CN108806792B - Deep learning face diagnosis system - Google Patents
Deep learning face diagnosis system Download PDFInfo
- Publication number
- CN108806792B CN108806792B CN201711255031.1A CN201711255031A CN108806792B CN 108806792 B CN108806792 B CN 108806792B CN 201711255031 A CN201711255031 A CN 201711255031A CN 108806792 B CN108806792 B CN 108806792B
- Authority
- CN
- China
- Prior art keywords
- neural network
- convolutional neural
- module
- deep learning
- layer
- 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
- 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
- G06F18/24—Classification techniques
- G06F18/243—Classification techniques relating to the number of classes
- G06F18/24323—Tree-organised classifiers
-
- 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
- G06V40/172—Classification, e.g. identification
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Health & Medical Sciences (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a facial diagnosis method and system based on deep learning. The invention uses a large number of face images with disease labels to train the deep learning network, and after repeated iteration and parameter updating, the deep learning network can automatically and effectively extract the face characteristics including eyes, ears, noses, mouths and eyebrows, and find out the internal relation between the face characteristics and various diseases of human bodies, thereby effectively detecting and screening the diseases of new face images of testees, and the invention is a non-invasive computer-aided automatic diagnosis method. The system can update parameters by continuously collecting the face pictures of the patient and then training, so that the prediction judgment result is more accurate. The method and the system can effectively solve the problem of difficult disease detection in poverty-stricken laggard areas, so that people can conveniently and quickly carry out non-invasive automatic disease detection and screening through terminals such as mobile phones, computers and the like, and diseases can be treated in time, thereby improving the living quality level of human beings.
Description
Technical Field
The invention belongs to the fields of computer vision, machine learning and medicine, and particularly relates to a computer method and a system for diagnosing diseases through facial diagnosis based on deep learning.
Background
Over two thousand years ago, the Chinese ancient book Huangdi's internal classic records that the blood and qi flow from the twelve main meridians and the three hundred and sixty five main meridians to the face and the orifices. This indicates that pathological changes of the five zang-organs and six fu-organs of a person are manifested in the relevant areas of the face. In China, an experienced doctor can grasp the whole body and local lesion conditions of a patient by observing facial features of the patient, and the diagnosis mode is called 'facial diagnosis'. The disadvantage of the diagnostic method is that it requires a lot of experience of the doctor to have a high accuracy.
With the development of science and technology, big data and deep learning technology develop rapidly in recent years. The deep learning technology is that a multi-level neural network is utilized, and a large amount of data are trained, so that a computer can learn and understand complex data such as images and sounds and can make corresponding behaviors, which is sometimes better than that of a human being. The deep learning network can extract contrast features which are high in complexity and difficult for human beings to understand.
In 2006, Geoffrey Hinton, professor Geoffrey, canada, proposed an algorithm for deep learning. He and his student Ruslan Salakhutdinov published a paper in science, from which he triggered a hot tide of deep learning studies. In 2012, the deep convolutional neural network structure AlexNet proposed by the university of toronto team won the champion in the ImageNet large-scale visual recognition challenge race (ILSVRC). In 2014, Google corporation and oxford university visual geometry groups each achieved excellent performance in ILSVRC using their deep convolutional neural networks GoogleNet and VGGNet, which are more structurally complex than AlexNet. In 2015, Revzmin et al, Microsoft Asian institute, using the residual error network ResNet, captured the ILSVRC champion, and obtained the best results of the year in image classification, target detection, semantic segmentation, and other tasks. The residual network has a greater depth, up to 152 layers. In 2016, the integrated deep neural network performed well in ILSVRC tasks.
Therefore, in the current big data era, people can utilize increasingly complex depth models to analyze and extract effective information contained in mass data, so that more accurate prediction can be made on unknown data. The essence of deep learning is that learning features are extracted by constructing a machine learning model with a plurality of hidden layers and a large amount of training data, so that the accuracy of classification or prediction is improved. A large number of research experiments show that the characteristic learning method based on big data deep learning can be used for more effectively expressing the intrinsic information of the data than the method of artificially constructing the characteristic. The invention is a method for realizing the purpose of diagnosing diseases through the face by utilizing big data deep learning.
Disclosure of Invention
The invention aims to conveniently and rapidly detect and screen diseases for all human beings and improve the quality level of life of the human beings, in particular to poor and lagged regions.
