CN110634557B - Medical care resource auxiliary allocation method and system based on deep neural network - Google Patents
Medical care resource auxiliary allocation method and system based on deep neural network Download PDFInfo
- Publication number
- CN110634557B CN110634557B CN201910781429.1A CN201910781429A CN110634557B CN 110634557 B CN110634557 B CN 110634557B CN 201910781429 A CN201910781429 A CN 201910781429A CN 110634557 B CN110634557 B CN 110634557B
- Authority
- CN
- China
- Prior art keywords
- human body
- target human
- neural network
- image
- image data
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
Images
Classifications
-
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0012—Biomedical image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-based segmentation
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H30/00—ICT specially adapted for the handling or processing of medical images
-
- G—PHYSICS
- G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
- G16H—HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA
- G16H40/00—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices
- G16H40/20—ICT specially adapted for the management or administration of healthcare resources or facilities; ICT specially adapted for the management or operation of medical equipment or devices for the management or administration of healthcare resources or facilities, e.g. managing hospital staff or surgery rooms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30004—Biomedical image processing
Abstract
The invention discloses a medical care resource auxiliary allocation method and system based on a deep neural network, wherein the method comprises the following steps: acquiring image data based on a high-definition camera, and performing image enhancement processing on the image data to obtain enhanced image data; performing target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image; performing posture feature extraction processing on the segmented target human body image based on supervised learning to obtain the posture feature of the target human body; and carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on the deep neural network, and carrying out medical care resource allocation on the target human body based on a prediction result. In the implementation of the invention, the observation of the target non-contact posture state is realized, and the efficiency of scheduling and distributing medical care resources is improved.
Description
Technical Field
The invention relates to the technical field of intelligent old age care, in particular to a medical care resource auxiliary allocation method and system based on a deep neural network.
Background
Medical care resource allocation of the current hospital mainly depends on a traditional hospital information management system, and mainly only simply collects, stores and manages medical care information, and manually arranges work by depending on experience of management personnel. The traditional method of allocating by means of daily experience of a manager is influenced by subjectivity and is difficult to ensure scientificity of an allocation plan, and is objectively influenced and difficult to ensure execution efficiency of the allocation plan.
Aiming at health condition management and disease prevention prediction of long-term patients and old people under sleep and rest conditions, based on an artificial intelligence deep learning technology, data deep mining and physiological data modeling are carried out on image related information of the patients and the old people, demand prediction and optimal configuration of medical care resources are served, clinical decision is assisted, and the problems of difficult disease analysis, shortage of medical care resources and reasonable configuration of medical care personnel are solved.
Mainly solves the following problems:
1) and (4) non-contact observation of the body state of the target. Adopt high definition digtal camera, can obtain disease, old man at the health characteristic of true natural state, especially to the monitoring under some sleep states, carry check out test set influence rest quality, too rely on medical personnel to make rounds of wards, increase working strength, and efficiency is lower.
2) And (4) combining visual sensing and body state evaluation and early warning prediction of a historical database. Artificial intelligence and deep learning technology are introduced, mass medical information data are fully mined, a body state image evaluation system based on a database is obtained, certain early warning and forecasting can be achieved, medical staff are served, medical resource distribution is assisted, and efficiency is improved.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides a medical care resource auxiliary allocation method and system based on a deep neural network, which can realize the observation of a target non-contact posture state and improve the efficiency of medical care resource scheduling and distribution.
In order to solve the above technical problem, an embodiment of the present invention provides a medical care resource assisted allocation method based on a deep neural network, where the method includes:
acquiring image data based on a high-definition camera, and performing image enhancement processing on the image data to obtain enhanced image data;
performing target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
performing posture feature extraction processing on the segmented target human body image based on supervised learning to obtain the posture feature of the target human body;
and carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on the deep neural network, and carrying out medical care resource auxiliary allocation on the target human body based on a prediction result.
Optionally, the performing image enhancement processing on the image data to obtain enhanced image data includes:
and carrying out image enhancement preprocessing on the image data based on a frequency domain enhancement algorithm to obtain enhanced image data.
Optionally, the performing, by using the convolutional neural network model, target human body detection and segmentation processing on the enhanced image data to obtain a segmented target human body image includes:
inputting the enhanced image data as original data of a convolutional neural network model, and dividing the enhanced image data into n image cell units based on the number of channels;
inputting the n image cell units into a backbone network of the convolutional neural network model, and performing limit filtering processing on a result output by the backbone network to obtain a limit processing result;
inputting the limiting processing result into a classifier of the second classification to carry out logistic regression, and judging whether a target human body exists or not;
if the target human body exists, inputting the limiting processing result into a locator module for positioning processing, acquiring bounding boxes coordinates of a series of anchor boxes of the target human body, and acquiring an accurate position coordinate frame based on a non-maximum suppression algorithm;
and storing the corresponding coordinate area of the enhanced image data based on the accurate position coordinate frame to obtain the segmented target human body image.
Optionally, the inputting the n image cell units into a backbone network of the convolutional neural network model, and performing constraint filtering processing on a result output by the backbone network to obtain a constraint processing result, includes:
inputting the n image cell units into a backbone network of a convolutional neural network model, and obtaining a characteristic tensor map of each image cell unit through a series of convolutional layers and pooling layers;
and limiting and filtering the characteristic tensor map of each image cell unit through a threshold value adjusting module to obtain a limiting and processing result.
Optionally, the performing, on the basis of the convolutional neural network model, target human body detection and segmentation processing on the enhanced image data to obtain a segmented target human body image further includes:
and performing preliminary abnormal state early warning prediction processing based on the time period and the segmented target human body image position, and performing early warning on corresponding medical care personnel based on a preliminary prediction result.
Optionally, the performing, based on supervised learning, a body state feature extraction process on the segmented target human body image to obtain a body state feature of the target human body includes:
inputting the segmented target human body image as input image data into a designed convolutional neural network for body state feature extraction to obtain generalized abstract body state features, wherein the input image data is obtained by changing the size of the segmented target human body image into a uniform image size;
the training data of the designed convolutional neural network are image data of a target human body, artificial marking and marking are carried out on the training data according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed convolutional neural network based on supervised learning.
