnel, t r Indicates the attendance of the clock-in person for the rth time; F i (t r ) represents the facial image feature vector corresponding to the r-th attendance check-in of the clock-in staff;

[0047] Build an LSTM time series model:

[0048] Determine the model parameters: input feature vector dimension n, number of hidden layer units h, learning rate α;

[0049] Construct the LSTM network structure: input layer, receiving standardized feature vectors; LSTM layer, capturing the long-term dependencies of time series through LSTM units; fully connected layer, mapping the output of LSTM layer to predicted values;

[0050] Use mean squared error as the loss function: Select the optimization algorithm Adam to train the model and minimize the loss function; input the standardized feature vector into the LSTM model, perform forward propagation, and calculate the output; calculate the loss and perform back propagation to update the model parameters;

[0051] Where u is a positive integer, u∈{1,2,...,r}, representing a sequence of moments; f u Indicates t u The real eigenvector at time f u,v Indicates t u The predicted feature vector at time instant;

[0052] Update the eigenvector norm:

[0053] Among them, F i,update Indicates the clock-in person D i Updated eigenvector criterion; F i,std Indicates the clock-in person D i The standard of the eigenvector before updating; β is the smoothing factor;

[0054] Update the facial image feature vector standards for all clock-in personnel.

[0055] In the case where features may change over time, the predicted feature vector can help capture the changing trend of the feature vector of the clock-in person. Through model prediction, potential changes can be taken into account, not just relying on past collected data.

[0056] Compared with the prior art, the beneficial effects of the present invention are: the present invention can update the feature vector standard in real time or periodically according to the actual changes, so that it can always accurately reflect the current facial features of the clocking-in personnel, thereby better adapting to the dynamic changes of personnel characteristics and reducing the failure of attendance recognition caused by changes in personnel appearance; the present invention uses the SVM classification model to construct a segmentation hyperplane, which has good performance in processing binary classification problems, can effectively distinguish the feature vectors of different clocking-in personnel, and adjusts the parameters by minimizing the classification error and maximizing the interval between the two classes, so that the constructed hyperplane can classify the feature vector more accurately, which provides a reliable basis for the subsequent selection of feature vector standards and improves the classification performance of the overall model. BRIEF DESCRIPTION OF THE DRAWINGS

[0057] Figure 1 A schematic diagram of a flow chart of an intelligent data management system for smart attendance according to the present invention;

[0058] Figure 2 The present invention is a schematic diagram of the steps of an intelligent data management method applied to smart attendance. DETAILED DESCRIPTION

[0059] 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.

[0060] Example: Figure 1-Figure 2 As shown, the present invention provides a technical solution, an intelligent data management system applied to smart attendance, the system comprising: an image processing module, a standard determination module, a sample matching module and a standard update module;

[0061] The image processing module is used to collect attendance images of clock-in personnel and perform face recognition to obtain face images; the standard determination module is used to extract features of face images and form feature vector standards; the sample matching module is used to perform face recognition and feature vector matching on newly collected attendance images of clock-in personnel; the standard update module is used to periodically update the feature vector standards according to changes in feature vectors of face images of clock-in personnel at different times;

[0062] The output end of the image processing module is connected to the input end of the standard determination module; the output end of the standard determination module is connected to the input end of the sample matching module; the output end of the sample matching module is connected to the input end of the standard updating module.

[0063] The image processing module includes an image acquisition unit, a face recognition unit and an image processing unit;

[0064] The image acquisition unit is used to collect the attendance image of the clock-in personnel; the face recognition unit is used to identify the attendance image using a face recognition algorithm; the image processing unit is used to crop and standardize the attendance image according to the recognition situation to obtain a face image;

[0065] The output end of the image acquisition unit is connected to the input end of the face recognition unit; the output end of the face recognition unit is connected to the input end of the image processing unit; the output end of the image processing unit is connected to the input end of the standard determination module.