In order to achieve the purpose, the inventor provides a method and a system for non-invasive automatic detection and screening of diseases only through face images based on a big data deep learning technology. The method comprises the following steps:
s100: and collecting a proper face image sample with a disease diagnosis result label, and establishing a face image database with the disease diagnosis label.
S200: and combining a convolution layer, a pooling layer, a batch normalization function ReLu, a residual error structure and an inclusion structure into a mixing module, then combining a dimensionality reduction module, and establishing a deep convolutional neural network by a plurality of combinations in a cascade mode.
S300: and (5) preprocessing the picture acquired in the S100, taking the preprocessed picture as training data, and training a deep convolutional neural network model to obtain network model parameters.
S400: and (5) extracting the features of the training data by using the trained deep convolutional neural network model in the step (S300), and training a random forest classifier by using the extracted features to obtain a classification model.
S500: inputting the attempted image, extracting the features of the human face image to be classified by using the convolutional neural network model in S300, and then obtaining the disease category to which the human face image to be detected belongs by using the classifier model obtained in S400.
The invention is also characterized in that:
step S100 is to add a disease diagnosis result label according to a disease diagnosis report issued at that time in a hospital while acquiring a face image of a subject. And establishing a database of Niger species (black species), Mongolian species (yellow species) and Caucasian species (caucasian species) according to different skin colors. Preprocessing the collected pictures, wherein the preprocessed pictures have the same size and are input in three color dimensions of red, green and blue.
Step S200 uses a hybrid module combining a residual structure and an inclusion structure in the network construction. The input data is processed by Batch Normalization (Batch Normalization) in the network construction, and the pooling layer uses a global average pooling method.
In step S300, the preprocessed face image is divided into two parts, namely a training sample and a testing sample, the deep convolutional neural network is trained by the training sample, and the weight is updated by adopting a supervised training mode and a back propagation algorithm. And training the deep convolutional neural network using a Graphics Processor (GPU). And finally, inputting the test sample into a deep convolution neural network for testing. If the pre-trained network is adopted, the fine-tuning technology is required to be assisted.
In step S400, the trained deep convolutional neural network model is used as a feature extractor, and a classifier is trained by using the extracted features to obtain a classification model. This step of the function requires the use of a multivariate classifier model.
The disease types which can be detected and screened in step S500 include hyperthyroidism, hypothyroidism, hansheng disease, acromegaly, renal failure, neonatal pathological jaundice, Down 'S syndrome, thalassemia, DiGeorge' S syndrome and mitral stenosis. More and more disease species will be added.
The invention has the beneficial effect that the invention can automatically extract the characteristics under the data driving of the training image with the disease diagnosis result label through the deep convolutional neural network. The deep convolutional neural network can avoid the additional preprocessing process of the traditional image recognition algorithm. A residual structure is applied in this deep convolutional neural network to solve the problem of accuracy degradation as the network deepens.
The features extracted by the deep convolutional neural network can be prevented from being influenced by color illumination and the like, and more complex features which are difficult to describe by human can be extracted. The deep convolutional neural network and the classifiers are effectively combined, so that not only can rich characteristics be extracted, but also a proper classifier can be selected according to the characteristics of a data set of the classifier, and the accuracy of the whole classification model is improved.
Drawings
FIG. 1 is a schematic diagram of the overall machine learning algorithm of the present invention;
FIG. 2 is a schematic diagram of the structure of the deep convolutional neural network of the present invention;
fig. 3 is a schematic diagram of the system architecture of the present invention.
Detailed Description
In order to explain technical contents, structural features, and objects and effects of the present invention in detail, the following detailed description is given with reference to the accompanying drawings in conjunction with the embodiments.
Referring to fig. 1, in order to solve the technical problem of the present invention, one technical solution adopted by the present invention is to provide a method based on deep learning, including the following steps:
s100: collecting a proper face image sample, adding a disease type label according to a disease diagnosis report issued by a hospital during collection, and establishing a face image database with a disease diagnosis result label. Databases are divided into three major categories, the Niger species (black), Mongolian species (yellow), and Caucasian species (caucasian). Preprocessing the collected pictures, wherein the preprocessed pictures have red, green and blue three-color dimension input. The matconvnet code based on the Matlab platform below is an example of a preprocessing step.