Optionally, the first layer of the designed convolutional neural network is a convolutional layer with a convolutional kernel of 7 × 7 and a sliding step of 2, and after passing through the Relu activation function, the second layer is a convolutional layer with a convolutional kernel of 3 × 3 and a sliding step of 2; constructing a convolution module for reuse; the designed convolutional neural network adopts a full-connection design, and the network depth is realized by repeating the block times.
Optionally, the performing, by the deep neural network, abnormal state early warning and prediction processing on the posture features of the target human body includes:
extracting data with relevance from a target human body condition database to construct a deep neural network, and extracting and processing feature information with strong relevance through a single-layer wider network structure to obtain a first extracted feature;
the method comprises the steps that discrete data in a target human body condition database are input from one or more discrete phases of a finite set, are converted into sparse tensor representations through embedding and heat coding, are input into a deep neural network of softmax, second extraction features are obtained, and the second extraction features are subjected to connection operation;
fully connecting tensor results of the first extracted features and the second extracted features with feature tensors abstracted out by the feature extraction sub-networks to construct a combined neural network; and (3) solving abnormal state early warning prediction based on a random gradient descent optimization algorithm by taking a cross entropy function as a target function and taking a normal and abnormal situation of a target human body as a logistic regression through a Relu activation function.
Optionally, the target human body condition database is composed of a general database and a target human body association database; the general database stores common data information of people; the target human body association database stores individual data information of corresponding specific target human bodies; the target human body condition database also stores relevant data recorded by medical staff on a traditional medical information system based on the condition of the target person, and the relevant data comprises: age, sex, weather, sleep, exercise, disease condition and cause related information.
In addition, an embodiment of the present invention further provides a system for assisted medical care resource allocation based on a deep neural network, where the system includes:
the image acquisition and processing module: the system comprises a high-definition camera, a storage unit and a display unit, wherein the high-definition camera is used for acquiring image data and performing image enhancement processing on the image data to obtain enhanced image data;
an image segmentation module: the system is used for carrying out target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
a feature extraction module: the system is used for extracting the body state characteristics of the segmented target human body image based on supervised learning to obtain the body state characteristics of the target human body;
the early warning and resource allocation module: the system is used for carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on the deep neural network and carrying out medical care resource auxiliary allocation on the target human body based on the prediction result.
In the embodiment of the invention, the high-definition camera is used for acquiring image data to realize non-contact observation of the state of the target human body; the system combines visual sensing and body state assessment and early warning prediction of a historical database, introduces artificial intelligence and deep learning technology, fully excavates massive medical information data, obtains a body state image assessment system based on the database, realizes abnormal state early warning prediction, and carries out auxiliary allocation of medical care resources according to related abnormal state early warning prediction, thereby providing allocation efficiency and increasing the working efficiency of medical care personnel.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the embodiments or the prior art descriptions will be briefly described below, it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings can be obtained by those skilled in the art without creative efforts.
FIG. 1 is a schematic flow chart of a medical care resource assisted deployment method based on a deep neural network according to an embodiment of the present invention;
FIG. 2 is a schematic view of a flow structure of a convolutional neural network model for performing target human body detection and segmentation according to an embodiment of the present invention;
FIG. 3 is a schematic flow structure diagram of a deep neural network model for performing early warning and prediction of abnormal states in an embodiment of the present invention;
fig. 4 is a schematic structural component diagram of a healthcare resource assisted deployment system based on a deep neural network in an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Examples
Referring to fig. 1, fig. 1 is a schematic flow chart illustrating a medical care resource assisted deployment method based on a deep neural network according to an embodiment of the present invention.
As shown in fig. 1, a method for assisted medical care resource deployment based on a deep neural network includes:
s11: acquiring image data based on a high-definition camera, and performing image enhancement processing on the image data to obtain enhanced image data;
in a specific implementation process of the present invention, the performing image enhancement processing on the image data to obtain enhanced image data includes: and carrying out image enhancement preprocessing on the image data based on a frequency domain enhancement algorithm to obtain enhanced image data.
Specifically, according to the fact that corresponding high-definition cameras are installed or additionally installed in hospitals and nursing homes, video streams are collected through the high-definition cameras, image data are obtained, and the image data are subjected to image enhancement; the image data is preprocessed by a frequency domain enhancement method, noise elimination is mainly completed, illumination compensation is performed according to indoor illumination environment change, the brightness of a darkroom environment is enhanced, and the dark of an environment with excessively high brightness is compensated.
S12: performing target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
in a specific implementation process of the present invention, the performing target human body detection and segmentation processing on the enhanced image data based on the convolutional neural network model to obtain a segmented target human body image includes: inputting the enhanced image data as original data of a convolutional neural network model, and dividing the enhanced image data into n image cell units based on the number of channels; inputting the n image cell units into a backbone network of the convolutional neural network model, and performing limit filtering processing on a result output by the backbone network to obtain a limit processing result; inputting the limiting processing result into a classifier of the second classification for carrying out logistic regression, and judging whether a target human body exists or not; if the target human body exists, inputting the limiting processing result into a locator module for positioning processing, acquiring bounding boxes coordinates of a series of anchor boxes of the target human body, and acquiring an accurate position coordinate frame based on a non-maximum suppression algorithm; and storing the corresponding coordinate area of the enhanced image data based on the accurate position coordinate frame to obtain the segmented target human body image.
Further, the inputting n image cell units into a back bone network of the convolutional neural network model, and performing constraint filtering processing on a result output by the back bone network to obtain a constraint processing result, includes: inputting n image cell units into a back bone network of a convolutional neural network model, and obtaining a characteristic tensor map of each image cell unit through a series of convolutional layers and pooling layers; and limiting and filtering the characteristic tensor maps of the image cell units by a threshold value adjusting module to obtain a limiting and processing result.