[0066] The standard determination module includes a feature extraction unit, a hyperplane construction unit, a distance calculation unit and a standard determination unit;

[0067] The feature extraction unit is used to use a pre-trained model to extract features from a face image and form a feature vector; the hyperplane construction unit is used to construct hyperplanes for different punching personnel to classify the feature vectors of the face images of different punching personnel; the distance calculation unit is used to calculate the distance between all feature vectors corresponding to the face image of the same punching personnel and the hyperplane corresponding to the punching personnel; the standard determination unit is used to select the feature vector corresponding to the maximum distance as the feature vector standard;

[0068] The output end of the feature extraction unit is connected to the input end of the hyperplane construction unit; the output end of the hyperplane construction unit is connected to the input end of the distance calculation unit; the output end of the distance calculation unit is connected to the input end of the standard determination unit; the output end of the standard determination unit is connected to the input end of the sample matching module.

[0069] The sample matching module includes a new sample recognition unit, a similarity calculation unit and a face matching unit;

[0070] The new sample recognition unit is used to perform face recognition on the newly collected attendance image of the punching person to obtain the feature vector of the newly collected sample; the similarity calculation unit is used to calculate the cosine similarity between the new sample and all feature vector standards; the face matching unit is used to set a threshold and select the punching person corresponding to the feature vector standard with the highest cosine similarity value and meeting the threshold range as the face matching result;

[0071] The output end of the new sample recognition unit is connected to the input end of the similarity calculation unit; the output end of the similarity calculation unit is connected to the input end of the face matching unit; the output end of the face matching unit is connected to the input end of the standard update module.

[0072] The standard update module includes an image preprocessing unit, a timing model unit and a standard update unit;

[0073] The image preprocessing unit is used to obtain feature vectors and perform standardization processing; the time series model unit is used to build a time series model to analyze the trend of feature vector data changing over time; the standard updating unit is used to update the feature vector standard;

[0074] The output end of the image preprocessing unit is connected to the input end of the timing model unit; the output end of the timing model unit is connected to the input end of the standard updating unit.

[0075] An intelligent data management method for smart attendance, the method comprising the following steps:

[0076] Step 1: Collect attendance images of clock-in personnel and perform face recognition to obtain face images;

[0077] Step 2, extract the features of the face image and form a feature vector standard;

[0078] Step 3: Perform face recognition and feature vector matching on the newly collected attendance images of clock-in personnel;

[0079] Step 4: Update the feature vector standard according to the changes in the facial image feature vectors of the people who punch in at different times.

[0080] In step 1, the clock-in personnel are marked as: [D1, D2, …, D m ]; For clock-in personnel D i , collect its attendance images and mark them as: [K1,K2,…,K H(Di) ];

[0081] Among them, m is a positive integer, which indicates the number of clock-in personnel, D m represents the mth collector; i∈{1,2,…,m}, represents the sequence of clock-in personnel, D i represents the i-th clock-in person; H(Di) represents the number of attendance images of the i-th clock-in person; K H(Di) represents the H(Di)th attendance image of the i-th clock-in person;

[0082] For the attendance image K of the clock-in personnel Di j, perform face recognition on it: use the face detection algorithm HaarCascades to identify the face area in the attendance image; according to the face detection result, crop the face area from the attendance image; adjust and re-crop the cropped face area, standardize and normalize the face area, and obtain a face image P with a fixed size and a fixed pixel range. j ;

[0083] Where j is a positive integer, j∈{1,2,…,H(Di)}, representing the i-th collected personnel attendance image sequence, K j represents the jth attendance image of the i-th collected person; P j Represents the attendance image K j The corresponding face image;

[0084] Perform face recognition on all attendance images to obtain face images.

[0085] In step 2, for the clock-in person D i The face image P j , use the pre-trained deep learning model VGGFace to extract features from face images and obtain a feature vector of fixed length: F i,j =[F1,F2,…,F n ]; Standardize the feature vector;

[0086] Among them, F i,j Indicates the clock-in person D i The face image P j Corresponding feature vector; n is a positive integer, indicating the number of features, F n represents the nth feature;

[0087] Perform feature extraction on each face image to obtain the corresponding feature vector;

[0088] For clock-in personnel D i H(Di) groups of feature vectors corresponding to the face image are used to construct the hyperplane segmentation D i The facial image feature vector of and the facial image feature vectors of other clock-in personnel are:

[0089] The punch-in person D i The feature vector of is marked as positive, and the feature vectors of other clock-in personnel are marked as negative;

[0090] The SVM classification model is used to construct a segmentation hyperplane; the hyperplane is represented as: ω T x+b=0;

[0091] Among them, ω represents the normal vector, b is the bias, and x represents the input feature vector;

[0092] The model is trained based on the collected feature vectors, and the parameters ω and b are adjusted by minimizing the classification error and maximizing the margin between the two classes;

[0093] Calculate the eigenvector F i,j The distance from the hyperplane: d(F i,j )=|ω T ·F i,j +b| / ||ω||;

[0094] Among them, |ω T ·F i,j +b| is the eigenvector F i,j The algebraic distance to the hyperplane represents the projection of the eigenvector in the direction of the hyperplane normal vector; ||ω|| is the modulus of the normal vector, representing the length of the normal vector of the hyperplane;

[0095] Calculate the number of clock-in personnel D i The distance between the eigenvectors corresponding to all face images and the hyperplane, the eigenvector corresponding to the maximum distance is selected as the eigenvector standard;

[0096] Selecting the feature vector with the largest distance as the standard can ensure that the selected feature vector best represents the characteristics of the person punching the card. This feature vector has the largest separation from the hyperplane in the feature space and can better reflect the identity characteristics of the person punching the card.

[0097] Construct hyperplanes for the facial image feature vectors of all clock-in personnel, and calculate the feature vector standards corresponding to all clock-in personnel.

[0098] In step 3, face recognition is performed on the newly collected attendance image of the clock-in personnel to obtain the feature vector of the newly collected sample; a threshold A is set, and the cosine similarity between the feature vector of the newly collected sample and the standard feature vector of each clock-in personnel is calculated respectively: Sim = (F new ·F std ) / (||F new ||||F std ||); select the check-in personnel with the highest cosine similarity value and the feature vector standard that meets the threshold range as the face matching result; when the similarity does not meet the threshold, it is indicated as a matching anomaly;

[0099] Among them, F new Represents the feature vector of the newly collected sample; F std Denotes the eigenvector norm.

[0100] In step 4, for the clock-in person D i , record its t1~t rAttendance image data at each moment; face recognition is performed to obtain face image data, and feature extraction is performed to obtain feature vectors; the feature vectors are standardized to obtain feature vectors [F i (t1),F i (t2),...,F i (t r )];

[0101] Among them, r is a positive integer, which indicates the number of attendance times of clock-in personnel, t r Indicates the attendance of the clock-in person for the rth time; F i (t r ) represents the facial image feature vector corresponding to the r-th attendance check-in of the clock-in staff;

[0102] Build an LSTM time series model:

[0103] Determine the model parameters: input feature vector dimension n, number of hidden layer units h, learning rate α;

[0104] Construct the LSTM network structure: input layer, receiving standardized feature vectors; LSTM layer, capturing the long-term dependencies of time series through LSTM units; fully connected layer, mapping the output of LSTM layer to predicted values;

[0105] Use mean squared error as the loss function: Select the optimization algorithm Adam to train the model and minimize the loss function; input the standardized feature vector into the LSTM model, perform forward propagation, and calculate the output; calculate the loss and perform back propagation to update the model parameters;

[0106] Where u is a positive integer, u∈{1,2,...,r}, representing a sequence of moments; f u Indicates t u The real eigenvector at time f u,v Indicates t u The predicted feature vector at time instant;

[0107] Update the eigenvector norm:

[0108] Among them, F i,update Indicates the clock-in person D i Updated eigenvector criterion; F i,std Indicates the clock-in person D i The standard of the eigenvector before updating; β is the smoothing factor;

[0109] Update the facial image feature vector standards for all clock-in personnel.

[0110] In the case where features may change over time, the predicted feature vector can help capture the changing trend of the feature vector of the clock-in person. Through model prediction, potential changes can be taken into account, not just relying on past collected data.