The read image is first normalized and processed,
img0＝imresize(img0，net.meta.normalization.imageSize(1：2))；
the mean value of the image is then subtracted,
img0＝img0-net.meta.normalization.averagelmage；
s200: and (3) establishing a deep convolutional neural network by utilizing a convolutional layer, a pooling layer, a batch normalization and activation function ReLu assisted by a residual error structure and an inclusion structure.
The GPU has excellent parallel computing capability, so that great help can be provided for big data computation, and in the process, the GPU is used for training and deducing the deep convolutional neural network.
And (3) rolling layers: the input images are convoluted by using different convolution kernels to obtain a feature map, and each convolution kernel detects specific features on all positions on the input feature map, so that weight sharing on the same input feature map is realized. Each input feature map element is x, the weight connecting the input feature map element and the output feature map element is w, the output feature map element is y, the bias parameter is b, f is an activation function, and the formula is as follows:
the activation function adopts ReLu, and the formula is as follows:
f(x)＝max(0，x)
a pooling layer: unimportant samples in the feature map can be removed through down-sampling, training parameters are reduced, and therefore the degree of fitting of the over-model is reduced. Here we use global average pooling (global average pooling) to average the ensemble for each feature map.
Random deactivation layer (Dropout): in the training process of the deep learning network, the neural unit of the deep learning network is temporarily discarded from the network according to a certain probability, so that overfitting can be effectively prevented, and the value is 0.8.
Residual structure: the residual structure means that each residual module has two paths, one path is an input feature, the other path is a residual obtained by performing convolution operation on the input feature for multiple times, and finally, the input feature and the residual are added:
H(x)＝F(x)+x
x represents input features, f (x) represents residual features, and h (x) represents residual module output features.
The inclusion structure: the structure formed by convolution layers (such as 1 × 1, 3 × 3) with different sizes in parallel increases the width of the neural network and the adaptability of the network nerves to the scale.
The Hybrid module combining the residual structure and the inclusion structure makes the neural network deeper and wider.
S300: and (5) taking the picture learning sample obtained in the step (S100) as training data, training a convolutional neural network model, and obtaining network model parameters. In the process of training the convolutional neural network, a supervised training mode and a back propagation algorithm are adopted to update the weight. The process is as follows:
calculating the error between the sample output characteristic and the theoretical label value:
the error is transmitted to each layer along the reverse path of the convolutional neural network, and the weight is updated according to a gradient descent algorithm:
eta is the learning rate, delta is the error term, and the weight update formula is as follows:
Δwij＝ηδjxij
with the continuous update of the convolution kernel weight value, the output error of the convolution neural network is continuously reduced and finally tends to be stable. If the pre-trained network is adopted, the fine-tuning technology is required to be assisted.
S400: and (5) extracting the characteristics of the training data by using the trained convolutional neural network model in the step (S300), and training by using a multivariate classifier to obtain a multivariate classification model. Here, a random forest algorithm is used as an example, and a training set is used to generate a plurality of classification decision trees, and the decision trees form a random forest. In the prediction process, each tree can predict a result, and finally, the class with the most classified results in all the decision trees is taken as the final result, the splitting index adopted for generating the classified decision trees is a kini coefficient, and the definition formula is as follows:
in the classification problem, there is a class k, and the probability that a sample point belongs to the kth class is pkThe sample capacity is | D |, | C |kIs of class CkThe number of samples.
S500: referring to fig. 3, a human face image is input by a subject at a terminal, the human face image is transmitted to a cloud computing platform, features of the human face image to be classified are extracted by using a convolutional neural network model in S300, then a classifier model obtained in S400 is used to obtain a disease class to which the human face image to be classified belongs, and the disease class is returned to the subject. The disease types which can be detected and screened in S500 at present comprise hyperthyroidism, hypothyroidism, hansheng disease, acromegaly, renal failure, neonatal pathological jaundice, Down' S syndrome, thalassemia, DiGeorge syndrome and mitral stenosis. More and more disease categories will be added to this invention in the future.
The above description is only an embodiment of the present invention, and not intended to limit the scope of the present invention, and all modifications of equivalent structures and equivalent processes performed by the present specification and drawings, or directly or indirectly applied to other related technical fields, are included in the scope of the present invention.