Further, the performing target human body detection and segmentation processing on the enhanced image data based on the convolutional neural network model to obtain a segmented target human body image further includes: and performing preliminary abnormal state early warning prediction processing based on the time period and the segmented target human body image position, and early warning to corresponding medical personnel based on a preliminary prediction result.
Specifically, please refer to fig. 2, fig. 2 is a schematic view of a flow structure of a convolutional neural network model for detecting and segmenting a target human body according to an embodiment of the present invention; as shown in fig. 2, real-time tracking of a target human body is required to be achieved, so that the problem of real-time performance of the system method needs to be considered in an important point, and for a high-definition video stream acquired by a high-definition camera, the target human body only occupies a small part of the whole picture, so that detection and segmentation of the target human body in a camera view scene are considered, whether the target human body exists or not and at which position needs to be found and located, image data of the target human body is segmented from an original video image, a picture with a small size is obtained, further feature extraction processing is facilitated, and real-time performance is guaranteed.
Aiming at task requirements, the designed neural network mainly solves two problems: 1) judging whether a target human body exists in the image; 2) and extracting the position of the target human body in the picture. In order to solve the two problems, the real-time problem in practical application is considered at the same time, and an end-to-end Convolutional Neural Network (CNN) for human body detection and positioning is designed. The network structure can rapidly and accurately judge whether a target human body exists in the current high-definition camera video image, and finally output the target human body position (1 or more) or the non-target human body positions if the target human body position exists. Wherein, the backhaul adopts an open-source Yolo-v3-lite basic network, designs a two-classification (with or without target human body) classifier and an NMS locator based on a regression method. The specific implementation process is as follows:
(first step) inputting the picture of the enhanced image data as the original data of the convolutional neural network model, wherein the image size of the original data is (width) x Height (Height) x channel number (channel)), and is determined by the used acquisition camera, and the specifications are generally as follows; the observation shows that the speed of processing a high-definition image of a large frame at a time is slow, and in order to further improve the real-time performance, the original data is divided into (SxS) n image cell units (grid cells); taking a 1080P three-channel diagram as an example, an image input is 1920x1080x3, and 6x6 divided into 320x180x3 is 32 image cell units.
And (step two), inputting the segmented image cell units into a backbone network, obtaining a characteristic tensor map of each cell unit through a series of convolution layers and pooling layers (thinning according to yolo), and performing a limit filtering treatment on the characteristic tensor map through a threshold adjusting module (threshold).
(third step) inputting the result of the second step into a classifier of the second classification for logistic regression, judging whether the target human body is detected, inputting the result of the second step into a locator module to obtain bounding boxes coordinates of a series of anchor boxes of the target human body, and obtaining the most accurate position coordinate box by adopting an NMS (non-maximum suppression) algorithm; and storing the photo area data corresponding to the coordinate frame for further processing.
After the segmented target human body image is obtained, corresponding preliminary abnormal state early warning prediction needs to be performed, according to a work and rest scheme designed by a doctor for the actual situation of the target human body and according to the state of the target human body corresponding to the current time, the following four situations can be preliminarily classified as shown in the following table:
when the target person detection network model detects that the time (T _ threshold) that the target human body is in an abnormal state reaches a certain threshold value is obtained, for example, when the target human body should sleep, the target human body is not in a bed for a long time or the target human body should go to exercise and activity, and the target human body always rests in the bed, the relevant medical personnel needs to be reminded to give an early warning.
S13: performing posture feature extraction processing on the segmented target human body image based on supervised learning to obtain the posture feature of the target human body;
in a specific implementation process of the present invention, the performing, based on supervised learning, a body state feature extraction process on the segmented target human body image to obtain the body state feature of the target human body includes: inputting the segmented target human body image as input image data into a designed convolutional neural network for body state feature extraction to obtain generalized abstract body state features, wherein the input image data is obtained by changing the size of the segmented target human body image into a uniform image size; the training data of the designed convolutional neural network are image data of a target human body, artificial marking and marking are carried out on the training data according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed convolutional neural network based on supervised learning.
Further, the first layer of the designed convolutional neural network is a convolutional layer with a convolutional kernel of 7 × 7 and a sliding step of 2, and after passing through the Relu activation function, the second layer is a convolutional layer with a convolutional kernel of 3 × 3 and a sliding step of 2; constructing a convolution module for reuse; the designed convolutional neural network adopts a full-connection design, and the network depth is realized by repeating the block times.
Specifically, the segmented target human body image obtained in S12 is input as input image data to a designed convolutional neural network for body state feature extraction, and the designed convolutional neural network performs feature extraction on image information to obtain generalized abstract features. The training data set is image data of a target human body, manual marking and marking are carried out according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed network by adopting a supervised learning method.
The specific implementation process is as follows: the method comprises the following steps that (first step), pictures obtained by target human body detection are input as original data, the picture data Resize (size change) is processed into uniform picture sizes, in a design network, the size is (224x224x3), different sizes of Resize can have certain influence on actual effect, the main influence is the distance between a camera and a target person, the state of the target person when the target person rests on a bed is taken as an example, when the camera is close to the bed, the target person is detected to be clear, the picture size is large, the corresponding size of the Resize is large, otherwise, the opposite is true, the higher resolution effect is maintained as much as possible, the clear image features are extracted accurately, and the effect is better. Limited by practical conditions, the situation that the size is small can also occur, but the aspect ratio can be ensured to be (1: 3' -1:1) as much as possible; (second step) the first layer is the convolutional layer with convolution kernel 7x7 and sliding step size 2. The second layer is the convolutional layer with convolution kernel 3x3 and sliding step size 2, via the Relu activation function. A reusable convolution module is constructed; and (third step) considering the real-time requirement, adopting the design idea of a full connection layer, properly deepening the depth of the network structure on the premise of not increasing more parameters, and ensuring the generalization capability of the network model to the characteristics. Setting the depth of the network according to the requirement, and realizing the depth by repeating the block times; and (fourth step), the feature extraction network is coupled to the whole combined network as a sub-network, and the final model parameters are obtained through further training.