[0111] In this embodiment:

[0112] Step 1: Mark the clock-in personnel as [employee 1, employee 2, employee 3]; for employee 2, collect his attendance image and mark it as [attendance image 1, attendance image 2, attendance image 3];

[0113] For employee 2's attendance image 1, face recognition is performed on it: the face detection algorithm Haar Cascades is used to identify the face area in the attendance image; according to the face detection result, the face area is cropped from the attendance image; the cropped face area is adjusted and re-cropped, and the face area is standardized and normalized to obtain a face image 1 with a fixed size and a fixed pixel range;

[0114] Perform face recognition on all attendance images to obtain face images.

[0115] Step 2: for employee 2's face image 1, use the pre-trained deep learning model VGGFace to extract features from the face image to obtain a fixed-length feature vector of employee 2's face image 1; and standardize the feature vector;

[0116] Perform feature extraction on each face image to obtain the corresponding feature vector;

[0117] For the three sets of feature vectors corresponding to employee 2's facial image, a hyperplane is constructed to separate the feature vector of employee 2's facial image from the feature vectors of other clock-in personnel's facial images:

[0118] Mark the feature vector of employee 2 as positive, and the feature vectors of other clock-in employees as negative.

[0119] Use the SVM classification model to construct the segmentation hyperplane;

[0120] The model is trained based on the collected feature vectors, and the parameters are adjusted by minimizing the classification error and maximizing the margin between the two classes;

[0121] Calculate the distance between the fixed-length feature vector of employee 2’s face image 1 and the hyperplane;

[0122] Calculate the distance between the eigenvectors corresponding to all the facial images of employee 2 and the hyperplane, and select the eigenvector corresponding to the maximum distance as the eigenvector standard;

[0123] Selecting the feature vector with the largest distance as the standard can ensure that the selected feature vector best represents the characteristics of the person punching the card. This feature vector has the largest separation from the hyperplane in the feature space and can better reflect the identity characteristics of the person punching the card.

[0124] Construct hyperplanes for the facial image feature vectors of all clock-in personnel, and calculate the feature vector standards corresponding to all clock-in personnel.

[0125] Step 3, perform face recognition on the newly collected attendance image of the punching personnel to obtain the feature vector of the newly collected sample; set the threshold, and calculate the cosine similarity between the feature vector of the newly collected sample and the feature vector standard of each punching personnel; select the punching personnel corresponding to the feature vector standard with the highest cosine similarity value and meeting the threshold range as the face matching result; when the similarity does not meet the threshold, it is indicated as a matching anomaly;

[0126] Step 4: for employee 2, record the attendance image data from time t1 to t5; perform face recognition to obtain face image data, perform feature extraction to obtain feature vectors; perform standardization on the feature vectors to obtain the feature vectors of employee 2's face image from time t1 to t5;

[0127] Constructing LSTM time series model: Determine model parameters; Construct LSTM network structure: Input layer, receiving standardized feature vectors; LSTM layer, capturing long-term dependencies of time series through LSTM units; Fully connected layer, mapping the output of LSTM layer to predicted values;

[0128] Use mean square error as the loss function; select the optimization algorithm Adam to train the model and minimize the loss function; input the standardized feature vector into the LSTM model, perform forward propagation, and calculate the output; calculate the loss and perform back propagation to update the model parameters;

[0129] Update the feature vector standard according to the formula; update the facial image feature vector standard of all clock-in personnel.

[0130] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. An intelligent data management system for smart attendance, characterized by: The system includes: an image processing module, a standard determination module, a sample matching module and a standard updating module; The image processing module is used to collect attendance images of clock-in personnel and perform face recognition to obtain face images; the standard determination module is used to extract features of face images and form feature vector standards; the sample matching module is used to perform face recognition and feature vector matching on newly collected attendance images of clock-in personnel; the standard update module is used to periodically update the feature vector standards according to changes in feature vectors of face images of clock-in personnel at different times; The output end of the image processing module is connected to the input end of the standard determination module; the output end of the standard determination module is connected to the input end of the sample matching module; the output end of the sample matching module is connected to the input end of the standard updating module.

2. According to claim 1, an intelligent data management system for smart attendance, characterized in that: The image processing module includes an image acquisition unit, a face recognition unit and an image processing unit; The image acquisition unit is used to collect the attendance image of the clock-in personnel; the face recognition unit is used to identify the attendance image using a face recognition algorithm; the image processing unit is used to crop and standardize the attendance image according to the recognition situation to obtain a face image; The output end of the image acquisition unit is connected to the input end of the face recognition unit; the output end of the face recognition unit is connected to the input end of the image processing unit; the output end of the image processing unit is connected to the input end of the standard determination module.

3. According to claim 2, an intelligent data management system for smart attendance, characterized in that: The standard determination module includes a feature extraction unit, a hyperplane construction unit, a distance calculation unit and a standard determination unit; The feature extraction unit is used to use a pre-trained model to extract features from a face image and form a feature vector; the hyperplane construction unit is used to construct hyperplanes for different punching personnel to classify the feature vectors of the face images of different punching personnel; the distance calculation unit is used to calculate the distance between all feature vectors corresponding to the face image of the same punching personnel and the hyperplane corresponding to the punching personnel; the standard determination unit is used to select the feature vector corresponding to the maximum distance as the feature vector standard; The output end of the feature extraction unit is connected to the input end of the hyperplane construction unit; the output end of the hyperplane construction unit is connected to the input end of the distance calculation unit; the output end of the distance calculation unit is connected to the input end of the standard determination unit; the output end of the standard determination unit is connected to the input end of the sample matching module.

4. According to claim 3, an intelligent data management system for smart attendance is characterized in that: The sample matching module includes a new sample recognition unit, a similarity calculation unit and a face matching unit; The new sample recognition unit is used to perform face recognition on the newly collected attendance image of the punching person to obtain the feature vector of the newly collected sample; the similarity calculation unit is used to calculate the cosine similarity between the new sample and all feature vector standards; the face matching unit is used to set a threshold and select the punching person corresponding to the feature vector standard with the highest cosine similarity value and meeting the threshold range as the face matching result; The output end of the new sample recognition unit is connected to the input end of the similarity calculation unit; the output end of the similarity calculation unit is connected to the input end of the face matching unit; the output end of the face matching unit is connected to the input end of the standard update module.

5. According to claim 4, an intelligent data management system for smart attendance, characterized in that: The standard update module includes an image preprocessing unit, a timing model unit and a standard update unit; The image preprocessing unit is used to obtain feature vectors and perform standardization processing; the time series model unit is used to build a time series model to analyze the trend of feature vector data changing over time; the standard updating unit is used to update the feature vector standard; The output end of the image preprocessing unit is connected to the input end of the timing model unit; the output end of the timing model unit is connected to the input end of the standard updating unit.

6. An intelligent data management method for smart attendance, characterized in that: The method comprises the following steps: Step 1: Collect attendance images of clock-in personnel and perform face recognition to obtain face images; Step 2, extract the features of the face image and form a feature vector standard; Step 3: Perform face recognition and feature vector matching on the newly collected attendance images of clock-in personnel; Step 4: Update the feature vector standard according to the changes in the facial image feature vectors of the people who punch in at different times.

7. The intelligent data management method for smart attendance according to claim 6 is characterized in that: In step 1, the clock-in personnel are marked as: [D1, D2, …, D m ]; For clock-in personnel D i , collect its attendance images and mark them as: [K1,K2,…,K H(Di) ]; Among them, m is a positive integer, which indicates the number of clock-in personnel, D m represents the mth collector; i∈{1,2,…,m}, represents the sequence of clock-in personnel, D i represents the i-th clock-in person; H(Di) represents the number of attendance images of the i-th clock-in person; K H(Di) represents the H(Di)th attendance image of the i-th clock-in person; For the attendance image K of the clock-in personnel Di j , perform face recognition on it: use the face detection algorithm Haar Cascades to identify the face area in the attendance image; according to the face detection result, crop the face area from the attendance image; adjust and re-crop the cropped face area, standardize and normalize the face area, and obtain a face image P with a fixed size and a fixed pixel range. j ; Where j is a positive integer, j∈{1,2,…,H(Di)}, representing the i-th collected personnel attendance image sequence, K j represents the jth attendance image of the i-th collected person; P j Represents the attendance image K j The corresponding face image; Perform face recognition on all attendance images to obtain face images.