Claims (5)
1. A deep learning face diagnosis system, which detects and screens diseases through human face features, is characterized by comprising the following steps:
s100: collecting a face image sample with a disease diagnosis result label, and establishing a face image database with the disease diagnosis label, wherein the types of the disease diagnosis label comprise hyperthyroidism, hypothyroidism, hansheng disease, acromegaly, renal failure, neonatal pathological jaundice, down syndrome, thalassemia, diqiao syndrome and mitral stenosis;
s200: constructing a deep convolutional neural network, wherein the neural network comprises an input layer, a convolutional layer, a first pooling layer, a plurality of combination modules, a second pooling layer, a random inactivation layer (dropout) and a classification layer which are sequentially connected; the combined module consists of a mixing module followed by a dimension reduction (reduction) module; in the mixing module, an excitation function layer is connected firstly, then the mixing module is divided into two paths, one path is a direct output characteristic, the other path is divided into three parallel sub-paths, each sub-path comprises convolution layers with different numbers, the outputs of the three parallel sub-paths are spliced together to be used as the input of the other convolution layer, finally the direct output characteristic and the output characteristic of the other convolution layer are added, and then an excitation function is used for processing; the mixing module can increase the adaptability to the scale, avoid the accuracy reduction along with the deepening of the network and increase the depth and the width of the network structure;
s300: preprocessing the image acquired in the step S100, taking the preprocessed image and a label corresponding to the preprocessed image as training data, and training a deep convolutional neural network model to obtain network model parameters; extracting the characteristics of training data by using a trained S200 deep convolutional neural network model, and carrying out supervised training on a random forest classifier by using the extracted characteristics and a corresponding label to obtain a random forest classification model;
s400: and inputting a face image by a subject at a terminal, transmitting the face image to a cloud computing platform, extracting the features of the face image to be classified by using the trained deep convolutional neural network, obtaining the disease category to which the face image to be classified belongs by using the forest classifier model obtained in the S300, and returning the disease category to the subject.
2. The deep learning interview system of claim 1 wherein said step S100 further comprises categorizing the database into three broad categories according to skin tone: a database of Niger Roman race, Mongolian race, and Caucasian race; the preprocessed images in step S300 all have the same size and three color dimension inputs of red, green and blue,
the preprocessing step of matconvnet codes based on the Matlab platform comprises the following steps:
firstly, normalizing the read image:
img0＝imresize(img0，net.meta.normalization.imageSize(1∶2))
the image mean is then subtracted:
img0＝img0-net.meta.normalization.averageImage。
3. the deep learning face diagnosis system according to claim 1, wherein the implementation of step S200 uses a hybrid module combining a residual structure and an inclusion structure in the network construction, and the output formula of the residual structure is as follows:
H(x)＝F(x)+x
x represents input features, F (x) represents residual features, H (x) represents residual module output features;
the output formula of the convolution layer in the deep convolutional neural network is as follows:
each input characteristic diagram element is x, the weight value connecting the input characteristic diagram element and the output characteristic diagram element is w, the output characteristic diagram element is y, the bias parameter is b, and f is an activation function;
the ReLu function is used as the activation function in the deep convolutional neural network, and the output formula is as follows:
f(x)＝max(0，x)
the second pooling layer uses a global average pooling method, which is to calculate an overall average value for each feature map, and the output formula is as follows:
wherein, ykRepresenting the global average pooled output value, x, from the kth feature mapkpqRepresents the element at (p, q) in the kth feature map region R, | R | represents the number of all elements of the kth feature map;
the purpose of the random deactivation layer (dropout) is to prevent overfitting.
4. The deep learning surface diagnosis system of claim 1, wherein the deep learning surface diagnosis system of step S300 is implemented by using a Graphics Processing Unit (GPU) to train a deep convolutional neural network by adopting a supervised training mode and a back propagation algorithm to update weights, and the process is as follows:
calculating the error between the sample output characteristic and the theoretical label value:
the error is transmitted to each layer along the reverse path of the convolutional neural network, and the weight is updated according to a gradient descent algorithm:
eta is the learning rate, delta is the error term, and the weight update formula is as follows:
Δwij＝ηδjxij
extracting image features and label training by using a convolutional neural network to generate a plurality of classification decision trees, wherein the decision trees form a random forest; during prediction, each tree can predict a result, and finally, the class with the most classified results in all the decision trees is taken as a final result, and the splitting index adopted for generating the classified decision trees is a kini coefficient:
the probability that a sample point belongs to class k is pkThe sample capacity is | D |, | C |kIs of class CkIn this embodiment, k is 11.