S14: and carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on the deep neural network, and carrying out medical care resource auxiliary allocation on the target human body based on a prediction result.
In a specific implementation process of the present invention, the performing abnormal state early warning and prediction processing on the body state characteristics of the target human body based on the deep neural network includes: extracting data with relevance from a target human body condition database to construct a deep neural network, and extracting and processing feature information with strong relevance through a single-layer wider network structure to obtain a first extracted feature; the method comprises the steps that discrete data in a target human body condition database are input from one or more discrete phases of a finite set, are converted into sparse tensor representations through embedding and heat coding, are input into a deep neural network of softmax, second extraction features are obtained, and the second extraction features are subjected to connection operation; fully connecting tensor results of the first extracted features and the second extracted features with feature tensors abstracted out by the feature extraction sub-networks to construct a combined neural network; and (3) solving abnormal state early warning prediction based on a random gradient descent optimization algorithm by taking a cross entropy function as a target function and taking a normal and abnormal situation of a target human body as a logistic regression through a Relu activation function.
Further, the target human body condition database consists of a general database and a target human body association database; the general database stores common data information of people; the target human body association database stores individual data information of corresponding specific target human bodies; the target human body condition database also stores relevant data recorded by medical staff on a traditional medical information system based on the condition of the target person, and the relevant data comprises: age, sex, climate, sleep, exercise, disease condition, and cause related information.
Specifically, please refer to fig. 3, fig. 3 is a schematic flow structure diagram of a deep neural network model for performing abnormal state early warning prediction in the embodiment of the present invention; as shown in fig. 3, a state evaluation system is designed based on a width-depth joint neural network to realize early warning prediction of an abnormal state of a target person; the network model is designed as a width and depth combined neural network, and the functions of the following parts are comprehensively realized: 1) performing data mining on all data collected by the database; 2) further excavating implicit correlation between the images and discrete data of the database by combining abstract characteristics of the target human body image data information; 3) and further deducing and predicting the physical condition of the target human body to realize the early warning and predicting function.
The specific implementation process is as follows:
(first step) constructing a wide neural network among data with certain relevance in a database, and extracting characteristic information with strong relevance through a single-layer wide network structure; (second step) the discrete data in the database is one or more discrete phase inputs from a finite set, which can be regarded as classified input data, and is converted into sparse tensor representation through one-hot (Embedding) introduction; inputting the characteristics into a deep neural network of softmax, and acquiring the characteristics to perform connection operation (coordination); and (third step), connecting (participating) the feature Tensor (Tensor) results of the first step and the second step with the feature Tensor abstracted out by the feature extraction sub-network to construct a joint neural network. Through Relu activation function, taking cross entropy function as target function, taking normal and abnormal situations of target person as logistic regression, and adopting random gradient descent optimization method to solve; and (fourth step) carrying out a large amount of data training on the body state evaluation model of the target person, acquiring image data acquired by a high-definition camera and various data in a database through a training data set, obtaining model parameters of the whole wide-depth neural network after training, deploying the model in a cloud server after the training reaches a certain stability degree, training the model by using subsequent updating data, and updating regularly, so that the accuracy is better, the database is richer, and more target persons can be served.
The database consists of a general database and a target person association database, wherein the general database concerns common data information of people and is probably related to disease common knowledge; the database associated with the target person is primarily concerned with the individual data information of the corresponding specific target person. The data of the database may have the conditions of large discreteness, unobvious relevance and the like, but the deep neural network obtained through the training of the database can complete data mining according to relevant input and output result conclusions. Data collection may be based on relevant data recorded by healthcare workers on traditional medical information systems based on the condition of the target person, including but not limited to such things as: relevant information such as age, sex, weather, sleep, movement, morbidity, cause and the like is introduced, and some key habits and information (possibly needing to be collected again) in the long-term life process of the target person are introduced to establish a required database; the database is updated regularly, and the corresponding network model is upgraded regularly.
In the embodiment of the invention, the high-definition camera is used for acquiring image data to realize non-contact observation of the state of the target human body; the system combines visual sensing and body state assessment and early warning prediction of a historical database, introduces artificial intelligence and deep learning technology, fully excavates massive medical information data, obtains a body state image assessment system based on the database, realizes abnormal state early warning prediction, and carries out auxiliary allocation of medical care resources according to related abnormal state early warning prediction, thereby providing allocation efficiency and increasing the working efficiency of medical care personnel.
Examples
Referring to fig. 2, fig. 2 is a schematic structural component diagram of a healthcare resource assisted deployment system based on a deep neural network according to an embodiment of the present invention.
As shown in fig. 2, a system for assisted medical resource deployment based on a deep neural network includes:
the image acquisition processing module 11: the system comprises a high-definition camera, a storage unit and a display unit, wherein the high-definition camera is used for acquiring image data and performing image enhancement processing on the image data to obtain enhanced image data;
in a specific implementation process of the present invention, the performing image enhancement processing on the image data to obtain enhanced image data includes: and carrying out image enhancement preprocessing on the image data based on a frequency domain enhancement algorithm to obtain enhanced image data.
Specifically, according to the fact that corresponding high-definition cameras are installed or additionally installed in hospitals and nursing homes, video streams are collected through the high-definition cameras, image data are obtained, and the image data are subjected to image enhancement; the image data is preprocessed by adopting a frequency domain enhancement method, noise elimination is mainly completed, illumination compensation is performed according to indoor illumination environment change, the brightness of a darkroom environment is enhanced, and the dark of an over-bright brightness environment is compensated.