8. The intelligent data management method for smart attendance according to claim 7 is characterized in that: In step 2, for the clock-in person D i The face image P j , use the pre-trained deep learning model VGGFace to extract features from face images and obtain a feature vector of fixed length: F i,j =[F1,F2,…,F n ]; Standardize the feature vector; Among them, F i,j Indicates the clock-in person D i The face image P j Corresponding feature vector; n is a positive integer, indicating the number of features, F n represents the nth feature; Perform feature extraction on each face image to obtain the corresponding feature vector; For clock-in personnel D i H(Di) groups of feature vectors corresponding to the face image are used to construct the hyperplane segmentation D i The facial image feature vector of and the facial image feature vectors of other clock-in personnel are: The punch-in person D i The feature vector of is marked as positive, and the feature vectors of other clock-in personnel are marked as negative; The SVM classification model is used to construct a segmentation hyperplane; the hyperplane is represented as: ω T x+b=0; Among them, ω represents the normal vector, b is the bias, and x represents the input feature vector; The model is trained based on the collected feature vectors, and the parameters ω and b are adjusted by minimizing the classification error and maximizing the margin between the two classes; Calculate the eigenvector F i,j The distance from the hyperplane: d(F i,j )=|ω T ·F i,j +b| / ||ω||; Among them, |ω T ·F i,j +b| is the eigenvector F i,j The algebraic distance to the hyperplane represents the projection of the eigenvector in the direction of the hyperplane normal vector; ||ω|| is the modulus of the normal vector, representing the length of the normal vector of the hyperplane; Calculate the number of clock-in personnel D i The distance between the eigenvectors corresponding to all face images and the hyperplane, the eigenvector corresponding to the maximum distance is selected as the eigenvector standard; Construct hyperplanes for the facial image feature vectors of all clock-in personnel, and calculate the feature vector standards corresponding to all clock-in personnel.

9. The intelligent data management method for smart attendance according to claim 8, characterized in that: In step 3, face recognition is performed on the newly collected attendance image of the clock-in personnel to obtain the feature vector of the newly collected sample; a threshold A is set, and the cosine similarity between the feature vector of the newly collected sample and the standard feature vector of each clock-in personnel is calculated respectively: Sim = (F new ·F std ) / (||F new ||||F std ||); select the check-in personnel with the highest cosine similarity value and the feature vector standard that meets the threshold range as the face matching result; when the similarity does not meet the threshold, it is indicated as a matching anomaly; Among them, F new Represents the feature vector of the newly collected sample; F std Denotes the eigenvector norm.

10. The intelligent data management method for smart attendance according to claim 9 is characterized in that: In step 4, for the clock-in person D i , record its t1~t r Attendance image data at all times; face recognition is performed to obtain face image data, and feature extraction is performed to obtain feature vectors; The eigenvector is normalized to obtain the eigenvector [F i (t1),F i (t2),...,F i (t r )]; Among them, r is a positive integer, which indicates the number of attendance times of clock-in personnel, t r Indicates the attendance of the clock-in person for the rth time; F i (t r ) represents the facial image feature vector corresponding to the r-th attendance check-in of the clock-in staff; Build an LSTM time series model: Determine the model parameters: input feature vector dimension n, number of hidden layer units h, learning rate α; Construct the LSTM network structure: input layer, receiving standardized feature vectors; LSTM layer, capturing the long-term dependencies of time series through LSTM units; fully connected layer, mapping the output of LSTM layer to predicted values; Use mean squared error as the loss function: Select the optimization algorithm Adam to train the model and minimize the loss function; input the standardized feature vector into the LSTM model, perform forward propagation, and calculate the output; calculate the loss and perform back propagation to update the model parameters; Where u is a positive integer, u∈{1,2,...,r}, representing a sequence of moments; f u Indicates t u The real eigenvector at time f u,v Indicates t u The predicted feature vector at time instant; Update the eigenvector norm: Among them, F i,update Indicates the clock-in person D i Updated eigenvector criterion; F i,std Indicates the clock-in person D i The standard of the eigenvector before updating; β is the smoothing factor; Update the facial image feature vector standards for all clock-in personnel.

Images