5. The deep learning face diagnosis system according to claim 1, wherein the system comprises a terminal and a cloud computing platform; the terminal comprises a face image acquisition module, an information sending module and an information receiving module; the cloud computing platform comprises an image receiving module, a face image database with a disease diagnosis result label, a deep convolutional neural network training module, a deep convolutional neural network classification prediction module and a classification prediction result information sending module; the user can input the face image to the cloud computing platform through the terminal equipment including a desktop computer, a smart phone, a tablet computer and a notebook computer, the facial features are automatically extracted through the deep convolutional neural network, the classification model is used for disease detection, and the system returns a disease detection classification result to the terminal equipment.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2017103453755 | 2017-05-03 | ||
| CN201710345375 | 2017-05-03 |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108806792A CN108806792A (en) | 2018-11-13 |
| CN108806792B true CN108806792B (en) | 2022-01-04 |
Family
ID=64095168
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711255031.1A Active CN108806792B (en) | 2017-05-03 | 2017-11-25 | Deep learning face diagnosis system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108806792B (en) |
Families Citing this family (21)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109741823A (en) * | 2018-11-23 | 2019-05-10 | 杭州电子科技大学 | A kind of pneumothorax aided diagnosis method based on deep learning |
| JP7395604B2 (en) * | 2018-12-04 | 2023-12-11 | ジャン・ルオウェイ | Automatic image-based skin diagnosis using deep learning |
| CN109685968A (en) * | 2018-12-15 | 2019-04-26 | 西安建筑科技大学 | A kind of the identification model building and recognition methods of the banknote image defect based on convolutional neural networks |
| CN109766920A (en) * | 2018-12-18 | 2019-05-17 | 任飞翔 | Article characteristics Model Calculating Method and device based on deep learning |
| CN109785311B (en) * | 2019-01-14 | 2021-06-04 | 深圳和而泰数据资源与云技术有限公司 | Disease diagnosis device, electronic equipment and storage medium |
| CN109920538B (en) * | 2019-03-07 | 2022-11-25 | 中南大学 | Zero sample learning method based on data enhancement |
| CN111951950B (en) * | 2019-05-16 | 2023-11-24 | 中国科学院自动化研究所 | Three-dimensional data medical classification system based on deep learning |
| CN110363075B (en) * | 2019-06-03 | 2020-03-20 | 南京特殊教育师范学院 | Suspicious ill face detection system based on big data server |
| CN111353520A (en) * | 2019-07-03 | 2020-06-30 | 孙彤 | Method for identifying vermilion ink time sequence |
| CN111104961A (en) * | 2019-10-31 | 2020-05-05 | 太原理工大学 | Method for classifying breast cancer based on improved MobileNet network |
| CN111489325B (en) * | 2019-12-16 | 2023-07-28 | 杭州电子科技大学 | Instant identification device for hyperthyroidism disease and control method thereof |
| CN111341459A (en) * | 2020-02-28 | 2020-06-26 | 上海交通大学医学院附属上海儿童医学中心 | Training method of classified deep neural network model and genetic disease detection method |
| CN111951934A (en) * | 2020-08-20 | 2020-11-17 | 陈文立 | Novel acromegaly screening system and screening method thereof |
| CN111803024B (en) * | 2020-08-27 | 2022-10-04 | 张桦 | System for measuring eyeball protrusion degree based on deep learning algorithm |
| CN112163512A (en) * | 2020-09-25 | 2021-01-01 | 杨铠郗 | Autism spectrum disorder face screening method based on machine learning |
| CN112687393A (en) * | 2020-12-29 | 2021-04-20 | 康奥生物科技(天津)股份有限公司 | Disease prediction system based on big data |
| CN113012823A (en) * | 2021-03-30 | 2021-06-22 | 海口讯婕城医疗科技有限公司 | Cloud platform system for electrocardiograph |
| CN113139597B (en) * | 2021-04-19 | 2022-11-04 | 中国人民解放军91054部队 | Statistical thought-based image distribution external detection method |
| CN113284613A (en) * | 2021-05-24 | 2021-08-20 | 暨南大学 | Face diagnosis system based on deep learning |
| CN113625094A (en) * | 2021-09-11 | 2021-11-09 | 沈阳工业大学 | Subway platform screen door fault diagnosis system |
| WO2023161956A1 (en) * | 2022-02-28 | 2023-08-31 | Saanvi Mehra | Device and method for assessing a subject for down syndrome |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102426652A (en) * | 2011-10-10 | 2012-04-25 | 北京工业大学 | Traditional Chinese medicine face color identifying and retrieving method based on image analysis |
| CN102973253A (en) * | 2012-10-31 | 2013-03-20 | 北京大学 | Method and system for monitoring human physiological indexes by using visual information |
| JP2014182424A (en) * | 2013-03-18 | 2014-09-29 | Seiko Epson Corp | Printing equipment, printing system, and printed matter |
| CN104699931A (en) * | 2013-12-09 | 2015-06-10 | 广州华久信息科技有限公司 | Neural network blood pressure prediction method and mobile phone based on human face |
| CN105590102A (en) * | 2015-12-30 | 2016-05-18 | 中通服公众信息产业股份有限公司 | Front car face identification method based on deep learning |
-
2017
- 2017-11-25 CN CN201711255031.1A patent/CN108806792B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102426652A (en) * | 2011-10-10 | 2012-04-25 | 北京工业大学 | Traditional Chinese medicine face color identifying and retrieving method based on image analysis |
| CN102973253A (en) * | 2012-10-31 | 2013-03-20 | 北京大学 | Method and system for monitoring human physiological indexes by using visual information |
| JP2014182424A (en) * | 2013-03-18 | 2014-09-29 | Seiko Epson Corp | Printing equipment, printing system, and printed matter |
| CN104699931A (en) * | 2013-12-09 | 2015-06-10 | 广州华久信息科技有限公司 | Neural network blood pressure prediction method and mobile phone based on human face |
| CN105590102A (en) * | 2015-12-30 | 2016-05-18 | 中通服公众信息产业股份有限公司 | Front car face identification method based on deep learning |
Non-Patent Citations (1)
| Title |
|---|
| 基于面象特征的中医体质自动辨识***研究;梁玉梅;《中国优秀硕士学位论文全文数据库 医药卫生科技辑》;20170315(第3期);E056-67 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108806792A (en) | 2018-11-13 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108806792B (en) | Deep learning face diagnosis system | |
| Sahu et al. | FINE_DENSEIGANET: Automatic medical image classification in chest CT scan using Hybrid Deep Learning Framework | |
| Salido et al. | Using deep learning to detect melanoma in dermoscopy images | |
| CN110276745B (en) | Pathological image detection algorithm based on generation countermeasure network | |
| CN109117864A (en) | Coronary heart disease risk prediction technique, model and system based on heterogeneous characteristic fusion | |
| CN108257135A (en) | The assistant diagnosis system of medical image features is understood based on deep learning method | |
| CN109493342B (en) | Skin disease picture lesion type classification method based on deep learning | |
| CN111223553A (en) | Two-stage deep migration learning traditional Chinese medicine tongue diagnosis model | |
| CN107516312A (en) | A kind of Chinese medicine complexion automatic classification method using shallow-layer neutral net | |
| CN113610118B (en) | Glaucoma diagnosis method, device, equipment and method based on multitasking course learning | |
| Singh et al. | Collaboration of features optimization techniques for the effective diagnosis of glaucoma in retinal fundus images | |
| CN114398979A (en) | Ultrasonic image thyroid nodule classification method based on feature decoupling | |
| CN112668486A (en) | Method, device and carrier for identifying facial expressions of pre-activated residual depth separable convolutional network | |
| Elayaraja et al. | An efficient approach for detection and classification of cancer regions in cervical images using optimization based CNN classification approach | |
| CN111462082A (en) | Focus picture recognition device, method and equipment and readable storage medium | |
| CN117010971B (en) | Intelligent health risk providing method and system based on portrait identification | |
| CN110992309B (en) | Fundus image segmentation method based on deep information transfer network | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans | |
| Kotiyal et al. | Diabetic retinopathy binary image classification using PySpark | |
| CN115661185A (en) | Fundus image blood vessel segmentation method and system | |
| Yasmin et al. | Impact of fuzziness for skin lesion classification with transformer-based model | |
| Wu et al. | Mscan: Multi-scale channel attention for fundus retinal vessel segmentation | |
| Charulatha et al. | Computer-Aided Design for skin disease Identification and Categorization Using Image Processing | |
| Hatode et al. | Evolution and Testimony of Deep Learning Algorithm for Diabetic Retinopathy Detection | |
| CN110517783A (en) | Model training method, device and heart disease diagnosis device, equipment, medium |
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 |