The image segmentation module 12: the system is used for carrying out target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
in a specific implementation process of the present invention, the performing target human body detection and segmentation processing on the enhanced image data based on the convolutional neural network model to obtain a segmented target human body image includes: inputting the enhanced image data as original data of a convolutional neural network model, and dividing the enhanced image data into n image cell units based on the number of channels; inputting the n image cell units into a backbone network of the convolutional neural network model, and performing constraint filtering processing on a result output by the backbone network to obtain a constraint processing result; inputting the limiting processing result into a classifier of the second classification to carry out logistic regression, and judging whether a target human body exists or not; if the target human body exists, inputting the limiting processing result into a locator module for positioning processing, acquiring bounding boxes coordinates of a series of anchor boxes of the target human body, and acquiring an accurate position coordinate frame based on a non-maximum suppression algorithm; and storing the corresponding coordinate area of the enhanced image data based on the accurate position coordinate frame to obtain the segmented target human body image.
Further, the inputting n image cell units into a back bone network of the convolutional neural network model, and performing constraint filtering processing on a result output by the back bone network to obtain a constraint processing result, includes: inputting n image cell units into a back bone network of a convolutional neural network model, and obtaining a characteristic tensor map of each image cell unit through a series of convolutional layers and pooling layers; and limiting and filtering the characteristic tensor map of each image cell unit through a threshold value adjusting module to obtain a limiting and processing result.
Further, the performing target human body detection and segmentation processing on the enhanced image data based on the convolutional neural network model to obtain a segmented target human body image further includes: and performing preliminary abnormal state early warning prediction processing based on the time period and the segmented target human body image position, and early warning to corresponding medical personnel based on a preliminary prediction result.
Specifically, please refer to fig. 2, fig. 2 is a schematic view of a flow structure of a convolutional neural network model for detecting and segmenting a target human body according to an embodiment of the present invention; as shown in fig. 2, real-time tracking of a target human body is required to be implemented, so that the real-time problem of the method of the present system needs to be mainly considered, and for a high-definition video stream acquired by a high-definition camera, the target human body only occupies a small part of the whole picture, so that detection and segmentation of the target human body in a camera view scene need to be considered, whether the target human body exists or at which position needs to be found and positioned, and image data of the target human body is segmented from an original video image, so as to obtain a picture with a small size, which facilitates further feature extraction processing, and ensures real-time.
Aiming at task requirements, the designed neural network mainly solves two problems: 1) judging whether a target human body exists in the image; 2) and extracting the position of the target human body in the picture. In order to solve the two problems, the real-time problem in practical application is considered at the same time, and an end-to-end Convolutional Neural Network (CNN) for human body detection and positioning is designed. The network structure can quickly and accurately judge whether a target human body exists in the current high-definition camera video image, and finally output None or target human body position (1 or more) if the target human body position exists. Wherein, backbone adopts open-source Yolo-v3-lite basic network, and designs a two-classification (with or without target human body) classifier and an NMS locator based on regression method. The specific implementation process is as follows:
(first step) inputting the picture of the enhanced image data as the original data of the convolutional neural network model, wherein the image size of the original data is (width) x Height (Height) x channel number (channel)), and is determined by the used acquisition camera, and the specifications are generally as follows; the observation shows that the speed of processing a high-definition image of a large frame at a time is slow, and in order to further improve the real-time performance, the original data is divided into (SxS) n image cell units (grid cells); taking a 1080P three-channel diagram as an example, an image input is 1920x1080x3, and 6x6 divided into 320x180x3 is 32 image cell units.
And (step two), inputting the segmented image cell units into a backbone network, obtaining a characteristic tensor map of each cell unit through a series of convolution layers and pooling layers (refined according to yolo), and performing a limiting filtering treatment on the characteristic tensor map through a threshold adjusting module (threshold).
(third step) inputting the result of the second step into a classifier of the second classification for logistic regression, judging whether the object human body is detected, inputting the result of the second step into a locator module to obtain bounding boxes coordinates of a series of anchor boxes of the object human body, and obtaining the most accurate position coordinate box by adopting an NMS (non-maximum suppression) algorithm; and storing the photo area data corresponding to the coordinate frame for further processing.
After the segmented target human body image is obtained, corresponding preliminary abnormal state early warning prediction needs to be performed, according to a work and rest scheme designed by a doctor for the actual situation of the target human body and according to the state of the target human body corresponding to the current time, the following four situations can be preliminarily classified as shown in the following table:
when the target person detection network model detects that the time (T _ threshold) that the target human body is in an abnormal state reaches a certain threshold value is obtained, for example, when the target human body should sleep, the target human body is not in a bed for a long time or the target human body should go to exercise and activity, and the target human body always rests in the bed, the relevant medical personnel needs to be reminded to give an early warning.
The feature extraction module 13: the system is used for extracting the body state characteristics of the segmented target human body image based on supervised learning to obtain the body state characteristics of the target human body;
in a specific implementation process of the present invention, the obtaining of the posture characteristics of the target human body by performing posture characteristic extraction processing on the segmented target human body image based on supervised learning includes: inputting the segmented target human body image as input image data into a designed convolutional neural network for body state feature extraction to obtain generalized abstract body state features, wherein the input image data is obtained by changing the size of the segmented target human body image into a uniform image size; the training data of the designed convolutional neural network are image data of a target human body, artificial marking and marking are carried out on the training data according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed convolutional neural network based on supervised learning.
Further, the first layer of the designed convolutional neural network is a convolutional layer with a convolutional kernel of 7 × 7 and a sliding step of 2, and after passing through the Relu activation function, the second layer is a convolutional layer with a convolutional kernel of 3 × 3 and a sliding step of 2; constructing a convolution module for reuse; the designed convolutional neural network adopts a full-connection design, and the network depth is realized by repeating the block times.
Specifically, the segmented target human body image obtained in S12 is input as input image data to a designed convolutional neural network for body state feature extraction, and the designed convolutional neural network performs feature extraction on image information to obtain generalized abstract features. The training data set is the image data of the target human body, manual marking and marking are carried out according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed network by adopting a supervision learning method.
