WO2022105118A1 - 基于图像的健康状态识别方法、装置、设备及存储介质 - Google Patents

基于图像的健康状态识别方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2022105118A1
WO2022105118A1 PCT/CN2021/090417 CN2021090417W WO2022105118A1 WO 2022105118 A1 WO2022105118 A1 WO 2022105118A1 CN 2021090417 W CN2021090417 W CN 2021090417W WO 2022105118 A1 WO2022105118 A1 WO 2022105118A1
Authority
WO
WIPO (PCT)
Prior art keywords
recognition model
feature
initial
training
model
Prior art date
Application number
PCT/CN2021/090417
Other languages
English (en)
French (fr)
Inventor
陈昊
Original Assignee
平安科技（深圳）有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 平安科技（深圳）有限公司 filed Critical 平安科技（深圳）有限公司
Publication of WO2022105118A1 publication Critical patent/WO2022105118A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
    • G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation

Definitions

  • the present application belongs to the technical field of artificial intelligence, and specifically relates to an image-based health state identification method, device, computer equipment and storage medium.
  • the realization of facial health status recognition technology mostly focuses on making judgments about whether customers are healthy and awake based on the customer's physiological signals, whole body images, facial images, and video influences.
  • the user's physical parameters such as blood pressure
  • the user's health status is comprehensively judged based on the user's physical parameters; Physical parameters need to use additional equipment to judge the user's health status.
  • the customer experience is often poor, which is not conducive to system integration, and the conditions for use are relatively strict.
  • the purpose of the embodiments of the present application is to propose an image-based health status recognition method, device, computer equipment and storage medium, so as to solve the problem that the existing face health status recognition technology has poor customer experience and cannot be adapted to mobile terminal services. technical issues.
  • the embodiments of the present application provide an image-based health state identification method, which adopts the following technical solutions:
  • An image-based health state identification method comprising:
  • the embodiments of the present application also provide an image-based health state identification device, which adopts the following technical solutions:
  • An image-based health state identification device comprising:
  • a first model training module configured to obtain a first training sample set from a preset database, and train a preset initial feature recognition model through the first training sample set to obtain a face feature recognition model;
  • the second model training module is used to collect the training results of the facial feature recognition model, and based on the training results, the initial relationship recognition model is trained to obtain a feature relationship recognition model, and a correction matrix is generated based on the feature relationship recognition model;
  • a facial feature recognition module used for acquiring the image to be recognized, and identifying the image to be recognized through a facial feature recognition model to obtain the facial features of the user in the image to be recognized;
  • the health status recognition module is used to correct the facial features of the image to be recognized based on the correction matrix, and identify the health status of the user according to the correction result.
  • the embodiment of the present application also provides a computer device, which adopts the following technical solutions:
  • a computer device includes a memory and a processor, wherein computer-readable instructions are stored in the memory, and the processor implements the following image-based health state identification method when executing the computer-readable instructions:
  • the embodiments of the present application also provide a computer-readable storage medium, which adopts the following technical solutions:
  • a computer-readable storage medium where computer-readable instructions are stored on the computer-readable storage medium, and when the computer-readable instructions are executed by a processor, the following image-based health state identification method is implemented:
  • the present application discloses an image-based health state recognition method, device, equipment and storage medium, which belong to the field of artificial intelligence. Generate a correction matrix for the facial feature recognition model.
  • feature extraction is performed through a pre-built face feature recognition model to obtain the user's facial features, and then the correction matrix is used to correct the user's facial feature parameters, and finally the facial features are corrected according to the correction results. to identify the health status.
  • the recognition accuracy of the face health state is improved, and the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources, which is conducive to reducing the system computing pressure and system integration pressure, and has strong adaptability. Easy to deploy in mobile terminals.
  • FIG. 1 shows an exemplary system architecture diagram to which the present application can be applied
  • FIG. 2 shows a flowchart of an embodiment of an image-based health state identification method according to the present application
  • Fig. 3 shows a flowchart of a specific implementation manner of step S201 in Fig. 2;
  • Fig. 4 shows a flowchart of a specific implementation manner of step S202 in Fig. 2;
  • FIG. 5 shows a schematic structural diagram of an embodiment of an image-based health state identification device according to the present application
  • FIG. 6 shows a schematic structural diagram of an embodiment of a computer device according to the present application.
  • the system architecture 100 may include terminal devices 101 , 102 , and 103 , a network 104 and a server 105 .
  • the network 104 is a medium used to provide a communication link between the terminal devices 101 , 102 , 103 and the server 105 .
  • the network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, among others.
  • the user can use the terminal devices 101, 102, 103 to interact with the server 105 through the network 104 to receive or send messages and the like.
  • Various communication client applications may be installed on the terminal devices 101 , 102 and 103 , such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social platform software, and the like.
  • the terminal devices 101, 102, and 103 can be various electronic devices that have a display screen and support web browsing, including but not limited to smart phones, tablet computers, e-book readers, MP3 players (Moving Picture Experts Group Audio Layer III, dynamic Picture Experts Compression Standard Audio Layer 3), MP4 (Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Compression Standard Audio Layer 4) Players, Laptops and Desktops, etc.
  • MP3 players Moving Picture Experts Group Audio Layer III, dynamic Picture Experts Compression Standard Audio Layer 3
  • MP4 Moving Picture Experts Group Audio Layer IV, Moving Picture Experts Compression Standard Audio Layer 4
  • the server 105 may be a server that provides various services, such as a background server that provides support for the pages displayed on the terminal devices 101 , 102 , and 103 .
  • the image-based health state identification method provided by the embodiments of the present application is generally performed by a server/terminal device, and accordingly, an image-based health state identification apparatus is generally set in the server/terminal device.
  • terminal devices, networks and servers in FIG. 1 are merely illustrative. There can be any number of terminal devices, networks and servers according to implementation needs.
  • the image-based health state identification method includes the following steps:
  • S201 Obtain a first training sample set from a preset database, and train a preset initial feature recognition model through the first training sample set to obtain a face feature recognition model.
  • eye state such as whether it is puffy, sunken, etc.
  • cheek state such as puffiness, sunken, etc.
  • skin state such as drunkenness and redness
  • sample data is obtained from a preset database.
  • the sample data are face images of different users.
  • the sample data is marked based on the above-mentioned facial features, and the marked sample data is combined to obtain the first
  • the training sample set is used to train a preset initial feature recognition model through the first training sample set to obtain a face feature recognition model.
  • the preset initial feature recognition model is a deep convolutional neural network model, such as a CNN model, and a convolutional neural network (Convolutional Neural Networks, CNN) is a type of feedforward neural network (Feedforward Neural Networks) is one of the representative algorithms of deep learning.
  • Convolutional neural network has the ability of representation learning and can perform shift-invariant classification of input information according to its hierarchical structure, so it is also called "shift-invariant artificial neural network”.
  • Convolutional neural network is constructed by imitating the visual perception mechanism of biology, which can perform supervised learning and unsupervised learning. Small computational effort to learn grid-like topology features, such as pixels and audio, with stable results and no additional feature engineering requirements on the data.
  • S202 Collect the training results of the facial feature recognition model, and train the initial relationship recognition model based on the training results to obtain a feature relationship recognition model, and generate a correction matrix based on the feature relationship recognition model.
  • the training results output during the training of the face feature recognition model are collected, the training results of the face feature recognition model are marked, and the marked training results of the face feature recognition model are combined to obtain a second training sample set,
  • the initial relationship recognition model is trained through the second training sample set, and the trained feature relationship recognition model is obtained.
  • the trained feature relationship recognition model is verified until the model is fitted, and the output result of the feature relationship recognition model is obtained.
  • a correction matrix is generated from the output result of , wherein the output result of the feature relation recognition model is in the form of a matrix.
  • the initial relationship recognition model is a graph convolutional neural network model, such as a GCN model, and a graph convolutional network (Graph Convolutional Network, GCN) is a neural network structure that has gradually become popular in recent years.
  • GCN Graph Convolutional Network
  • LSTM and CNN which can only be used for grid-based data
  • graph convolutional networks can process data with generalized topological graph structure and deeply explore its characteristics and laws, such as PageRank reference network, A series of irregular data with spatial topology map structure, such as social network, communication network, protein molecular structure, etc.
  • GCN has astonishingly designed a method of extracting features from graph data, which can perform end-to-end learning of node feature information and structural information at the same time. It is currently the best choice for graph data learning tasks, and graph convolution is extremely applicable. It is widely applicable to nodes and graphs of any topology.
  • S203 Acquire an image to be recognized, and identify the image to be recognized through a facial feature recognition model to obtain the facial features of the user in the image to be recognized.
  • the user's to-be-recognized image is acquired, and the to-be-recognized image is recognized by a trained facial feature recognition model to obtain the user's facial features.
  • a trained facial feature recognition model to obtain the user's facial features.
  • the electronic device (for example, the server/terminal device shown in FIG. 1 ) on which the image-based health state identification method runs may receive the health state identification instruction through a wired connection or a wireless connection.
  • the above wireless connection methods may include but are not limited to 3G/4G connection, WiFi connection, Bluetooth connection, WiMAX connection, Zigbee connection, UWB (ultra wideband) connection, and other wireless connection methods currently known or developed in the future .
  • multiple facial features output by the facial feature recognition model are obtained, wherein the output of the facial feature recognition model is in the form of a matrix, and the correction matrix is used to perform an inner product operation with the output results of the facial feature recognition model to complete the treatment process. Identify multiple facial features of the image for correction, wherein the weight matrix can be considered as a tool for correcting facial features. Then sigmoid processing is performed on the correction result, and the sigmoid processing result is output. The sigmoid processing result is a specific value, and the user's health status is judged by the sigmoid processing result.
  • model training is performed in combination with the CNN deep learning convolutional network model and the GCN graph convolutional network model to obtain a face feature recognition model and a feature relationship recognition model, and based on the feature relationship recognition model, a face is generated.
  • Correction matrix for the feature recognition model is generated.
  • the trained facial feature recognition model and correction matrix can be deployed to the mobile terminal.
  • the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources. It is beneficial to reduce the pressure of system operation and system integration, has strong adaptability, and is convenient for deployment in mobile terminals.
  • the present application discloses an image-based health state recognition method, device, equipment and storage medium, which belong to the field of artificial intelligence. Generate a correction matrix for the facial feature recognition model.
  • feature extraction is performed through a pre-built face feature recognition model to obtain the user's facial features, and then the correction matrix is used to correct the user's facial feature parameters, and finally the facial features are corrected according to the correction results. to identify the health status.
  • the recognition accuracy of the face health state is improved, and the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources, which is conducive to reducing the system computing pressure and system integration pressure, and has strong adaptability. Easy to deploy in mobile terminals.
  • FIG. 3 shows a flowchart of a specific implementation of step S201 in FIG. 2, obtaining a first training sample set from a preset database, and training initial feature recognition through the first training sample set model, and the steps of obtaining a face feature recognition model include:
  • sample data is obtained from a preset database.
  • the sample data are face images of different users
  • the sample data is labeled based on the facial features of the sample data
  • the labeled sample data is randomly combined to obtain
  • the first training sample set and the first verification data set for example, the labeled sample data can be randomly divided into 10 equal sample subsets, wherein 9 sample subsets are randomly combined as the first training sample set, and the remaining samples are divided into 10 equal sample subsets.
  • the subset serves as the first validation dataset. Import the first training sample set into the initial feature recognition model for model training to obtain the initial face feature recognition model, verify the initial face feature recognition model through the first verification data set, and output the verified face feature recognition model .
  • the facial feature recognition model can be quickly obtained. Model.
  • the steps of verifying the initial facial feature recognition model through the first verification data set and outputting the verified facial feature recognition model specifically include:
  • the backpropagation algorithm that is, the error backpropagation algorithm (Backpropagation algorithm, BP algorithm) is a learning algorithm suitable for multi-layer neuron networks. It is based on the gradient descent method and is used for the error of deep learning networks. calculate.
  • the input and output relationship of BP network is essentially a mapping relationship: the function completed by a BP neural network with n input and m output is a continuous mapping from n-dimensional Euclidean space to a finite field in m-dimensional Euclidean space. A map is highly nonlinear.
  • the learning process of BP algorithm consists of forward propagation process and back propagation process.
  • the input information is processed layer by layer through the hidden layer through the input layer and transmitted to the output layer, and then transferred to the back propagation, and the partial derivative of the objective function to the weight of each neuron is obtained layer by layer, which constitutes The gradient of the objective function to the weight vector is used as the basis for modifying the weight.
  • the first verification data set is imported into the initial face feature recognition model, and the recognition result is output; based on the recognition result and the first preset standard result, a back-propagation algorithm is used to perform fitting calculation to obtain the recognition error; Compared with the first preset error threshold, if the recognition error is greater than the first preset error threshold, the initial facial feature recognition model is iteratively updated based on the loss function of the initial facial feature recognition model, until the recognition error is less than or equal to the first. Until a preset error threshold is reached, the facial feature recognition model that has passed the verification is obtained.
  • the first preset standard result and the first preset error threshold may be set in advance.
  • the face feature recognition model is verified and iterated through the back-propagation algorithm to obtain a face feature recognition model that meets the requirements.
  • Fig. 4 shows a flowchart of a specific implementation of step S202 in Fig. 2, collecting the training results of the facial feature recognition model, and training the initial relationship recognition model based on the training results,
  • the steps of obtaining a feature relationship identification model and generating a correction matrix based on the feature relationship identification model include:
  • the training result of the facial feature recognition model is marked to obtain the label of the health status of the face.
  • the label of the health status of the face is such as flushing-drunk, swollen eyes-lack of rest, Dark circles - lack of rest, etc.
  • Joint distribution probability is referred to as joint distribution, which is the probability distribution of random vectors composed of two or more random variables.
  • the weight value of the influence of each health state label on the health state of the face is represented by the joint distribution probability. .
  • the first training sample set is imported into the face feature recognition model for training, each training result of the face feature recognition model is collected, and the obtained training results are marked to obtain a health status label, such as a training result showing If the user's face is flushed, it is marked as drunk, and if a training result shows dark circles around the user's eyes, it is marked as lack of rest.
  • a health status label such as a training result showing If the user's face is flushed, it is marked as drunk, and if a training result shows dark circles around the user's eyes, it is marked as lack of rest.
  • the preset initial relationship recognition model is trained according to the second training sample set, the feature relationship recognition model is obtained, the output result of the feature relationship recognition model is obtained, and the correction matrix is generated based on the output result.
  • the second training sample set and the second verification data set are obtained, and the initial relationship recognition model is paired by the second training sample set and the second verification data set.
  • a feature relationship recognition model is obtained, and a correction matrix is generated based on the output result of the feature relationship recognition model, and the feature relationship recognition model and correction matrix can be quickly obtained.
  • the method before the step of combining the health state labels based on the joint distribution probability to obtain the second training sample set and the second verification data set, the method further includes:
  • the health status labels are vectorized by the work2vec tool.
  • word2vec is a tool for word vector calculation.
  • word2vec can efficiently train on millions of dictionaries and hundreds of millions of data sets.
  • the training result obtained by word2vec - word embedding, can be very good. Measure the similarity between words.
  • Behind the word2vec algorithm is a shallow neural network.
  • word2vec is an open source tool for computing word vectors.
  • each health state label is vectorized by the work2vec tool to obtain the feature vector of each health state label, and the feature vectors are randomly combined according to the joint distribution probability to obtain the second training sample set and the second verification data set, Both the second training sample set and the second verification data set are data sets in the form of vectors.
  • an output result in the form of a matrix is obtained, so as to obtain a correction matrix.
  • steps of training the preset initial relationship recognition model according to the second training sample set to obtain the feature relationship recognition model specifically include:
  • the second training sample set is imported into the initial relationship recognition model for training, the initial feature relationship recognition model is obtained, the second verification data set is imported into the initial feature relationship recognition model for verification, and the prediction result is output.
  • the second preset standard result use the back-propagation algorithm to perform fitting calculation, obtain the prediction error, and compare the prediction error with the second preset error threshold. If the prediction error is greater than the second preset error threshold, based on the initial feature relationship
  • the loss function of the recognition model iteratively updates the initial feature relationship recognition model until the prediction error is less than or equal to the second preset error threshold, and the verified feature relationship recognition model is obtained.
  • the second preset standard result and the second preset error threshold may be set in advance.
  • the feature relationship identification model is verified and iterated through the back-propagation algorithm to obtain a feature relationship identification model that meets the requirements.
  • the initial feature relationship identification model is iteratively updated until the prediction error is less than or equal to the second preset error threshold, and the step of acquiring the verified feature relationship identification model specifically includes:
  • the loss function of the initial feature relationship recognition model is constructed, and the parameters of the feature relationship recognition model are optimized based on the constructed loss function until the loss function of the feature relationship recognition model reaches the minimum value.
  • the loss function L of the feature relationship recognition model is as follows:
  • c refers to the label number
  • y represents the true label value
  • ⁇ represents the sigmoid function
  • the initial feature relationship identification model is iteratively updated by constructing a loss function of the initial feature relationship identification model, and optimizing the parameters of the feature relationship identification model based on the constructed loss function.
  • the steps of obtaining an image to be recognized, and identifying the image to be recognized through a facial feature recognition model, to obtain the facial features of the user in the image to be recognized specifically include:
  • the facial features of the user in the to-be-recognized image are obtained by performing feature recognition on the face region in the to-be-recognized image through the facial feature recognition model.
  • the to-be-recognized image is acquired, and the to-be-recognized image is input into the trained facial feature recognition model to acquire the user's facial features in the to-be-recognized image.
  • the face area in the image to be recognized is recognized by the SSD model.
  • the image to be recognized is normalized, and a pre-trained SSD (Single Shot MultiBox Detector, general object detection) model is used to extract the face region.
  • the coordinates of the upper left and lower right points of the face frame can be extracted from the SSD model, and the face area is cut out from the original image according to the extracted coordinates, and scaled to 448x448 size, and then the scaled face area is Input to the facial feature recognition model to recognize the image to be recognized, and obtain the facial features of the user.
  • SSD Single Shot MultiBox Detector, general object detection
  • a pre-built SSD model is used to intercept the face region of the image to be detected, and then a pre-built face feature recognition model is used to extract the user's facial features to obtain a facial feature tensor. Then use the pre-built correction matrix to correct the user's face feature tensor, and finally identify the user's health status based on the correction result.
  • the facial features are corrected by a simple correction matrix, which improves the accuracy of the health state recognition, so the adaptability is strong, and the recognition process does not occupy a lot of computing resources.
  • the above image to be recognized can also be stored in a node of a blockchain.
  • the blockchain referred to in this application is a new application mode of computer technologies such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
  • Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information to verify its Validity of information (anti-counterfeiting) and generation of the next block.
  • the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
  • the aforementioned storage medium may be a non-volatile storage medium such as a magnetic disk, an optical disk, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM) or the like.
  • the present application provides an embodiment of an image-based health state identification device, and the device embodiment corresponds to the method embodiment shown in FIG. 2 , Specifically, the device can be applied to various electronic devices.
  • the apparatus for recognizing an image-based health state includes:
  • the first model training module 501 is configured to obtain a first training sample set from a preset database, and train a preset initial feature recognition model through the first training sample set to obtain a face feature recognition model;
  • the second model training module 502 is used to collect the training results of the facial feature recognition model, and based on the training results, the initial relationship recognition model is trained to obtain a feature relationship recognition model, and a correction matrix is generated based on the feature relationship recognition model;
  • the facial feature recognition module 503 is used to obtain the to-be-recognized image, and to recognize the to-be-recognized image through the facial feature recognition model to obtain the user's facial feature in the to-be-recognized image;
  • the health status recognition module 504 is configured to correct the facial features of the image to be recognized based on the correction matrix, and identify the health status of the user according to the correction result.
  • the first model training module 501 specifically includes:
  • a first labeling submodule used for obtaining sample data from a preset database, and labeling the sample data
  • the first combination submodule is used to randomly combine the labeled sample data to obtain a first training sample set and a first verification data set;
  • the first training submodule is used to import the first training sample set into the initial feature recognition model for model training to obtain the initial face feature recognition model;
  • the first verification sub-module is used to verify the initial face feature recognition model by using the first verification data set, and output the verified face feature recognition model.
  • the first verification unit specifically includes:
  • a first verification unit used for importing the first verification data set into the initial facial feature recognition model, and outputting the recognition result
  • a first error calculation unit configured to perform fitting calculation using a back-propagation algorithm based on the identification result and the first preset standard result, to obtain the identification error
  • the first iterative unit is used to compare the recognition error with the first preset error threshold, and if the recognition error is greater than the first preset error threshold, iteratively update the initial face feature recognition model until the recognition error is less than or equal to the first Until a preset error threshold is reached, the facial feature recognition model that has passed the verification is obtained.
  • the second model training module 502 specifically includes:
  • the second labeling sub-module is used to collect the training results of the face feature recognition model, and label the training results to obtain the health status label;
  • the joint distribution probability calculation sub-module is used to count the number of health state labels, and calculate the joint distribution probability of each health state label based on the number of health state labels;
  • the second combining submodule is used to combine the health status labels based on the joint distribution probability to obtain a second training sample set and a second verification data set;
  • the second training submodule is used to train the initial relationship recognition model according to the second training sample set to obtain the feature relationship recognition model
  • the correction matrix generation sub-module is used to obtain the output result of the feature relationship recognition model, and generate a correction matrix based on the output result.
  • the second training unit specifically includes:
  • the second training unit is used to import the second training sample set into the initial relationship recognition model for training, and obtain the initial feature relationship recognition model;
  • a second verification unit configured to verify the initial feature relationship identification model by using the second verification data set, and output a prediction result
  • a second error calculation unit configured to perform fitting calculation using a back-propagation algorithm based on the prediction result and the second preset standard result to obtain the prediction error
  • a second iterative unit configured to compare the prediction error with a second preset error threshold, and if the prediction error is greater than the second preset error threshold, iteratively update the initial feature relationship identification model until the prediction error is less than or equal to the second Until the preset error threshold is reached, the feature relationship recognition model that has passed the verification is obtained.
  • the second iterative unit specifically includes:
  • the loss function setting subunit is used to set the loss function of the initial feature relation recognition model
  • the second iterative subunit is used to iteratively update the initial feature relationship recognition model based on the loss function, until the output of the loss function reaches a minimum value;
  • the facial feature recognition module 503 specifically includes:
  • the face area recognition sub-module is used to obtain the image to be recognized, scan the image to be recognized, and identify the face area in the image to be recognized;
  • the facial feature recognition sub-module is used to perform feature recognition on the face region in the to-be-recognized image through the facial-feature recognition model to obtain the user's facial features in the to-be-recognized image.
  • the present application discloses an image-based health state recognition device, belonging to the field of artificial intelligence.
  • the pre-built facial feature recognition model and feature relationship recognition model are used to generate a correction matrix of the facial feature recognition model based on the feature relationship recognition model.
  • feature extraction is performed through a pre-built face feature recognition model to obtain the user's facial features, and then the correction matrix is used to correct the user's facial feature parameters, and finally the facial features are corrected according to the correction results. to identify the health status.
  • the recognition accuracy of the face health state is improved, and the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources, which is conducive to reducing the system computing pressure and system integration pressure, and has strong adaptability. Easy to deploy in mobile terminals.
  • FIG. 6 is a block diagram of the basic structure of a computer device according to this embodiment.
  • the computer device 6 includes a memory 61 , a processor 62 , and a network interface 63 that communicate with each other through a system bus. It should be pointed out that only the computer device 6 with components 61-63 is shown in the figure, but it should be understood that it is not required to implement all of the shown components, and more or less components may be implemented instead.
  • the computer device here is a device that can automatically perform numerical calculation and/or information processing according to pre-set or stored instructions, and its hardware includes but is not limited to microprocessors, special-purpose Integrated circuit (Application Specific Integrated Circuit, ASIC), programmable gate array (Field-Programmable Gate Array, FPGA), digital processor (Digital Signal Processor, DSP), embedded equipment, etc.
  • ASIC Application Specific Integrated Circuit
  • FPGA Field-Programmable Gate Array
  • DSP Digital Signal Processor
  • embedded equipment etc.
  • the computer equipment may be a desktop computer, a notebook computer, a palmtop computer, a cloud server and other computing equipment.
  • the computer device can perform human-computer interaction with the user through a keyboard, a mouse, a remote control, a touch pad or a voice control device.
  • the memory 61 includes at least one type of readable storage medium, and the readable storage medium includes flash memory, hard disk, multimedia card, card-type memory (for example, SD or DX memory, etc.), random access memory (RAM), static Random Access Memory (SRAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), Programmable Read Only Memory (PROM), Magnetic Memory, Magnetic Disk, Optical Disk, etc.
  • the memory 61 may be an internal storage unit of the computer device 6 , such as a hard disk or a memory of the computer device 6 .
  • the memory 61 may also be an external storage device of the computer device 6, such as a plug-in hard disk, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, flash memory card (Flash Card), etc.
  • the memory 61 may also include both the internal storage unit of the computer device 6 and its external storage device.
  • the memory 61 is generally used to store the operating system and various application software installed on the computer device 6 , such as computer-readable instructions for an image-based health state identification method, and the like.
  • the memory 61 can also be used to temporarily store various types of data that have been output or will be output.
  • the processor 62 may be a central processing unit (Central Processing Unit, CPU), a controller, a microcontroller, a microprocessor, or other data processing chips. This processor 62 is typically used to control the overall operation of the computer device 6 . In this embodiment, the processor 62 is configured to execute computer-readable instructions stored in the memory 61 or process data, such as computer-readable instructions for executing the image-based health state identification method.
  • CPU Central Processing Unit
  • controller central processing unit
  • microcontroller a microcontroller
  • microprocessor microprocessor
  • This processor 62 is typically used to control the overall operation of the computer device 6 .
  • the processor 62 is configured to execute computer-readable instructions stored in the memory 61 or process data, such as computer-readable instructions for executing the image-based health state identification method.
  • the network interface 63 may include a wireless network interface or a wired network interface, and the network interface 63 is generally used to establish a communication connection between the computer device 6 and other electronic devices.
  • the present application discloses computer equipment, which belongs to the field of artificial intelligence.
  • the method generates a correction matrix of the facial feature recognition model based on the feature relationship recognition model by using a pre-built face feature recognition model and a feature relationship recognition model.
  • face health status recognition feature extraction is performed through a pre-built face feature recognition model to obtain the user's facial features, and then the correction matrix is used to correct the user's facial feature parameters, and finally the facial features are corrected according to the correction results. to identify the health status.
  • the recognition accuracy of the face health state is improved, and the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources, which is conducive to reducing the system computing pressure and system integration pressure, and has strong adaptability. Easy to deploy in mobile terminals.
  • the present application also provides another implementation manner, that is, to provide a computer-readable storage medium
  • the computer-readable storage medium may be non-volatile or volatile
  • the computer-readable storage medium stores Computer readable instructions executable by at least one processor to cause the at least one processor to perform the steps of the image-based health state identification method as described above.
  • the application discloses a storage medium, which belongs to the field of artificial intelligence.
  • the method generates a correction matrix of the face feature recognition model based on the feature relationship recognition model by using a pre-built face feature recognition model and a feature relationship recognition model.
  • face health status recognition feature extraction is performed through a pre-built face feature recognition model to obtain the user's facial features, and then the correction matrix is used to correct the user's facial feature parameters, and finally the facial features are corrected according to the correction results. to identify the health status.
  • the recognition accuracy of the face health state is improved, and the correction of the user's facial feature parameters through the correction matrix will not take up a lot of computing resources and system resources, which is conducive to reducing the system computing pressure and system integration pressure, and has strong adaptability. Easy to deploy in mobile terminals.
  • the methods of the above embodiments can be implemented by means of software plus a necessary general hardware platform, and of course hardware can also be used, but in many cases the former is better implementation.
  • the technical solution of the present application can be embodied in the form of a software product in essence or in a part that contributes to the prior art, and the computer software product is stored in a storage medium (such as ROM/RAM, magnetic disk, CD-ROM), including several instructions to make a terminal device (which may be a mobile phone, a computer, a server, an air conditioner, or a network device, etc.) execute the methods described in the various embodiments of this application.
  • a storage medium such as ROM/RAM, magnetic disk, CD-ROM

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • General Engineering & Computer Science (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • Biomedical Technology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Multimedia (AREA)
  • Human Computer Interaction (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

一种基于图像的健康状态识别方法、装置、设备及存储介质，属于人工智能领域中的计算机视觉技术，所述方法通过第一训练样本集训练初始特征识别模型，得到人脸特征识别模型；采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，并基于特征关系识别模型生成校正矩阵；获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到用户的脸部特征；基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。此外，该方法还涉及区块链技术，待识别图像可存储于区块链中。所述方法可以显著提高人脸健康状态识别精度，同时适配性较强，方便部署于移动终端。

Description

基于图像的健康状态识别方法、装置、设备及存储介质
本申请要求于2020年11月17日提交中国专利局、申请号为202011286362.3，发明名称为“基于图像的健康状态识别方法、装置、设备及存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于人工智能技术领域，具体涉及一种基于图像的健康状态识别方法、装置、计算机设备及存储介质。
背景技术
随着人工智能在金融领域中的广泛使用，在移动端开展金融行为的场景越来越常见，而在移动端开展金融行为一般涉及到一套较为严格的审批业务，其中判定用户是否是在健康、头脑清醒的情况下做出的判断与操作，是确保送用户操作资料是否具备法律意义的前提。
目前，人脸健康状态识别技术的实现大多集中在基于客户生理信号、全身图像、脸部图像、视频影响来做出客户是否健康、是否清醒的判断。比如基于便携性设备(比如手环)获得用户的身体参数，如血压等，基于用户的身体参数综合判断用户的健康状态；但是，在进行健康状态识别过程中，发明人意识到现有的通过身体参数判断用户的健康状态，需要借助额外的设备，客户体验感往往不佳，不利于系统集成，并且使用的条件比较严苛。而通过拍摄用户一段全身视频影像或者图像，然后分析全身视频影像或者图像来对客户健康状态进行判断。而目前通过影像或者图像来对客户健康状态进行判断时，需要拍摄用户带有特定动作或者姿态的全身视频影像或者图像才能实现，而要求用户做出特定动作或者姿态客户体验感往往不佳，并且无法适应于移动端业务。
发明内容
本申请实施例的目的在于提出一种基于图像的健康状态识别方法、装置、计算机设备及存储介质，以解决现有的人脸健康状态识别技术客户体验感不佳，且无法适应于移动端业务的技术问题。
为了解决上述技术问题，本申请实施例提供一种基于图像的健康状态识别方法，采用了如下所述的技术方案：
一种基于图像的健康状态识别方法，包括：
从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵；
获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征；
基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
为了解决上述技术问题，本申请实施例还提供一种基于图像的健康状态识别装置，采用了如下所述的技术方案：
一种基于图像的健康状态识别装置，包括：
第一模型训练模块，用于从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
第二模型训练模块，用于采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵；
脸部特征识别模块，用于获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征；
健康状态识别模块，用于基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
为了解决上述技术问题，本申请实施例还提供一种计算机设备，采用了如下所述的技术方案：
一种计算机设备，包括存储器和处理器，存储器中存储有计算机可读指令，处理器执行计算机可读指令时实现如下的基于图像的健康状态识别方法：
从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵；
获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征；
基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
为了解决上述技术问题，本申请实施例还提供一种计算机可读存储介质，采用了如下所述的技术方案：
一种计算机可读存储介质，计算机可读存储介质上存储有计算机可读指令，计算机可读指令被处理器执行时实现如下的基于图像的健康状态识别方法：
从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵；
获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征；
基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
与现有技术相比，本申请实施例主要有以下有益效果：
本申请公开了一种基于图像的健康状态识别方法、装置、设备及存储介质，属于人工智能领域，所述方法通过预先构建的人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。在进行人脸健康状态识别时，通过预先构建的人脸特征识别模型进行特征提取，以获得用户的脸部特征，然后校正矩阵对用户的脸部特征参数进行校正，最后根据校正结果对人脸的健康状态进行识别。提高了人脸健康状态识别精度，同时通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
附图说明
为了更清楚地说明本申请中的方案，下面将对本申请实施例描述中所需要使用的附图作一个简单介绍，显而易见地，下面描述中的附图是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1示出了本申请可以应用于其中的示例性系统架构图；
图2示出了根据本申请的基于图像的健康状态识别方法的一个实施例的流程图；
图3示出了图2中步骤S201的一种具体实施方式的流程图；
图4示出了图2中步骤S202的一种具体实施方式的流程图；
图5示出了根据本申请的基于图像的健康状态识别装置的一个实施例的结构示意图；
图6示出了根据本申请的计算机设备的一个实施例的结构示意图。
具体实施方式
除非另有定义，本文所使用的所有的技术和科学术语与属于本申请的技术领域的技术人员通常理解的含义相同；本文中在申请的说明书中所使用的术语只是为了描述具体的实施例的目的，不是旨在于限制本申请；本申请的说明书和权利要求书及上述附图说明中的术语“包括”和“具有”以及它们的任何变形，意图在于覆盖不排他的包含。本申请的说明书和权利要求书或上述附图中的术语“第一”、“第二”等是用于区别不同对象，而不是用于描述特定顺序。
在本文中提及“实施例”意味着，结合实施例描述的特定特征、结构或特性可以包含在本申请的至少一个实施例中。在说明书中的各个位置出现该短语并不一定均是指相同的实施例，也不是与其它实施例互斥的独立的或备选的实施例。本领域技术人员显式地和隐式地理解的是，本文所描述的实施例可以与其它实施例相结合。
为了使本技术领域的人员更好地理解本申请方案，下面将结合附图，对本申请实施例中的技术方案进行清楚、完整地描述。
如图1所示，系统架构100可以包括终端设备101、102、103，网络104和服务器105。网络104用以在终端设备101、102、103和服务器105之间提供通信链路的介质。网络104可以包括各种连接类型，例如有线、无线通信链路或者光纤电缆等等。
用户可以使用终端设备101、102、103通过网络104与服务器105交互，以接收或发送消息等。终端设备101、102、103上可以安装有各种通讯客户端应用，例如网页浏览器应用、购物类应用、搜索类应用、即时通信工具、邮箱客户端、社交平台软件等。
终端设备101、102、103可以是具有显示屏并且支持网页浏览的各种电子设备，包括但不限于智能手机、平板电脑、电子书阅读器、MP3播放器(Moving Picture Experts Group Audio Layer III，动态影像专家压缩标准音频层面3)、MP4(Moving Picture Experts Group Audio Layer IV，动态影像专家压缩标准音频层面4)播放器、膝上型便携计算机和台式计算机等等。
服务器105可以是提供各种服务的服务器，例如对终端设备101、102、103上显示的页面提供支持的后台服务器。
需要说明的是，本申请实施例所提供的基于图像的健康状态识别方法一般由服务器/终端设备执行，相应地，基于图像的健康状态识别装置一般设置于服务器/终端设备中。
应该理解，图1中的终端设备、网络和服务器的数目仅仅是示意性的。根据实现需要，可以具有任意数目的终端设备、网络和服务器。
继续参考图2，示出了根据本申请的基于图像的健康状态识别的方法的一个实施例的流程图。所述的基于图像的健康状态识别方法，包括以下步骤：
S201，从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型。
随着人工智能在金融领域中的广泛使用，在移动端开展金融行为的场景越来越常见，而在移动端开展金融行为一般涉及到一套较为严格的审批业务，其中判定用户是否是在健康、头脑清醒的情况下做出的判断与操作，是确保送用户操作资料是否具备法律意义的前提。决定用户操作是否具有法律意义的重要因素在于确定客户在做此操作时是否神志清醒，在实际应用场景中，根据对业务场景的具体分析，导致客户不清醒的原因主要有以下三种情况：1.醉酒；2.极度欠缺休息；3.一些急慢性病。这些原因在用户脸上均会反映出相应的特征，具体可归纳为以下情况：眼部状态(比如是否浮肿，凹陷等)，脸颊状态(比如浮肿，凹陷等)，皮肤状态(比如醉酒发红，比如欠缺休息所导致的皮肤发干)。
具体的，从预设数据库中获取样本数据，在本申请中样本数据为不同用户的人脸图像，基于上述脸部特征对样本数据进行标注，得到对标注后的样本数据进行组合，得到第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型。其中，预设的初始特征识别模型为深度卷积神经网络模型，如CNN模型，卷积神经网络(Convolutional Neural Networks,CNN)是一类包含卷积计算且具有深度结构的前馈神经网络(Feedforward Neural Networks)，是深度学习(deep learning)的代表算法之一。卷积神经网络具有表征学习(representation learning)能力，能够按其阶层结构对输入信息进行平移不变分类(shift-invariant classification)，因此也被称为“平移不变人工神经网络”。卷积神经网络仿造生物的视知觉(visual perception)机制构建，可以进行监督学习和非监督学习，其卷积层内的卷积核参数共享和层间连接的稀疏性使得卷积神经网络能够以较小的计算量对格点化(grid-like topology)特征，例如像素和音频进行学习，有稳定的效果且对数据没有额外的特征工程要求。
S202，采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵。
具体的，采集人脸特征识别模型训练时输出的训练结果，对人脸特征识别模型的训练结果进行标注，对标注后的人脸特征识别模型的训练结果进行组合，得到第二训练样本集，通过第二训练样本集训练初始关系识别模型，得到训练完成的特征关系识别模型，对训练完成的特征关系识别模型进行验证直至模型拟合，获取特征关系识别模型的输出结果，基于特征关系识别模型的输出结果生成校正矩阵，其中，特征关系识别模型的输出结果为矩阵形式。
其中，初始关系识别模型为图卷积神经网络模型，如GCN模型，图卷积网络(Graph Convolutional Network，GCN)是近年来逐渐流行的一种神经网络结构。不同于只能用于网格结构(grid-based)数据的传统网络模型LSTM和CNN，图卷积网络能够处理具有广义拓扑图结构的数据，并深入发掘其特征和规律，例如PageRank引用网络、社交网络、通信网络、蛋白质分子结构等一系列具有空间拓扑图结构的不规则数据。GCN精妙地设计了一种从图数据中提取特征的方法，它能同时对节点特征信息与结构信息进行端对端学习，是目前对图数据学习任务的最佳选择，图卷积适用性极广，适用于任意拓扑结构的节点与图。
S203，获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征。
具体的，在接收到健康状态识别指令时，获取用户的待识别图像，并通过训练好的人脸特征识别模型对待识别图像进行识别，得到用户的脸部特征。需要说明的是，通过人脸特征识别模型对待识别图像进行特征识别，可以得到待识别图像中用户的多个脸部特征。
在本实施例中，基于图像的健康状态识别方法运行于其上的电子设备(例如图1所示的服务器/终端设备)可以通过有线连接方式或者无线连接方式接收到健康状态识别指令。需要指出的是，上述无线连接方式可以包括但不限于3G/4G连接、WiFi连接、蓝牙连接、WiMAX连接、Zigbee连接、UWB(ultra wideband)连接、以及其他现在已知或将来开发的无线连接方式。
S204，基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
具体的，获取人脸特征识别模型输出的多个脸部特征，其中，人脸特征识别模型输出的取矩阵形式，利用校正矩阵分别与人脸特征识别模型输出结果进行内积运算，以完成对待识别图像的多个人脸特征进行校正，其中，权重矩阵可认为是一个校正人脸特征的工具。然后对校正结果进行sigmoid处理，输出sigmoid处理结果，sigmoid处理结果为一个具体数值，通过sigmoid处理结果判断用户的健康状态。
在本申请具体的实施例中，结合CNN深度学习卷积网络模型和GCN图卷积网络模型进行进行模型训练，得到人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。然后可以将训练好的人脸特征识别模型和校正矩阵部署到移动终端，在进行健康状态识别时，通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
本申请公开了一种基于图像的健康状态识别方法、装置、设备及存储介质，属于人工智能领域，所述方法通过预先构建的人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。在进行人脸健康状态识别时，通过预先构建的人脸特征识别模型进行特征提取，以获得用户的脸部特征，然后校正矩阵对用户的脸部特征参数进行校正，最后根据校正结果对人脸的健康状态进行识别。提高了人脸健康状态识别精度，同时通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
进一步地，请参考图3，图3示出了图2中步骤S201的一种具体实施方式的流程图，从预设数据库中获取第一训练样本集，通过第一训练样本集训练初始特征识别模型，得到人脸特征识别模型的步骤，具体包括：
S301，从预设数据库中获取样本数据，对样本数据进行标注；
S302，对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集；
S303，将第一训练样本集导入到初始特征识别模型中进行模型训练，得到初始人脸特征识别模型；
S304，通过第一验证数据集对初始人脸特征识别模型进行验证，输出验证通过的人脸特征识别模型。
具体的，从预设数据库中获取样本数据，在本申请中样本数据为不同用户的人脸图像，基于样本数据的脸部特征对样本数据进行标注，对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集，如可以将标注后的样本数据随机分为10等份的样本子集，其中，随机组合9样本子集作为第一训练样本集，将剩余的样本子集作为第一验证数据集。将第一训练样本集导入到初始特征识别模型中进行模型训练，得到初始人脸特征识别模型，通过第一验证数据集对初始人脸特征识别模型进行验证，输出验证通过的人脸特征识别模型。在上述实施例，通过构建第一训练样本集和第一验证数据集，并分别通过第一训练样本集和第一验证数据集对初始特征识别模型进行训练和验证，可以快速获得人脸特征识别模型。
进一步地，通过第一验证数据集对初始人脸特征识别模型进行验证，输出验证通过的人脸特征识别模型的步骤，具体包括：
将第一验证数据集导入到初始人脸特征识别模型，输出识别结果；
基于识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；
将识别误差与第一预设误差阈值进行比较，若识别误差大于第一预设误差阈值，则对初始人脸特征识别模型进行迭代更新，直到识别误差小于或等于第一预设误差阈值为止，获取验证通过的人脸特征识别模型。
其中，反向传播算法，即误差反向传播算法(Backpropagation algorithm，BP算法)适合于多层神经元网络的一种学习算法，它建立在梯度下降法的基础上，用于深度学习网络的误差计算。BP网络的输入、输出关系实质上是一种映射关系：一个n输入m输出的BP神经网络所完成的功能是从n维欧氏空间向m维欧氏空间中一有限域的连续映射，这一映射具有高度非线性。BP算法的学习过程由正向传播过程和反向传播过程组成。在正向传播过程中，输入信息通过输入层经隐含层，逐层处理并传向输出层，并转入反向传播，逐层求出目标函数对各神经元权值的偏导数，构成目标函数对权值向量的梯量，以作为修改权值的依据。
具体的，将第一验证数据集导入到初始人脸特征识别模型，输出识别结果；基于识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；将识别误差与第一预设误差阈值进行比较，若识别误差大于第一预设误差阈值，则基于初始人脸特征识别模型的损失函数对初始人脸特征识别模型进行迭代更新，直到识别误差小于或等于第一预设误差阈值为止，获取验证通过的人脸特征识别模型。其中，第一预设标准结果和第一预设误差阈值可以提前设定。在上述实施例中，通过反向传播算法对人脸特征识别模型进行验证和迭代，得到符合要求的人脸特征识别模型。
进一步地，请参考图4，图4示出了图2中步骤S202的一种具体实施方式的流程图，采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵的步骤，具体包括：
S401，采集人脸特征识别模型的训练结果，并对训练结果进行标注，得到健康状态标签；
S402，统计健康状态标签的数量，并基于健康状态标签的数量计算每一个健康状态标签的联合分布概率；
S403，基于联合分布概率对健康状态标签进行组合，得到第二训练样本集和第二验证数据集；
S404，根据第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型；
S405，获取特征关系识别模型的输出结果，并基于输出结果生成校正矩阵。
其中，对人脸特征识别模型的训练结果进行标注，得到人脸健康状态的标签，在申请具体的实施例中，人脸健康状态的标签如脸部红晕-醉酒、眼部浮肿-欠缺休息、黑眼圈-欠缺休息等等。联合分布概率简称联合分布，是两个及以上随机变量组成的随机向量的概率分布，在本申请具体的实施例中，通过联合分布概率来表征各个健康状态标签对于人脸健康状态影响的权重值。
具体的，通过第一训练样本集导入人脸特征识别模型中进行训练，采集人脸特征识别模型的每一个训练结果，并对得到训练结果进行标注，得到健康状态标签，如某个训练结果显示用户脸部红晕，则标注为醉酒，某个训练结果显示用户眼部黑眼圈，则标注为欠缺休息。统计得到的所有健康状态标签的数量，并基于健康状态标签的数量计算每一个健康状态标签的联合分布概率，基于联合分布概率对健康状态标签进行组合，得到第二训练样本集和第二验证数据集，根据第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型，获取特征关系识别模型的输出结果，并基于输出结果生成校正矩阵。在上述实施例中，通过对人脸特征识别模型的训练结果进行标注和组合，得到二训练样本集和第二验证数据集，通过二训练样本集和第二验证数据集对对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型的输出结果生成校正矩阵，可以快速获得特征关系识别模型和校正矩阵。
在本申请一种具体的实施例中，在基于联合分布概率对健康状态标签进行组合，得到第二训练样本集和第二验证数据集的步骤之前，还包括：
通过work2vec工具对健康状态标签进行矢量化。
其中，work2vec一款用于词向量计算的工具，word2vec可以在百万数量级的词典和上亿的数据集上进行高效地训练，word2vec得到的训练结果——词向量(word embedding)，可以很好地度量词与词之间的相似性。word2vec算法的背后是一个浅层神经网络。另外需要强调的一点是，word2vec是一个计算word vector的开源工具。
具体的，通过work2vec工具对每一个健康状态标签进行矢量化，得到每一个健康状态标签的特征向量，根据联合分布概率对特征向量进行随机组合，得到第二训练样本集和第二验证数据集，第二训练样本集和第二验证数据集均为向量形式的数据集，通过将向量形式的数据集输入到特征关系识别模型，得到矩阵形式的输出结果，以便获得校正矩阵。
进一步地，根据第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型的步骤，具体包括：
将第二训练样本集导入到预设的初始关系识别模型进行进行训练，得到初始特征关系识别模型；
通过第二验证数据集对初始特征关系识别模型进行验证，输出预测结果；
基于预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差；
将预测误差与第二预设误差阈值进行比较，若预测误差大于第二预设误差阈值，则对初始特征关系识别模型进行迭代更新，直到预测误差小于或等于第二预设误差阈值为止，获取验证通过的特征关系识别模型。
具体的，将第二训练样本集导入到初始关系识别模型进行进行训练，得到初始特征关系识别模型，将第二验证数据集导入到初始特征关系识别模型进行验证，输出预测结果，基于预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差，将预测误差与第二预设误差阈值进行比较，若预测误差大于第二预设误差阈值，则基于初始特征关系识别模型的损失函数对初始特征关系识别模型进行迭代更新，直到预测误差小于或等于第二预设误差阈值为止，获取验证通过的特征关系识别模型。其中，第二预设标准结果和第二预设误差阈值可以提前设定。在上述实施例中，通过反向传播算法对特征关系识别模型进行验证和迭代，得到符合要求的特征关系识别模型。
进一步地，对初始特征关系识别模型进行迭代更新，直到预测误差小于或等于第二预设误差阈值为止，获取验证通过的特征关系识别模型的步骤，具体包括：
设置初始特征关系识别模型的损失函数；
基于损失函数对初始特征关系识别模型进行迭代更新，直至损失函数的输出达到最小值；
获取损失函数的输出达到最小值的特征关系识别模型。
具体的，构建初始特征关系识别模型的损失函数，基于构建的损失函数对特征关系识别模型进行参数优化，直至特征关系识别模型的损失函数达到最小值，特征关系识别模型的损失函数L具体如下：
Figure PCTCN2021090417-appb-000001
其中，c指的是标签号，y表示真实标签值，
Figure PCTCN2021090417-appb-000002
表示预测输出，σ表示sigmoid函数。
在上述实施例中，通过构建初始特征关系识别模型的损失函数，以及基于构建的损失函数对特征关系识别模型进行参数优化，以实现对初始特征关系识别模型进行迭代更新。
进一步地，获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征的步骤，具体包括：
获取待识别图像，并对待识别图像进行扫描，识别待识别图像中的人脸区域；
通过人脸特征识别模型对待识别图像中的人脸区域进行特征识别，得到待识别图像中用户的脸部特征。
具体的，在进行人脸健康状态识别时，获取待识别图像，并将待识别图像输入到训练好的人脸特征识别模型，获取待识别图像中的用户的脸部特征。其中，通过SSD模型识别待识别图像中的人脸区域。
在本申请具体的实施例中，在获取待识别图像后，对待识别图像进行归一化，使用预训练的SSD(Single Shot MultiBox Detector，通用物体检测)模型对人脸区域进行提取。由SSD模型可以提取得到人脸框的左上以及右下点的坐标，根据提取出的坐标将人脸区域由原图中截取出来，并将其缩放到448x448大小，再将缩放后的人脸区域输入到人脸特征识别模型对待识别图像进行识别，得到用户的脸部特征。
在本申请具体的实施例中，通过预先构建的SSD模型对待检测图像进行人脸区域截取，然后使用预先构建的人脸特征识别模型进行用户的人脸特征提取，获取人脸特征张量。然后使用预先构建的校正矩阵对用户人脸特征张量进行校正，最后基于校正结果识别用户的健康状态。在本申请中，通过一个简单的校正矩阵对人脸特征进行校正，提高了健康状态识别的精度，因而适配性很强，且识别过程不会占用大量的计算资源。
需要强调的是，为进一步保证上述待识别图像的私密和安全性，上述待识别图像还可以存储于一区块链的节点中。
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain)，本质上是一个去中心化的数据库，是一串使用密码学方法相关联产生的数据块，每一个数据块中包含了一批次网络交易的信息，用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机可读指令来指令相关的硬件来完成，该计算机可读指令可存储于一计算机可读取存储介质中，该计算机可读指令在执行时，可包括如上述各方法的实施例的流程。其中，前述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)等非易失性存储介质，或随机存储记忆体(Random Access Memory，RAM)等。
应该理解的是，虽然附图的流程图中的各个步骤按照箭头的指示依次显示，但是这些步骤并不是必然按照箭头指示的顺序依次执行。除非本文中有明确的说明，这些步骤的执行并没有严格的顺序限制，其可以以其他的顺序执行。而且，附图的流程图中的至少一部分步骤可以包括多个子步骤或者多个阶段，这些子步骤或者阶段并不必然是在同一时刻执行完成，而是可以在不同的时刻执行，其执行顺序也不必然是依次进行，而是可以与其他步骤或者其他步骤的子步骤或者阶段的至少一部分轮流或者交替地执行。
进一步参考图5，作为对上述图2所示方法的实现，本申请提供了一种基于图像的健康状态识别装置的一个实施例，该装置实施例与图2所示的方法实施例相对应，该装置具体可以应用于各种电子设备中。
如图5所示，本实施例所述的基于图像的健康状态识别装置包括：
第一模型训练模块501，用于从预设数据库中获取第一训练样本集，通过第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
第二模型训练模块502，用于采集人脸特征识别模型的训练结果，并基于训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于特征关系识别模型生成校正矩阵；
脸部特征识别模块503，用于获取待识别图像，并通过人脸特征识别模型对待识别图像进行识别，得到待识别图像中用户的脸部特征；
健康状态识别模块504，用于基于校正矩阵对待识别图像的人脸特征进行校正，根据校正结果识别用户的健康状态。
进一步地，第一模型训练模块501具体包括：
第一标注子模块，用于从预设数据库中获取样本数据，对样本数据进行标注；
第一组合子模块，用于对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集；
第一训练子模块，用于将第一训练样本集导入到初始特征识别模型中进行模型训练，得到初始人脸特征识别模型；
第一验证子模块，用于通过第一验证数据集对初始人脸特征识别模型进行验证，输出验证通过的人脸特征识别模型。
进一步地，第一验证单元具体包括：
第一验证单元，用于将第一验证数据集导入到初始人脸特征识别模型，输出识别结果；
第一误差计算单元，用于基于识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；
第一迭代单元，用于将识别误差与第一预设误差阈值进行比较，若识别误差大于第一预设误差阈值，则对初始人脸特征识别模型进行迭代更新，直到识别误差小于或等于第一预设误差阈值为止，获取验证通过的人脸特征识别模型。
进一步地，第二模型训练模块502具体包括：
第二标注子模块，用于采集人脸特征识别模型的训练结果，并对训练结果进行标注，得到健康状态标签；
联合分布概率计算子模块，用于统计健康状态标签的数量，并基于健康状态标签的数量计算每一个健康状态标签的联合分布概率；
第二组合子模块，用于基于联合分布概率对健康状态标签进行组合，得到第二训练样本集和第二验证数据集；
第二训练子模块，用于根据第二训练样本集对初始关系识别模型进行训练，得到特征关系识别模型；
校正矩阵生成子模块，用于获取特征关系识别模型的输出结果，并基于输出结果生成校正矩阵。
进一步地，第二训练单元具体包括：
第二训练单元，用于将第二训练样本集导入到初始关系识别模型进行进行训练，得到初始特征关系识别模型；
第二验证单元，用于通过第二验证数据集对初始特征关系识别模型进行验证，输出预测结果；
第二误差计算单元，用于基于预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差；
第二迭代单元，用于将预测误差与第二预设误差阈值进行比较，若预测误差大于第二预设误差阈值，则对初始特征关系识别模型进行迭代更新，直到预测误差小于或等于第二预设误差阈值为止，获取验证通过的特征关系识别模型。
进一步地，第二迭代单元具体包括：
损失函数设置子单元，用于设置初始特征关系识别模型的损失函数；
第二迭代子单元，用于基于损失函数对初始特征关系识别模型进行迭代更新，直至损失函数的输出达到最小值；
获取损失函数的输出达到最小值的特征关系识别模型。
进一步地，脸部特征识别模块503具体包括：
人脸区域识别子模块，用于获取待识别图像，并对待识别图像进行扫描，识别待识别图像中的人脸区域；
脸部特征识别子模块，用于通过人脸特征识别模型对待识别图像中的人脸区域进行特征识别，得到待识别图像中用户的脸部特征。
本申请公开了一种基于图像的健康状态识别装置，属于人工智能领域，通过预先构建的人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。在进行人脸健康状态识别时，通过预先构建的人脸特征识别模型进行特征提取，以获得用户的脸部特征，然后校正矩阵对用户的脸部特征参数进行校正，最后根据校正结果对人脸的健康状态进行识别。提高了人脸健康状态识别精度，同时通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
为解决上述技术问题，本申请实施例还提供计算机设备。具体请参阅图6，图6为本实施例计算机设备基本结构框图。
所述计算机设备6包括通过系统总线相互通信连接存储器61、处理器62、网络接口63。需要指出的是，图中仅示出了具有组件61-63的计算机设备6，但是应理解的是，并不要求实施所有示出的组件，可以替代的实施更多或者更少的组件。其中，本技术领域技术人员可以理解，这里的计算机设备是一种能够按照事先设定或存储的指令，自动进行数值计算和/或信息处理的设备，其硬件包括但不限于微处理器、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程门阵列(Field－Programmable Gate Array，FPGA)、数字处理器(Digital Signal Processor，DSP)、嵌入式设备等。
所述计算机设备可以是桌上型计算机、笔记本、掌上电脑及云端服务器等计算设备。所述计算机设备可以与用户通过键盘、鼠标、遥控器、触摸板或声控设备等方式进行人机交互。
所述存储器61至少包括一种类型的可读存储介质，所述可读存储介质包括闪存、硬盘、多媒体卡、卡型存储器(例如，SD或DX存储器等)、随机访问存储器(RAM)、静态随机访问存储器(SRAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、可编程只读存储器(PROM)、磁性存储器、磁盘、光盘等。在一些实施例中，所述存储器61可以是所述计算机设备6的内部存储单元，例如该计算机设备6的硬盘或内存。在另一些实施例中，所述存储器61也可以是所述计算机设备6的外部存储设备，例如该计算机设备6上配备的插接式硬盘，智能存储卡(Smart Media Card,SMC)，安全数字(Secure Digital,SD)卡，闪存卡(Flash Card)等。当然，所述存储器61还可以既包括所述计算机设备6的内部存储单元也包括其外部存储设备。本实施例中，所述存储器61通常用于存储安装于所述计算机设备6的操作系统和各类应用软件，例如基于图像的健康状态识别方法的计算机可读指令等。此外，所述存储器61还可以用于暂时地存储已经输出或者将要输出的各类数据。
所述处理器62在一些实施例中可以是中央处理器(Central Processing Unit，CPU)、控制器、微控制器、微处理器、或其他数据处理芯片。该处理器62通常用于控制所述计算机设备6的总体操作。本实施例中，所述处理器62用于运行所述存储器61中存储的计算机可读指令或者处理数据，例如运行所述基于图像的健康状态识别方法的计算机可读指令。
所述网络接口63可包括无线网络接口或有线网络接口，该网络接口63通常用于在所述计算机设备6与其他电子设备之间建立通信连接。
本申请公开了计算机设备，属于人工智能领域，所述方法通过预先构建的人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。在进行人脸健康状态识别时，通过预先构建的人脸特征识别模型进行特征提取，以获得用户的脸部特征，然后校正矩阵对用户的脸部特征参数进行校正，最后根据校正结果对人脸的健康状态进行识别。提高了人脸健康状态识别精度，同时通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
本申请还提供了另一种实施方式，即提供一种计算机可读存储介质，所述计算机可读存储介质可以是非易失性，也可以是易失性，所述计算机可读存储介质存储有计算机可读指令，所述计算机可读指令可被至少一个处理器执行，以使所述至少一个处理器执行如上述的基于图像的健康状态识别方法的步骤。
本申请公开了一种存储介质，属于人工智能领域，所述方法通过预先构建的人脸特征识别模型和特征关系识别模型，并基于特征关系识别模型生成人脸特征识别模型的校正矩阵。在进行人脸健康状态识别时，通过预先构建的人脸特征识别模型进行特征提取，以获得用户的脸部特征，然后校正矩阵对用户的脸部特征参数进行校正，最后根据校正结果对人脸的健康状态进行识别。提高了人脸健康状态识别精度，同时通过校正矩阵对用户的脸部特征参数进行校正不会占用大量的计算资源和系统资源，有利于减轻系统运算压力和系统集成压力，适配性较强，方便部署于移动终端内。
通过以上的实施方式的描述，本领域的技术人员可以清楚地了解到上述实施例方法可借助软件加必需的通用硬件平台的方式来实现，当然也可以通过硬件，但很多情况下前者是更佳的实施方式。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质(如ROM/RAM、磁碟、光盘)中，包括若干指令用以使得一台终端设备(可以是手机，计算机，服务器，空调器，或者网络设备等)执行本申请各个实施例所述的方法。
显然，以上所描述的实施例仅仅是本申请一部分实施例，而不是全部的实施例，附图中给出了本申请的较佳实施例，但并不限制本申请的专利范围。本申请可以以许多不同的形式来实现，相反地，提供这些实施例的目的是使对本申请的公开内容的理解更加透彻全面。尽管参照前述实施例对本申请进行了详细的说明，对于本领域的技术人员来而言，其依然可以对前述各具体实施方式所记载的技术方案进行修改，或者对其中部分技术特征进行等效替换。凡是利用本申请说明书及附图内容所做的等效结构，直接或间接运用在其他相关的技术领域，均同理在本申请专利保护范围之内。

Claims (20)

  1. 一种基于图像的健康状态识别方法，包括：
    从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
    采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于所述特征关系识别模型生成校正矩阵；
    获取待识别图像，并通过所述人脸特征识别模型对所述待识别图像进行识别，得到所述待识别图像中用户的脸部特征；
    基于所述校正矩阵对所述待识别图像的人脸特征进行校正，根据校正结果识别所述用户的健康状态。
  2. 如权利要求1所述的基于图像的健康状态识别方法，其中，所述从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型的步骤，具体包括：
    从所述预设数据库中获取样本数据，对所述样本数据进行标注；
    对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集；
    将所述第一训练样本集导入到所述初始特征识别模型中进行模型训练，得到初始人脸特征识别模型；
    通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型。
  3. 如权利要求2所述的基于图像的健康状态识别方法，其中，所述通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型的步骤，具体包括：
    将所述第一验证数据集导入到所述初始人脸特征识别模型，输出识别结果；
    基于所述识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；
    将所述识别误差与第一预设误差阈值进行比较，若所述识别误差大于第一预设误差阈值，则对所述初始人脸特征识别模型进行迭代更新，直到所述识别误差小于或等于所述第一预设误差阈值为止，获取验证通过的所述人脸特征识别模型。
  4. 如权利要求1所述的基于图像的健康状态识别方法，其中，所述采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，并基于所述特征关系识别模型生成校正矩阵的步骤，具体包括：
    采集所述人脸特征识别模型的训练结果，并对所述训练结果进行标注，得到健康状态标签；
    统计所述健康状态标签的数量，并基于所述健康状态标签的数量计算每一个所述健康状态标签的联合分布概率；
    基于所述联合分布概率对所述健康状态标签进行组合，得到第二训练样本集和第二验证数据集；
    根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型；
    获取所述特征关系识别模型的输出结果，并基于所述输出结果生成校正矩阵。
  5. 如权利要求4所述的基于图像的健康状态识别方法，其中，所述根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型的步骤，具体包括：
    将所述第二训练样本集导入到所述预设的初始关系识别模型进行进行训练，得到初始特征关系识别模型；
    通过所述第二验证数据集对所述初始特征关系识别模型进行验证，输出预测结果；
    基于所述预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差；
    将所述预测误差与第二预设误差阈值进行比较，若所述预测误差大于第二预设误差阈值，则对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型。
  6. 如权利要求5所述的基于图像的健康状态识别方法，其中，所述对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型的步骤，具体包括：
    设置所述初始特征关系识别模型的损失函数；
    基于所述损失函数对所述初始特征关系识别模型进行迭代更新，直至所述损失函数的输出达到最小值；
    获取所述损失函数的输出达到最小值的所述特征关系识别模型。
  7. 如权利要求1至6任意一项所述的基于图像的健康状态识别方法，其中，所述获取待识别图像，并通过所述人脸特征识别模型对所述待识别图像进行识别，得到所述待识别图像中用户的脸部特征的步骤，具体包括：
    获取所述待识别图像，并对所述待识别图像进行扫描，识别所述待识别图像中的人脸区域；
    通过所述人脸特征识别模型对所述待识别图像中的人脸区域进行特征识别，得到所述待识别图像中用户的脸部特征。
  8. 一种基于图像的健康状态识别装置，包括：
    第一模型训练模块，用于从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
    第二模型训练模块，用于采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于所述特征关系识别模型生成校正矩阵；
    脸部特征识别模块，用于获取待识别图像，并通过所述人脸特征识别模型对所述待识别图像进行识别，得到所述待识别图像中用户的脸部特征；
    健康状态识别模块，用于基于所述校正矩阵对所述待识别图像的人脸特征进行校正，根据校正结果识别所述用户的健康状态。
  9. 一种计算机设备，包括存储器和处理器，所述存储器中存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现如下所述的基于图像的健康状态识别方法：
    从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
    采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于所述特征关系识别模型生成校正矩阵；
    获取待识别图像，并通过所述人脸特征识别模型对所述待识别图像进行识别，得到所述待识别图像中用户的脸部特征；
    基于所述校正矩阵对所述待识别图像的人脸特征进行校正，根据校正结果识别所述用户的健康状态。
  10. 如权利要求9所述的计算机设备，其中，所述从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型的步骤，具体包括：
    从所述预设数据库中获取样本数据，对所述样本数据进行标注；
    对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集；
    将所述第一训练样本集导入到所述初始特征识别模型中进行模型训练，得到初始人脸特征识别模型；
    通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型。
  11. 如权利要求10所述的计算机设备，其中，所述通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型的步骤，具体包括：
    将所述第一验证数据集导入到所述初始人脸特征识别模型，输出识别结果；
    基于所述识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；
    将所述识别误差与第一预设误差阈值进行比较，若所述识别误差大于第一预设误差阈值，则对所述初始人脸特征识别模型进行迭代更新，直到所述识别误差小于或等于所述第一预设误差阈值为止，获取验证通过的所述人脸特征识别模型。
  12. 如权利要求9所述的计算机设备，其中，所述采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，并基于所述特征关系识别模型生成校正矩阵的步骤，具体包括：
    采集所述人脸特征识别模型的训练结果，并对所述训练结果进行标注，得到健康状态标签；
    统计所述健康状态标签的数量，并基于所述健康状态标签的数量计算每一个所述健康状态标签的联合分布概率；
    基于所述联合分布概率对所述健康状态标签进行组合，得到第二训练样本集和第二验证数据集；
    根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型；
    获取所述特征关系识别模型的输出结果，并基于所述输出结果生成校正矩阵。
  13. 如权利要求12所述的计算机设备，其中，所述根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型的步骤，具体包括：
    将所述第二训练样本集导入到所述预设的初始关系识别模型进行进行训练，得到初始特征关系识别模型；
    通过所述第二验证数据集对所述初始特征关系识别模型进行验证，输出预测结果；
    基于所述预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差；
    将所述预测误差与第二预设误差阈值进行比较，若所述预测误差大于第二预设误差阈值，则对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型。
  14. 如权利要求13所述的计算机设备，其中，所述对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型的步骤，具体包括：
    设置所述初始特征关系识别模型的损失函数；
    基于所述损失函数对所述初始特征关系识别模型进行迭代更新，直至所述损失函数的输出达到最小值；
    获取所述损失函数的输出达到最小值的所述特征关系识别模型。
  15. 一种计算机可读存储介质，所述计算机可读存储介质上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现如下所述的基于图像的健康状态识别方法：
    从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型；
    采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，基于所述特征关系识别模型生成校正矩阵；
    获取待识别图像，并通过所述人脸特征识别模型对所述待识别图像进行识别，得到所述待识别图像中用户的脸部特征；
    基于所述校正矩阵对所述待识别图像的人脸特征进行校正，根据校正结果识别所述用户的健康状态。
  16. 如权利要求15所述的计算机可读存储介质，其中，所述从预设数据库中获取第一训练样本集，通过所述第一训练样本集训练预设的初始特征识别模型，得到人脸特征识别模型的步骤，具体包括：
    从所述预设数据库中获取样本数据，对所述样本数据进行标注；
    对标注后的样本数据进行随机组合，得到第一训练样本集和第一验证数据集；
    将所述第一训练样本集导入到所述初始特征识别模型中进行模型训练，得到初始人脸特征识别模型；
    通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型。
  17. 如权利要求16所述的计算机可读存储介质，其中，所述通过所述第一验证数据集对所述初始人脸特征识别模型进行验证，输出验证通过的所述人脸特征识别模型的步骤，具体包括：
    将所述第一验证数据集导入到所述初始人脸特征识别模型，输出识别结果；
    基于所述识别结果与第一预设标准结果，使用反向传播算法进行拟合计算，获取识别误差；
    将所述识别误差与第一预设误差阈值进行比较，若所述识别误差大于第一预设误差阈值，则对所述初始人脸特征识别模型进行迭代更新，直到所述识别误差小于或等于所述第一预设误差阈值为止，获取验证通过的所述人脸特征识别模型。
  18. 如权利要求15所述的计算机可读存储介质，其中，所述采集所述人脸特征识别模型的训练结果，并基于所述训练结果对初始关系识别模型进行训练，得到特征关系识别模型，并基于所述特征关系识别模型生成校正矩阵的步骤，具体包括：
    采集所述人脸特征识别模型的训练结果，并对所述训练结果进行标注，得到健康状态标签；
    统计所述健康状态标签的数量，并基于所述健康状态标签的数量计算每一个所述健康状态标签的联合分布概率；
    基于所述联合分布概率对所述健康状态标签进行组合，得到第二训练样本集和第二验证数据集；
    根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型；
    获取所述特征关系识别模型的输出结果，并基于所述输出结果生成校正矩阵。
  19. 如权利要求18所述的计算机可读存储介质，其中，所述根据所述第二训练样本集对预设的初始关系识别模型进行训练，得到特征关系识别模型的步骤，具体包括：
    将所述第二训练样本集导入到所述预设的初始关系识别模型进行进行训练，得到初始特征关系识别模型；
    通过所述第二验证数据集对所述初始特征关系识别模型进行验证，输出预测结果；
    基于所述预测结果与第二预设标准结果，使用反向传播算法进行拟合计算，获取预测误差；
    将所述预测误差与第二预设误差阈值进行比较，若所述预测误差大于第二预设误差阈值，则对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型。
  20. 如权利要求19所述的计算机可读存储介质，其中，所述对所述初始特征关系识别模型进行迭代更新，直到所述预测误差小于或等于所述第二预设误差阈值为止，获取验证通过的所述特征关系识别模型的步骤，具体包括：
    设置所述初始特征关系识别模型的损失函数；
    基于所述损失函数对所述初始特征关系识别模型进行迭代更新，直至所述损失函数的输出达到最小值；
    获取所述损失函数的输出达到最小值的所述特征关系识别模型。
PCT/CN2021/090417 2020-11-17 2021-04-28 基于图像的健康状态识别方法、装置、设备及存储介质 WO2022105118A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202011286362.3 2020-11-17
CN202011286362.3A CN112395979B (zh) 2020-11-17 2020-11-17 基于图像的健康状态识别方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2022105118A1 true WO2022105118A1 (zh) 2022-05-27

Family

ID=74605823

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/090417 WO2022105118A1 (zh) 2020-11-17 2021-04-28 基于图像的健康状态识别方法、装置、设备及存储介质

Country Status (2)

Country Link
CN (1) CN112395979B (zh)
WO (1) WO2022105118A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115270993A (zh) * 2022-08-23 2022-11-01 南通思诺船舶科技有限公司 柴油机组状态检测方法及系统
CN117237800A (zh) * 2023-08-01 2023-12-15 广州智在信息科技有限公司 基于人工智能的作物生长监控方法及计算机设备
CN117668528A (zh) * 2024-02-01 2024-03-08 成都华泰数智科技有限公司 基于物联网的天然气调压器故障检测方法及系统

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112395979B (zh) * 2020-11-17 2024-05-10 平安科技（深圳）有限公司 基于图像的健康状态识别方法、装置、设备及存储介质
CN112861788A (zh) * 2021-03-10 2021-05-28 中电健康云科技有限公司 基于脸色识别技术判断健康状况的方法
CN113158842A (zh) * 2021-03-31 2021-07-23 中国工商银行股份有限公司 一种识别方法、系统、设备及介质
CN113361402B (zh) * 2021-06-04 2023-08-18 北京百度网讯科技有限公司 识别模型的训练方法、确定准确率的方法、装置和设备
CN117031971B (zh) * 2023-07-18 2024-04-19 东莞莱姆森科技建材有限公司 基于智能镜的智能家具设备调节方法、装置、设备及介质

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108509905A (zh) * 2018-03-30 2018-09-07 百度在线网络技术（北京）有限公司 健康状态评估方法、装置、电子设备及存储介质
CN109119163A (zh) * 2018-07-16 2019-01-01 安徽昱康智能科技有限公司 基于面部识别的健康检测方法及其系统
CN109190535A (zh) * 2018-08-23 2019-01-11 南京邮电大学 一种基于深度学习的人脸面部肤色分析方法和系统
CN109492557A (zh) * 2018-10-28 2019-03-19 北京工业大学 一种基于au共生关系约束的自然场景下面部表情识别方法
CN110729053A (zh) * 2019-10-11 2020-01-24 平安医疗健康管理股份有限公司 数据处理方法、装置、计算机设备和存储介质
CN111612133A (zh) * 2020-05-20 2020-09-01 广州华见智能科技有限公司 基于人脸图像多阶段关系学习的内脏器官特征编码方法
CN112395979A (zh) * 2020-11-17 2021-02-23 平安科技（深圳）有限公司 基于图像的健康状态识别方法、装置、设备及存储介质

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11026634B2 (en) * 2017-04-05 2021-06-08 doc.ai incorporated Image-based system and method for predicting physiological parameters
CN107633203A (zh) * 2017-08-17 2018-01-26 平安科技（深圳）有限公司 面部情绪识别方法、装置及存储介质
WO2019245085A1 (en) * 2018-06-22 2019-12-26 Jun Ho Park Method, apparatus and medium for performing 3d model creation of people and identification of people via model matching
CN111227789A (zh) * 2018-11-29 2020-06-05 百度在线网络技术（北京）有限公司 人体健康监护方法和装置
CN109685011A (zh) * 2018-12-25 2019-04-26 北京华航无线电测量研究所 一种基于深度学习的地下管线检测识别方法
CN111860362A (zh) * 2020-07-24 2020-10-30 北京百度网讯科技有限公司 生成人脸图像校正模型及校正人脸图像的方法和装置

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108509905A (zh) * 2018-03-30 2018-09-07 百度在线网络技术（北京）有限公司 健康状态评估方法、装置、电子设备及存储介质
CN109119163A (zh) * 2018-07-16 2019-01-01 安徽昱康智能科技有限公司 基于面部识别的健康检测方法及其系统
CN109190535A (zh) * 2018-08-23 2019-01-11 南京邮电大学 一种基于深度学习的人脸面部肤色分析方法和系统
CN109492557A (zh) * 2018-10-28 2019-03-19 北京工业大学 一种基于au共生关系约束的自然场景下面部表情识别方法
CN110729053A (zh) * 2019-10-11 2020-01-24 平安医疗健康管理股份有限公司 数据处理方法、装置、计算机设备和存储介质
CN111612133A (zh) * 2020-05-20 2020-09-01 广州华见智能科技有限公司 基于人脸图像多阶段关系学习的内脏器官特征编码方法
CN112395979A (zh) * 2020-11-17 2021-02-23 平安科技（深圳）有限公司 基于图像的健康状态识别方法、装置、设备及存储介质

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115270993A (zh) * 2022-08-23 2022-11-01 南通思诺船舶科技有限公司 柴油机组状态检测方法及系统
CN117237800A (zh) * 2023-08-01 2023-12-15 广州智在信息科技有限公司 基于人工智能的作物生长监控方法及计算机设备
CN117668528A (zh) * 2024-02-01 2024-03-08 成都华泰数智科技有限公司 基于物联网的天然气调压器故障检测方法及系统
CN117668528B (zh) * 2024-02-01 2024-04-12 成都华泰数智科技有限公司 基于物联网的天然气调压器故障检测方法及系统

Also Published As

Publication number Publication date
CN112395979A (zh) 2021-02-23
CN112395979B (zh) 2024-05-10

Similar Documents

Publication Publication Date Title
WO2022105118A1 (zh) 基于图像的健康状态识别方法、装置、设备及存储介质
US11487995B2 (en) Method and apparatus for determining image quality
WO2021077984A1 (zh) 对象识别方法、装置、电子设备及可读存储介质
US10635893B2 (en) Identity authentication method, terminal device, and computer-readable storage medium
WO2022142006A1 (zh) 基于语义识别的话术推荐方法、装置、设备及存储介质
WO2019119505A1 (zh) 人脸识别的方法和装置、计算机装置及存储介质
WO2020024484A1 (zh) 用于输出数据的方法和装置
WO2022174491A1 (zh) 基于人工智能的病历质控方法、装置、计算机设备及存储介质
WO2022105117A1 (zh) 一种图像质量评价的方法、装置、计算机设备及存储介质
WO2020238353A1 (zh) 数据处理方法和装置、存储介质及电子装置
WO2023016007A1 (zh) 人脸识别模型的训练方法、装置及计算机程序产品
WO2022188697A1 (zh) 提取生物特征的方法、装置、设备、介质及程序产品
WO2021139316A1 (zh) 建立表情识别模型方法、装置、计算机设备及存储介质
CN112418059A (zh) 一种情绪识别的方法、装置、计算机设备及存储介质
CN112995414B (zh) 基于语音通话的行为质检方法、装置、设备及存储介质
CN113707299A (zh) 基于问诊会话的辅助诊断方法、装置及计算机设备
CN113420690A (zh) 基于感兴趣区域的静脉识别方法、装置、设备及存储介质
CN113254491A (zh) 一种信息推荐的方法、装置、计算机设备及存储介质
CN112668482A (zh) 人脸识别训练方法、装置、计算机设备及存储介质
WO2021068613A1 (zh) 面部识别方法、装置、设备及计算机可读存储介质
CN114282258A (zh) 截屏数据脱敏方法、装置、计算机设备及存储介质
WO2022073341A1 (zh) 基于语音语义的疾病实体匹配方法、装置及计算机设备
CN113343898A (zh) 基于知识蒸馏网络的口罩遮挡人脸识别方法、装置及设备
WO2022142032A1 (zh) 手写签名校验方法、装置、计算机设备及存储介质
CN113643283A (zh) 一种人体衰老状况的检测方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21893274

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21893274

Country of ref document: EP

Kind code of ref document: A1