The specific implementation process is as follows: the method comprises the following steps of (first step) inputting a picture obtained by detecting a target human body as original data, processing picture data Resize (changing size) into uniform picture size, wherein the size is (224x224x3) in a design network, the different sizes of Resize can have certain influence on actual effect, the main influence is the distance between a camera and the target person, taking the state of the target person when the target person is resting on a bed as an example, when the camera is close to the bed, the detected target person is clear, the picture size is large, the size of the corresponding Resize is large, otherwise, the opposite is true, the higher resolution effect is kept as much as possible, the clear image feature is accurately extracted, and the effect is better. Limited by practical conditions, the situation that the size is small can also occur, but the aspect ratio can be ensured to be (1: 3' -1:1) as much as possible; (second step) the first layer is the convolutional layer with convolution kernel 7x7 and sliding step size 2. The second layer is the convolutional layer with convolution kernel 3x3 and sliding step size 2, via the Relu activation function. A reusable convolution module is constructed; and (third step) considering the real-time requirement, adopting a design idea of a full connection layer, properly deepening the depth of a network structure on the premise of not increasing more parameters, and ensuring the generalization capability of a network model to the characteristics. Setting the depth of the network according to the requirement, and realizing the depth by repeating the block times; and (fourth step), coupling the feature extraction network as a sub-network into the whole combined network, and further training to obtain final model parameters.
The early warning and resource allocation module 14: the system is used for carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on the deep neural network and carrying out medical care resource auxiliary allocation on the target human body based on the prediction result.
In a specific implementation process of the present invention, the performing abnormal state early warning and prediction processing on the body state characteristics of the target human body based on the deep neural network includes: extracting data with relevance from a target human body condition database to construct a deep neural network, and extracting and processing feature information with strong relevance through a single-layer wider network structure to obtain a first extracted feature; the method comprises the steps that discrete data of a target human body condition database are input from one or more discrete phases of a finite set, are converted into sparse tensor representation through embedding and embedding of thermal coding, are input into a deep neural network of softmax, second extraction features are obtained, and the second extraction features are subjected to connection operation; fully connecting tensor results of the first extracted features and the second extracted features with feature tensors abstracted out by the feature extraction sub-networks to construct a combined neural network; and (3) solving abnormal state early warning prediction based on a random gradient descent optimization algorithm by taking a cross entropy function as a target function and taking a normal and abnormal situation of a target human body as a logistic regression through a Relu activation function.
Further, the target human body condition database consists of a general database and a target human body association database; the general database stores common data information of people; the target human body association database stores individual data information of corresponding specific target human bodies; the target human body condition database also stores relevant data recorded by medical staff on a traditional medical information system based on the condition of the target person, and the relevant data comprises: age, sex, weather, sleep, exercise, disease condition and cause related information.
Specifically, please refer to fig. 3, fig. 3 is a schematic flow structure diagram of a deep neural network model for performing abnormal state early warning prediction in the embodiment of the present invention; as shown in fig. 3, a state evaluation system is designed based on a width-depth joint neural network to realize early warning prediction of an abnormal state of a target person; the network model is designed into a width and depth combined neural network, and the functions of the following parts are comprehensively realized: 1) performing data mining on all data collected by the database; 2) further excavating implicit correlation between the images and discrete data of the database by combining abstract characteristics of the target human body image data information; 3) and further deducing and predicting the physical condition of the target human body to realize the early warning and predicting function.
The specific implementation process is as follows:
(first step) constructing a wide neural network among data with certain relevance in a database, and extracting characteristic information with strong relevance through a single-layer wide network structure; (second step) the discrete data in the database is one or more discrete phase inputs from a finite set, can be regarded as classified input data, and is converted into sparse tensor characterization through thermal coding (one-hot) introduction Embedding (Embedding); inputting the characteristics into a deep neural network of softmax, and acquiring the characteristics to perform connection operation (collocation); and (third step), connecting (competition) the feature Tensor (Tensor) results of the first step and the second step with the feature Tensor abstracted by the feature extraction sub-network to construct a joint neural network. Through a Relu activation function, taking a cross entropy function as a target function, performing logistic regression on normal and abnormal conditions of a target person, and solving by adopting a random gradient descent optimization method; and (fourth step), carrying out mass data training on the established target person body state evaluation model, collecting image data acquired by a high-definition camera and various data in a database through training data, obtaining model parameters of the whole wide-depth neural network after training, deploying the model in a cloud server after the training reaches a certain stability degree, training the model by using subsequent updated data, and updating the model at regular time, thereby ensuring better precision, richer database and more target human bodies to be served.
The database consists of a general database and a target person association database, wherein the general database concerns common data information of people and is probably related to disease common knowledge; the database associated with the target person is primarily concerned with the individual data information of the corresponding specific target person. The data of the database may have the conditions of large discreteness, unobvious relevance and the like, but the deep neural network obtained through the training of the database can complete data mining according to relevant input and output result conclusions. Data collection may be based on relevant data recorded by healthcare workers on traditional medical information systems based on the condition of the target person, including but not limited to such things as: relevant information such as age, sex, weather, sleep, movement, morbidity, cause and the like is introduced, and some key habits and information (possibly needing to be collected again) in the long-term life process of a target person are introduced to establish a required database; the database is updated regularly, and the corresponding network model is upgraded regularly.
In the embodiment of the invention, the high-definition camera acquires image data to realize non-contact type target human body state observation; the system combines visual sensing and body state assessment and early warning prediction of a historical database, introduces artificial intelligence and deep learning technology, fully excavates massive medical information data, obtains a body state image assessment system based on the database, realizes abnormal state early warning prediction, and carries out auxiliary allocation of medical care resources according to related abnormal state early warning prediction, thereby providing allocation efficiency and increasing the working efficiency of medical care personnel.
Those skilled in the art will appreciate that all or part of the steps in the methods of the above embodiments may be implemented by associated hardware instructed by a program, which may be stored in a computer-readable storage medium, and the storage medium may include: read Only Memory (ROM), Random Access Memory (RAM), magnetic or optical disks, and the like.
In addition, the method and the system for assisted medical care resource allocation based on the deep neural network provided by the embodiment of the present invention are described in detail, and a specific embodiment is adopted herein to explain the principle and the implementation manner of the present invention, and the description of the embodiment is only used to help understanding the method and the core idea of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, there may be variations in the specific embodiments and the application scope, and in summary, the content of the present specification should not be construed as a limitation to the present invention.
Claims (9)
1. A medical care resource auxiliary allocation method based on a deep neural network is characterized by comprising the following steps:
acquiring image data based on a high-definition camera, and performing image enhancement processing on the image data to obtain enhanced image data;
performing target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
performing posture feature extraction processing on the segmented target human body image based on supervised learning to obtain the posture feature of the target human body;
carrying out abnormal state early warning prediction processing on the body state characteristics of the target human body based on a deep neural network, and carrying out medical care resource auxiliary allocation on the target human body based on a prediction result;
the early warning and prediction processing of the abnormal state of the body state characteristics of the target human body based on the deep neural network comprises the following steps:
extracting data with strong relevance from a target human body condition database to construct a deep neural network, and extracting and processing the characteristic information with strong relevance through a single-layer wide network structure to obtain a first extracted characteristic;
the weak-relevance discrete data in the target human body condition database are input from one or more discrete phases of a finite set, are introduced into embedding and converted into sparse tensor representation through thermal coding, are input into a deep neural network of softmax, obtain second extraction features, and are subjected to connection operation;
fully connecting tensor results of the first extracted features and the second extracted features with feature tensors abstracted out by the feature extraction sub-networks to construct a combined neural network; through a Relu activation function, taking a cross entropy function as a target function, performing logistic regression on normal and abnormal conditions of a target human body, and performing early warning prediction on solving abnormal states based on a random gradient descent optimization algorithm;
the feature extraction sub-network is a network for carrying out body state feature extraction on the segmented target human body image based on supervised learning.
2. The auxiliary medical resource deployment method according to claim 1, wherein the image enhancement processing on the image data to obtain enhanced image data includes:
and carrying out image enhancement preprocessing on the image data based on a frequency domain enhancement algorithm to obtain enhanced image data.
3. The auxiliary medical resource allocation method according to claim 1, wherein the performing target human body detection and segmentation processing on the enhanced image data based on the convolutional neural network model to obtain a segmented target human body image comprises:
inputting the enhanced image data as original data of a convolutional neural network model, and dividing the enhanced image data into n image cell units based on the number of channels;
inputting the n image cell units into a backbone network of the convolutional neural network model, and performing limit filtering processing on a result output by the backbone network to obtain a limit processing result;
inputting the limiting processing result into a classifier of the second classification for carrying out logistic regression, and judging whether a target human body exists or not;
if the target human body exists, inputting the limiting processing result into a locator module for positioning processing, acquiring bounding boxes coordinates of a series of anchor boxes of the target human body, and acquiring an accurate position coordinate frame based on a non-maximum suppression algorithm;
and storing the corresponding coordinate area of the enhanced image data based on the accurate position coordinate frame to obtain the segmented target human body image.
4. The auxiliary medical resource allocation method according to claim 3, wherein the step of inputting the n image cell units into a back bone network of the convolutional neural network model and performing constraint filtering on the results output by the back bone network to obtain constraint processing results comprises:
inputting the n image cell units into a backbone network of a convolutional neural network model, and obtaining a characteristic tensor map of each image cell unit through a series of convolutional layers and pooling layers;
and limiting and filtering the characteristic tensor map of each image cell unit through a threshold value adjusting module to obtain a limiting and processing result.
5. The auxiliary medical resource allocation method according to claim 1, wherein the convolution-based neural network model performs target human body detection and segmentation on the enhanced image data, and further comprises, after obtaining a segmented target human body image:
and performing preliminary abnormal state early warning prediction processing based on the time period and the segmented target human body image position, and early warning to corresponding medical personnel based on a preliminary prediction result.
6. The medical care resource assisted allocation method according to claim 1, wherein the obtaining of the posture characteristics of the target human body by performing posture characteristic extraction processing on the segmented target human body image based on supervised learning comprises:
inputting the segmented target human body image as input image data into a designed convolutional neural network for body state feature extraction to obtain generalized abstract body state features, wherein the input image data is obtained by changing the size of the segmented target human body image into a uniform image size;
the training data of the designed convolutional neural network are image data of a target human body, artificial marking and marking are carried out on the training data according to normal and abnormal conditions, and parameter stabilization and optimization are carried out on the designed convolutional neural network based on supervised learning.
7. The method for assisted medical care resource allocation according to claim 6, wherein the first layer of the designed convolutional neural network is a convolutional layer with a convolutional kernel of 7x7 and a sliding step of 2, and after passing through the Relu activation function, the second layer is a convolutional layer with a convolutional kernel of 3x3 and a sliding step of 2; constructing a convolution module for reuse; the designed convolutional neural network adopts a full-connection design, and the network depth is realized by repeating the block times.
8. The assisted medical resource allocation method according to claim 1, wherein the target body condition database is composed of a general database and a target body association database; the general database stores common data information of people; the target human body association database stores individual data information of corresponding specific target human bodies; the target human body condition database also stores relevant data recorded by medical staff on a traditional medical information system based on the condition of the target person, and the relevant data comprises: age, sex, climate, sleep, exercise, disease condition, and cause related information.
9. A depth neural network-based healthcare resource assisted deployment system, comprising:
the image acquisition and processing module: the system comprises a high-definition camera, a storage unit and a display unit, wherein the high-definition camera is used for acquiring image data and performing image enhancement processing on the image data to obtain enhanced image data;
an image segmentation module: the system is used for carrying out target human body detection and segmentation processing on the enhanced image data based on a convolutional neural network model to obtain a segmented target human body image;
a feature extraction module: the system is used for extracting the body state characteristics of the segmented target human body image based on supervised learning to obtain the body state characteristics of the target human body;
the early warning and resource allocation module: the system is used for carrying out abnormal state early warning and prediction processing on the body state characteristics of the target human body based on the deep neural network and carrying out medical care resource auxiliary allocation on the target human body based on the prediction result;
the early warning and prediction processing of the abnormal state of the body state characteristics of the target human body based on the deep neural network comprises the following steps:
extracting data with strong relevance from a target human body condition database to construct a deep neural network, and extracting and processing feature information with strong relevance through a single-layer wide network structure to obtain a first extracted feature;
the weak-relevance discrete data in the target human body condition database are input from one or more discrete phases of a finite set, are introduced into embedding and converted into sparse tensor representation through thermal coding, are input into a deep neural network of softmax, obtain second extraction features, and are subjected to connection operation;
fully connecting tensor results of the first extracted features and the second extracted features with feature tensors abstracted out by the feature extraction sub-networks to construct a combined neural network; through Relu activation function, taking a cross entropy function as a target function, taking two conditions of normal and abnormal of a target human body as logistic regression, and carrying out early warning prediction on solving abnormal states based on a random gradient descent optimization algorithm;
the feature extraction sub-network is a network for carrying out body state feature extraction on the segmented target human body image based on supervised learning.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910781429.1A CN110634557B (en) | 2019-08-23 | 2019-08-23 | Medical care resource auxiliary allocation method and system based on deep neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910781429.1A CN110634557B (en) | 2019-08-23 | 2019-08-23 | Medical care resource auxiliary allocation method and system based on deep neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110634557A CN110634557A (en) | 2019-12-31 |
| CN110634557B true CN110634557B (en) | 2022-08-23 |
Family
ID=68970574
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910781429.1A Active CN110634557B (en) | 2019-08-23 | 2019-08-23 | Medical care resource auxiliary allocation method and system based on deep neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110634557B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113408321B (en) * | 2020-03-16 | 2023-08-22 | 中国人民解放军战略支援部队信息工程大学 | Real-time target detection method and device for lightweight image and video data |
| CN114281185B (en) * | 2021-04-25 | 2023-10-27 | 浙江壹体科技有限公司 | Body state identification and somatosensory interaction system and method based on embedded platform |
| CN117409978B (en) * | 2023-12-15 | 2024-04-19 | 贵州大学 | Disease prediction model construction method, system, device and readable storage medium |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407903A (en) * | 2016-08-31 | 2017-02-15 | 四川瞳知科技有限公司 | Multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method |
| CN107145878A (en) * | 2017-06-01 | 2017-09-08 | 重庆邮电大学 | Old man's anomaly detection method based on deep learning |
| CN109993122A (en) * | 2019-04-02 | 2019-07-09 | 中国石油大学(华东) | A kind of pedestrian based on depth convolutional neural networks multiplies staircase anomaly detection method |
-
2019
- 2019-08-23 CN CN201910781429.1A patent/CN110634557B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407903A (en) * | 2016-08-31 | 2017-02-15 | 四川瞳知科技有限公司 | Multiple dimensioned convolution neural network-based real time human body abnormal behavior identification method |
| CN107145878A (en) * | 2017-06-01 | 2017-09-08 | 重庆邮电大学 | Old man's anomaly detection method based on deep learning |
| CN109993122A (en) * | 2019-04-02 | 2019-07-09 | 中国石油大学(华东) | A kind of pedestrian based on depth convolutional neural networks multiplies staircase anomaly detection method |
Non-Patent Citations (2)
| Title |
|---|
| 卷积神经网络在人物检测上的应用;张翼飞等;《2019年全国公共安全通信学术研讨会》;20190815;第2节 * |
| 多通道卷积神经网络图像识别方法;易超人等;《河南科技大学学报（自然科学版）》;20180630;第38卷(第3期);第1-3节 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110634557A (en) | 2019-12-31 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110634557B (en) | Medical care resource auxiliary allocation method and system based on deep neural network | |
| US11922626B2 (en) | Systems and methods for automatic detection and quantification of pathology using dynamic feature classification | |
| CN112163634A (en) | Example segmentation model sample screening method and device, computer equipment and medium | |
| CN114693615A (en) | Deep learning concrete bridge crack real-time detection method based on domain adaptation | |
| CN106650804B (en) | A kind of face sample cleaning method and system based on deep learning feature | |
| CN114359727A (en) | Tea disease identification method and system based on lightweight optimization Yolo v4 | |
| WO2022178946A1 (en) | Melanoma image recognition method and apparatus, computer device, and storage medium | |
| CN107437068A (en) | Pig individual discrimination method based on Gabor direction histograms and pig chaeta hair pattern | |
| CN114240874A (en) | Bone age assessment method and device based on deep convolutional neural network and feature fusion and computer readable storage medium | |
| CN113822185A (en) | Method for detecting daily behavior of group health pigs | |
| CN110688980A (en) | Human body posture classification method based on computer vision | |
| CN110264407A (en) | Image Super-resolution model training and method for reconstructing, device, equipment and storage medium | |
| CN117115045A (en) | Method for improving medical image data quality based on Internet generation type artificial intelligence | |
| CN116012332A (en) | Hierarchical graph-based pathological image primary tumor stage multi-example learning method, frame, equipment and medium | |
| CN111709941A (en) | Lightweight automatic deep learning system and method for pathological image | |
| Lizarazo et al. | Segmentation of remotely sensed imagery: moving from sharp objects to fuzzy regions | |
| CN102867214A (en) | Counting management method for people within area range | |
| US20230334868A1 (en) | Surgical phase recognition with sufficient statistical model | |
| CN108710901A (en) | A kind of deformity of spine screening system and method based on deep learning | |
| CN114550918A (en) | Mental disorder evaluation method and system based on drawing characteristic data | |
| CN109726691B (en) | Monitoring method | |
| Liu et al. | Study on Pig Body Condition Scoring Based on Deep Learning Model EfficientNet-B0 | |
| CN112381032B (en) | Indoor unattended rapid detection method for resisting human body posture interference | |
| Cai et al. | Sheep Face Detection and Disease Prevention Based on Convolutional Neural Network | |
| Ghofrani Esfahani | Delineation of agricultural field boundaries using random sets |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |