CN107862694A - A kind of hand-foot-and-mouth disease detecting system based on deep learning - Google Patents
A kind of hand-foot-and-mouth disease detecting system based on deep learning Download PDFInfo
- Publication number
- CN107862694A CN107862694A CN201711376026.6A CN201711376026A CN107862694A CN 107862694 A CN107862694 A CN 107862694A CN 201711376026 A CN201711376026 A CN 201711376026A CN 107862694 A CN107862694 A CN 107862694A
- Authority
- CN
- China
- Prior art keywords
- hand
- foot
- mouth disease
- focus
- module
- 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.)
- Pending
Links
- 208000020061 Hand, Foot and Mouth Disease Diseases 0.000 title claims abstract description 95
- 208000025713 Hand-foot-and-mouth disease Diseases 0.000 title claims abstract description 95
- 238000013135 deep learning Methods 0.000 title claims abstract description 21
- 238000012549 training Methods 0.000 claims abstract description 84
- 238000003745 diagnosis Methods 0.000 claims abstract description 34
- 238000001514 detection method Methods 0.000 claims abstract description 29
- 238000003062 neural network model Methods 0.000 claims abstract description 28
- 238000013528 artificial neural network Methods 0.000 claims abstract description 18
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 4
- 238000012545 processing Methods 0.000 claims description 41
- 238000004458 analytical method Methods 0.000 claims description 18
- 238000012937 correction Methods 0.000 claims description 18
- 230000003902 lesion Effects 0.000 claims description 10
- 206010027336 Menstruation delayed Diseases 0.000 claims description 7
- 238000007621 cluster analysis Methods 0.000 claims description 6
- 230000007935 neutral effect Effects 0.000 claims description 4
- 230000008859 change Effects 0.000 claims description 3
- 230000001537 neural effect Effects 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 3
- 238000005516 engineering process Methods 0.000 abstract description 8
- 238000012360 testing method Methods 0.000 abstract description 2
- 238000000034 method Methods 0.000 description 11
- 238000013473 artificial intelligence Methods 0.000 description 6
- 230000036541 health Effects 0.000 description 5
- 238000005457 optimization Methods 0.000 description 4
- 238000006243 chemical reaction Methods 0.000 description 3
- 210000001367 artery Anatomy 0.000 description 2
- 201000010099 disease Diseases 0.000 description 2
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 210000004218 nerve net Anatomy 0.000 description 2
- 201000009030 Carcinoma Diseases 0.000 description 1
- 206010033799 Paralysis Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 210000004556 brain Anatomy 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 239000003814 drug Substances 0.000 description 1
- 229940079593 drug Drugs 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000029058 respiratory gaseous exchange Effects 0.000 description 1
- 208000024891 symptom Diseases 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Classifications
-
- 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
- G06T7/70—Determining position or orientation of objects or cameras
- G06T7/73—Determining position or orientation of objects or cameras using feature-based methods
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/30088—Skin; Dermal
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Medical Informatics (AREA)
- Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
- Radiology & Medical Imaging (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A kind of hand-foot-and-mouth disease detecting system based on deep learning, including neural network model training module and hand-foot-and-mouth disease detection module；The neural network model training module is based on hand-foot-and-mouth disease focus sample set, constructs convolutional neural networks model, by analyzing the image in hand-foot-and-mouth disease focus sample set, obtains the Neural Network Diagnosis model of hand-foot-and-mouth disease；The image of the hand-foot-and-mouth disease detection module based on input, is judged using Neural Network Diagnosis model, obtains hand-foot-and-mouth disease diagnostic result.The system diagnoses depth learning technology applied to hand-foot-and-mouth disease automatically, and the key problem that hand-foot-and-mouth disease detects is converted into target detection problems, and by independently optimizing the accuracy of lifting testing result.
Description
Technical field：
The present invention relates to a kind of hand-foot-and-mouth disease detecting system based on deep learning.
Background technology：
With expanding economy, people are increasingly paid close attention to own health, to the demand of medical services and health service not
It is disconnected to increase, and China's medical and health services are largely also rested among the treatment service system of traditional sense.In recent years, depth
Study has become new frontline technology, is widely used in the fields such as graphical analysis, video parsing, automatic Pilot and unmanned plane,
The huge contradiction of current medical supply and demand promotes people that depth learning technology is applied into medical field, and be thus born " artificial intelligence
Can+medical treatment " technology, it refers to the artificial intelligence model based on deep learning, after forming model parameter by initialization, to warp
The data for crossing mark are trained, and model parameter is adjusted after there is error, then are aided in medical knowledge, by largely training it
After form accurately medical diagnosis model, realize the diagnosis to disease, bring the raising of medical efficiency and accuracy rate of diagnosis.
Artificial intelligence (AI) of today can be that medical science solves specific medical problem, and have been achieved for huge
Progress, realized by depth learning technology, be mainly used in medical robot, intellectual drug research and development, intelligent diagnosis and treatment, intelligent shadow
Ring the five big fields such as identification, intelligent health management, it might even be possible to solve many medically current insurmountable problems.Enumerate mesh
Preceding typical application case, such as the intelligent exoskeleton of Russian ExoAtlet companies can help paralysed patient to carry out daily life
Action living, Stanford University can successfully differentiate cutaneum carcinoma, and Chinese Airdoc identifies BDR ability
Suitable with front three oculist, the product Arterys Cardio DL under the Arterys house flags in the U.S. obtain FDA batches
Standard, Baidu release Baidu's medical treatment brain, and it is diagnosed and the diagnosis 80% of BJ Univ Hospital doctor is identical, and artificial intelligence application at present exists
Medical field has been a trend.
Because doctor is a partially empirical industry, and artificial intelligence quickly can learn sample in the data of magnanimity
Eigen, artificial intelligence application can be freed doctor in medical diagnosis from the heavy labor repeated.At present, will be deep
Degree learning method is applied to portable medical, realizes the online autodiagnosis of patient, is realized especially for the autodiagnosis platform of hand-foot-and-mouth disease
Field is still a blank.Prior art does not have the plan of solution to this.
The content of the invention：
The purpose of the present invention is aiming at disadvantages mentioned above existing for prior art, there is provided a kind of hand based on deep learning
Sufficient stomatosis detecting system, depth learning technology is diagnosed automatically applied to hand-foot-and-mouth disease, the key problem that hand-foot-and-mouth disease is detected
Target detection problems are converted into, and by independently optimizing the accuracy of lifting testing result, are solved present in prior art
Problem.
The present invention is that technical scheme is used by solving above-mentioned technical problem：
A kind of hand-foot-and-mouth disease detecting system based on deep learning, including neural network model training module and hand-foot-and-mouth disease
Detection module；
The neural network model training module is based on hand-foot-and-mouth disease focus sample set, constructs convolutional neural networks model,
By analyzing the image in hand-foot-and-mouth disease focus sample set, the Neural Network Diagnosis model of hand-foot-and-mouth disease is obtained；
The image of the hand-foot-and-mouth disease detection module based on input, is judged using Neural Network Diagnosis model, is obtained
Hand-foot-and-mouth disease diagnostic result.
Preferably, in addition to correction module, the correction module are used for medical worker's analysis corrections hand-foot-and-mouth disease diagnosis knot
Fruit, and amendment data feedback to neural network model training module, neural network model training module is entered based on amendment data
The optimization of row Neural Network Diagnosis model.
Preferably, the neural network model training module includes sample acquisition module, image processing module, training mould
Block；
The sample acquisition module is used to obtain the training image in hand-foot-and-mouth disease focus sample set；
Described image processing module is used to obtain normalized images to training image progress standardization processing；
The training module is used to analyze obtained normalized images, and combines in hand-foot-and-mouth disease focus sample set
Diagnostic data carry out continual analysis training, obtain Neural Network Diagnosis model.
Preferably, the hand-foot-and-mouth disease detection module includes diagnosis object acquisition module, image processing module, diagnosis mould
Block；
The diagnosis object acquisition module is used to obtain diagnostic image to be diagnosed；
Described image processing module is used to obtain normalized images to diagnostic image progress standardization processing；
The diagnostic module is used to analyze normalized images, and draws brothers' mouth according to neural network model analysis
Sick diagnostic result.
Preferably, the correction module examines the hand-foot-and-mouth disease after hand-foot-and-mouth disease diagnostic result, medical worker's analysis corrections
Disconnected result, diagnostic image give neural network model training module as amendment data feedback.
Preferably, the standardization processing includes：Hand-foot-and-mouth disease focus point is labeled, training figure has been obtained to described
As carrying out batch processing, including unified form, equalization and denoising, then carry out extracting candidate frame and pre-training；Brothers' mouth
The normalized images for being labeled as obtaining the training image of focus carry out focus characteristic mark, form lesion information mark
Label, the focus, which includes hand-foot-and-mouth disease early stage, mid-term and the diseased region in late period, the mark, also includes label information and focus
The coordinate of upper left angle point and bottom right angle point in the normalized images, the label information refer to target whether belong to focus with
And the classification information of focus developing stage；The mark is by K mean algorithms to making the initial time chosen by hand in data set by oneself
Select frame to carry out cluster analysis, the statistical law of candidate frame is found, to cluster number k as anchor number, with k cluster centre
Box wide high parameter, which is defined, corrects anchor, obtains and the most similar initial candidate frame parameter of focus shape in normalized images.
Preferably, the standardization processing includes：Hand-foot-and-mouth disease focus point is labeled, diagnostic graph has been obtained to described
As carrying out batch processing, including unified form, equalization and denoising, then carry out extracting candidate frame and pre-training；Brothers' mouth
The normalized images for being labeled as obtaining the diagnostic image of focus carry out focus characteristic mark, form lesion information mark
Label, the focus, which includes hand-foot-and-mouth disease early stage, mid-term and the diseased region in late period, the mark, also includes label information and focus
The coordinate of upper left angle point and bottom right angle point in the normalized images, the label information refer to target whether belong to focus with
And the classification information of focus developing stage；The mark is by K mean algorithms to making the initial time chosen by hand in data set by oneself
Select frame to carry out cluster analysis, the statistical law of candidate frame is found, to cluster number k as anchor number, with k cluster centre
Box wide high parameter, which is defined, corrects anchor, obtains and the most similar initial candidate frame parameter of focus shape in normalized images.
Preferably, the training module uses Darknet19 preceding 23 layer network, constructs an average pond layer and one
Full articulamentum, pre-training is carried out on hand-foot-and-mouth disease focus sample set first, parameter is then carried out on the training sample set
Fine setting, the image advanced features of hand-foot-and-mouth disease focus are obtained, are output to next layer of neural metwork training module.
Preferably, the diagnostic module uses YOLOv2 neutral nets, by extracting candidate region from normalized images,
Lesions position and classification information are predicted using entire image feature, directly learns the global information of image；The candidate frame
Object detection method, realized by the comprehensive score for screening candidate frame, the confidence level of each candidate frame and candidate frame are predicted
Classification information is multiplied, and obtains comprehensive score, then carries out non-maxima suppression processing, and as continuous iteration is in progress, parameter is uninterrupted
Prediction block, progressively close to true frame, the true frame positional information of final output and classification information.
Preferably, described image processing module carries out format discriminance to training image or diagnostic image, and according to standardization
Image request carries out format conversion, while calculates the resolution ratio of training image or diagnostic image, and to resolution ratio less than setting threshold
The standardization figure of value is reacquired.
Compared with prior art, it is an advantage of the invention that：Pass through separately positioned neural network model training module and brothers
Stomatosis detection module improves the validity of whole system operation, and Neural Network Diagnosis model can be continuously improved；It is logical
Setting correction module is crossed, improves contiguity therebetween, the source of sample set is substantially expanded, so as to improve nerve net
The accuracy of network diagnostic model；The present invention can carry out effective detection to the image to be checked under different scenes, can realize brothers
The online autodiagnosis of stomatosis, while accuracy rate is improved, greatly improves detection speed, there is provided a kind of to classify and position
The new approaches combined, gratifying effect is achieved in hand-foot-and-mouth disease object detection field.
Brief description of the drawings：
Fig. 1 is the schematic flow sheet of the present invention.
Fig. 2 is the schematic flow sheet of image processing module.
Fig. 3 is diagnostic result schematic diagram.
Embodiment：
For the technical characterstic for illustrating this programme can be understood, below by embodiment, and its accompanying drawing is combined, to this hair
It is bright to be described in detail.
As Figure 1-3, a kind of hand-foot-and-mouth disease detecting system based on deep learning, including neural network model training mould
Block and hand-foot-and-mouth disease detection module；
The neural network model training module is based on hand-foot-and-mouth disease focus sample set, constructs convolutional neural networks model,
By analyzing the image in hand-foot-and-mouth disease focus sample set, the Neural Network Diagnosis model of hand-foot-and-mouth disease is obtained；
The image of the hand-foot-and-mouth disease detection module based on input, is judged using Neural Network Diagnosis model, is obtained
Hand-foot-and-mouth disease diagnostic result.
Also include correction module, the correction module is used for medical worker's analysis corrections hand-foot-and-mouth disease diagnostic result, and will
Data feedback is corrected to neural network model training module, neural network model training module is based on amendment data and carries out nerve net
The optimization of network diagnostic model.
The neural network model training module includes sample acquisition module, image processing module, training module,
The sample acquisition module is used to obtain the training image in hand-foot-and-mouth disease focus sample set；
Described image processing module is used to obtain normalized images to training image progress standardization processing；
The training module is used to analyze obtained normalized images, and combines in hand-foot-and-mouth disease focus sample set
Diagnostic data carry out continual analysis training, obtain Neural Network Diagnosis model.
The hand-foot-and-mouth disease detection module includes diagnosis object acquisition module, image processing module, diagnostic module,
The diagnosis object acquisition module is used to obtain diagnostic image to be diagnosed；
Described image processing module is used to obtain normalized images to diagnostic image progress standardization processing；
The diagnostic module is used to analyze normalized images, and draws brothers' mouth according to neural network model analysis
Sick diagnostic result.
The correction module by the hand-foot-and-mouth disease diagnostic result after hand-foot-and-mouth disease diagnostic result, medical worker's analysis corrections,
Diagnostic image gives neural network model training module as amendment data feedback.
The standardization processing includes：Hand-foot-and-mouth disease focus point is labeled, the training image that obtained is carried out
Batch processing, including unified form, equalization and denoising, then carry out extracting candidate frame and pre-training；The hand-foot-and-mouth disease focus
The normalized images that are obtained to the training image of being labeled as carry out focus characteristic mark, form lesion information label, it is described
Focus includes hand-foot-and-mouth disease early stage, mid-term and the diseased region in late period, and the mark also includes label information and focus in the rule
The coordinate of upper left angle point and bottom right angle point in generalized image, the label information refer to whether target belongs to focus and focus
The classification information of developing stage；The mark is entered by K mean algorithms to the initial candidate frame chosen by hand in self-control data set
Row cluster analysis, the statistical law of candidate frame is found, to cluster number k as anchor number, with k cluster centre box width
High parameter, which is defined, corrects anchor, obtains and the most similar initial candidate frame parameter of focus shape in normalized images.
The standardization processing includes：Hand-foot-and-mouth disease focus point is labeled, the diagnostic image that obtained is carried out
Batch processing, including unified form, equalization and denoising, then carry out extracting candidate frame and pre-training；The hand-foot-and-mouth disease focus
The normalized images that are obtained to the diagnostic image of being labeled as carry out focus characteristic mark, form lesion information label, it is described
Focus includes hand-foot-and-mouth disease early stage, mid-term and the diseased region in late period, and the mark also includes label information and focus in the rule
The coordinate of upper left angle point and bottom right angle point in generalized image, the label information refer to whether target belongs to focus and focus
The classification information of developing stage；The mark is entered by K mean algorithms to the initial candidate frame chosen by hand in self-control data set
Row cluster analysis, the statistical law of candidate frame is found, to cluster number k as anchor number, with k cluster centre box width
High parameter, which is defined, corrects anchor, obtains and the most similar initial candidate frame parameter of focus shape in normalized images.
The training module uses Darknet19 preceding 23 layer network, constructs an average pond layer and a full connection
Layer, carries out pre-training on hand-foot-and-mouth disease focus sample set first, small parameter perturbations is then carried out on the training sample set, are obtained
The image advanced features of hand-foot-and-mouth disease focus are obtained, are output to next layer of neural metwork training module.
The diagnostic module uses YOLOv2 neutral nets, and by extracting candidate region from normalized images, utilization is whole
Width characteristics of image predicts lesions position and classification information, directly learns the global information of image；The target inspection of the candidate frame
Survey method, realized by the comprehensive score for screening candidate frame, the classification of the confidence level of each candidate frame and candidate frame prediction is believed
Manner of breathing multiplies, and obtains comprehensive score, then carries out non-maxima suppression processing, as continuous iteration is in progress, the uninterrupted prediction block of parameter,
Closest to true frame, the true frame positional information of final output and classification information.
Described image processing module carries out format discriminance to training image or diagnostic image, and is required according to normalized images
Format conversion is carried out, while calculates the resolution ratio of training image or diagnostic image, and is less than the specification of given threshold to resolution ratio
Change figure to be reacquired.
The following steps application present invention can be used：
1st, the training image in hand-foot-and-mouth disease focus sample set is obtained；
2nd, standardization processing is carried out to training image and obtains normalized images；
3rd, obtained normalized images are analyzed, and combines the diagnostic data in hand-foot-and-mouth disease focus sample set and carry out
Continual analysis is trained, and obtains Neural Network Diagnosis model；
4th, diagnostic image to be diagnosed is obtained；
5th, standardization processing is carried out to diagnostic image and obtains normalized images；
6th, obtained normalized images are analyzed, and hand-foot-and-mouth disease diagnosis knot is drawn according to neural network model analysis
Fruit；
7th, medical worker's analysis corrections hand-foot-and-mouth disease diagnostic result, and amendment data feedback to neural network model is trained
Module, neural network model training module carry out the optimization of Neural Network Diagnosis model based on amendment data.
Below by an embodiment, illustrate implementation process of the invention：Open-Source Tools of the embodiment based on deep learning
TensorFlow realizes that the hardware environment of operation is NVIDIA GTX1080, and implementing procedure figure is as shown in figure 1, specific implement
Step is as follows：
In the present embodiment, focus sample set input module include a variety of disease locus of hand-foot-and-mouth disease focus, Symptoms,
Picture under light conditions, have benefited from YOLOv2 and comprise only convolutional layer and pond layer, samples pictures can include a variety of image chis
It is very little.
In the present embodiment, image processing module is as shown in Fig. 2 divide the hand-foot-and-mouth disease detection sample image of input
Analysis is handled, including expands training sample set, hand-foot-and-mouth disease focus point mark, batch standardization, extraction candidate frame and pre-training.
Expand training sample set, specific extending method includes image translation, horizontal/vertical overturns, rotates, cuts out at random,
Add Gaussian noise, Fuzzy Processing etc., and adjust tone, saturation degree and depth of exposure, if sample set does not have representativeness, it is difficult to select
Good feature is selected out, by expanding training sample set, increases the diversity of sample, to realize the robust for improving final Detection results
The purpose of property；Data set is subjected to batch standardization, including unified form, equalization, denoising etc..
It is K mean cluster method to be used on the Bounding Box of training set, automatically to extract candidate frame specific method
Get the statistical law of "current" model candidate frame, in k=5 can preferably balance model complexity and high recall rate, obtain
To with the most similar anchor dimensional parameters of training sample set focus shape.In order that preferably matching is true by Bounding Box
Frame, use following distance metric：
D (box, cnetroid)=1-IOU (box, centroid)
Wherein IOU is the friendship of target frame and true frame and ratio：
Wherein gt represents true frame, and centroid represents cluster centre frame.
Training using Darknet-19 networks as YOLOv2 foundation characteristic extraction network, comprising 19 Ge Juan basic units, 5
Maximum pond layer.Similar to VGG-16, Web vector graphic substantial amounts of 3*3 convolution kernels, each pond are double by port number.Use for reference
The thought of NIN (Network in Network), Web vector graphic are global average pond (Global Average Pooling)
Prediction, 1*1 convolution kernel is used between 3*3 convolution kernel, represented to compressive features.To reduce the training time, pre-training
The convolutional layer model parameter that method initializes to obtain with Darknet19 network trainings, takes into account accuracy rate and complexity, has reached net
The lifting of network performance.
Neural network model training module uses YOLOv2 neutral nets, and target detection is carried out to candidate frame.YOLOv2 god
Overfeat and SSD thought have been used for reference through network, while has introduced the anchor point (anchor) in Faster RCNN, has been different from
It is a kind of deep learning based on recurrence based on region candidate (Region Proposal) deep learning object detection method
Object detection method, detection speed is improved, realize and detect end to end, be a kind of convolution model target detection complete in real time
Network.YOLOv2 is by last layer of 1*1 of Darknet-19 convolutional layer instead three layers of 3*3 convolutional layer, port number 1024,
And increase 1*1 convolutional layer after each convolutional layer, output dimension is classification number needed for detection, is examined for hand-foot-and-mouth disease
Survey, containing 3 kinds of early stage, mid-term, late period classifications, corresponding 5 target frames, understood to contain 5 class anchor points by cluster result, need 40 altogether
Passage.YOLOv2 provides a kind of mechanism classified and position joint training, uses the sample set while learning objective of tape label
Confine position and classification judges.Different from YOLO method, YOLOv2 prediction coordinate bits of the bounding box with respect to grid cell
Put, each cell predicts that 5 bounding box, each bounding box include 5 predicted values：tx,ty,tw,thAnd to,
Scope changes between 0 to 1, and this makes parameter be easier to learn, and network model is more stable.Wherein, tx,tyIt is exactly bounding
Box centre coordinate, cx,cyIt is offsets of the grid cell relative to the image upper left corner；P is used respectivelywAnd phRepresent priori
Bounding box width and height, then prediction result be：
bx=σ (tx)+cx
by=σ (ty)+cy
P (object) * IOU (b, object)=σ (to)
Wherein, YOLOv2 carries out batch standardization after each convolutional layer, is not increasing the feelings of other normalized forms
Under condition, the constringent appearance for being obviously improved, avoiding over-fitting situation is just brought.
Input picture may have format differences because of image input module uploading device difference, add image pre-processing module
Legitimacy differentiation is carried out to input picture, whether check image form, which meets, requires, enters row format conversion to ineligible.
Meanwhile image definition evaluation is significant for graphical analysis and identification, input picture there may be focusing and obscure, no
Detect and position beneficial to later stage focus.The fog-level that image is carried out in this module is calculated, and excessively fuzzy picture is obtained again
Take.Wherein the calculating of fog-level is realized by Laplce's variance algorithm, suitable by adjusting one using OPENCV frameworks
Threshold value picture is divided into two classes：It is clear and fuzzy, re-entered for the picture requirement for being judged as fuzzy.
The data that neural network model training module is transmitted using the model data of training module to pretreatment module are carried out
Calculate, target detection is carried out by candidate frame, completes the detection and positioning of hand-foot-and-mouth disease focus, the focus coordinate in output image
Return information and classification information.
In the present embodiment, correction module has used online difficult sample method for digging, main to consider hand-foot-and-mouth disease focus
Contain more simple sample in sample set, and difficult sample is less.Health care professional is carried out by system to diagnostic result
Differentiate, missing inspection focus and focus classification inaccurate information be modified, and be committed to system, system classification error information again
Secondary to feed back to Neural Network Diagnosis model, the training and autonomous optimization that hand-foot-and-mouth disease Neural Network Diagnosis model can be achieved are learned
Practise, the discriminating power of Strengthens network, finally make training pattern more effective, constantly lifting diagnosis precision.
In the present embodiment, diagnostic result output is using target collimation mark as shown in figure 3, go out lesions position, and by position with
In classification information deposit output document.
Part is not described in detail by the present invention, is the known technology of those skilled in the art of the present technique.
Claims (10)
- A kind of 1. hand-foot-and-mouth disease detecting system based on deep learning, it is characterised in that：Including neural network model training module With hand-foot-and-mouth disease detection module；The neural network model training module is based on hand-foot-and-mouth disease focus sample set, constructs convolutional neural networks model, passes through Image in hand-foot-and-mouth disease focus sample set is analyzed, obtains the Neural Network Diagnosis model of hand-foot-and-mouth disease；The image of the hand-foot-and-mouth disease detection module based on input, is judged using Neural Network Diagnosis model, obtains brothers Stomatosis diagnostic result.
- A kind of 2. hand-foot-and-mouth disease detecting system based on deep learning according to claim 1, it is characterised in that：Also include Correction module, the correction module is used for medical worker's analysis corrections hand-foot-and-mouth disease diagnostic result, and will correct data feedback extremely Neural network model training module, neural network model training module carry out the excellent of Neural Network Diagnosis model based on amendment data Change.
- A kind of 3. hand-foot-and-mouth disease detecting system based on deep learning according to claim 1, it is characterised in that：The god Include sample acquisition module, image processing module, training module through network model training module；The sample acquisition module is used to obtain the training image in hand-foot-and-mouth disease focus sample set；Described image processing module is used to obtain normalized images to training image progress standardization processing；The training module is used to analyze obtained normalized images, and combines examining in hand-foot-and-mouth disease focus sample set Disconnected data carry out continual analysis training, obtain Neural Network Diagnosis model.
- A kind of 4. hand-foot-and-mouth disease detecting system based on deep learning according to claim 2, it is characterised in that：The hand Sufficient stomatosis detection module includes diagnosis object acquisition module, image processing module, diagnostic module；The diagnosis object acquisition module is used to obtain diagnostic image to be diagnosed；Described image processing module is used to obtain normalized images to diagnostic image progress standardization processing；The diagnostic module is used to analyze normalized images, and show that hand-foot-and-mouth disease is examined according to neural network model analysis Disconnected result.
- A kind of 5. hand-foot-and-mouth disease detecting system based on deep learning according to claim 4, it is characterised in that：The school Positive module is using the hand-foot-and-mouth disease diagnostic result after hand-foot-and-mouth disease diagnostic result, medical worker's analysis corrections, diagnostic image as repairing Correction data feeds back to neural network model training module.
- A kind of 6. hand-foot-and-mouth disease detecting system based on deep learning according to claim 3, it is characterised in that：The rule Generalized processing includes：Hand-foot-and-mouth disease focus point is labeled, batch processing, including system are carried out to the training image that obtained One form, equalization and denoising, then carry out extracting candidate frame and pre-training；The hand-foot-and-mouth disease focus is labeled as to the instruction Practice the normalized images that image obtains and carry out focus characteristic mark, form lesion information label, the focus includes hand-foot-and-mouth disease In early days, the diseased region in mid-term and late period, the mark also include label information and focus the upper left in the normalized images The coordinate of angle point and bottom right angle point, the label information refer to whether target belongs to the classification letter of focus and focus developing stage Breath；The mark carries out cluster analysis by K mean algorithms to the initial candidate frame chosen by hand in self-control data set, finds time The statistical law of frame is selected, to cluster number k as anchor number, is defined amendment by k cluster centre box wide high parameter Anchor, obtain and the most similar initial candidate frame parameter of focus shape in normalized images.
- A kind of 7. hand-foot-and-mouth disease detecting system based on deep learning according to claim 4, it is characterised in that：The rule Generalized processing includes：Hand-foot-and-mouth disease focus point is labeled, batch processing, including system are carried out to the diagnostic image that obtained One form, equalization and denoising, then carry out extracting candidate frame and pre-training；The hand-foot-and-mouth disease focus is labeled as examining described The normalized images that disconnected image obtains carry out focus characteristic mark, form lesion information label, and the focus includes hand-foot-and-mouth disease In early days, the diseased region in mid-term and late period, the mark also include label information and focus the upper left in the normalized images The coordinate of angle point and bottom right angle point, the label information refer to whether target belongs to the classification letter of focus and focus developing stage Breath；The mark carries out cluster analysis by K mean algorithms to the initial candidate frame chosen by hand in self-control data set, finds time The statistical law of frame is selected, to cluster number k as anchor number, is defined amendment by k cluster centre box wide high parameter Anchor, obtain and the most similar initial candidate frame parameter of focus shape in normalized images.
- A kind of 8. hand-foot-and-mouth disease detecting system based on deep learning according to claim 3, it is characterised in that：The instruction Practice preceding 23 layer network that module uses Darknet19, an average pond layer and a full articulamentum are constructed, first in brothers' mouth Pre-training is carried out on focus sample set, small parameter perturbations are then carried out on the training sample set, obtains hand-foot-and-mouth disease focus Image advanced features, be output to next layer of neural metwork training module.
- A kind of 9. hand-foot-and-mouth disease detecting system based on deep learning according to claim 4, it is characterised in that：It is described to examine Disconnected module uses YOLOv2 neutral nets, by extracting candidate region from normalized images, using entire image feature come pre- Lesions position and classification information are surveyed, directly learns the global information of image；The object detection method of the candidate frame, pass through screening The comprehensive score of candidate frame is realized, the confidence level of each candidate frame is multiplied with the classification information that candidate frame is predicted, integrated Score, then carry out non-maxima suppression processing, as continuous iteration is in progress, the uninterrupted prediction block of parameter, progressively close to true frame, The true frame positional information of final output and classification information.
- A kind of 10. hand-foot-and-mouth disease detecting system based on deep learning according to claim 3 or 4, it is characterised in that：Institute State image processing module and format discriminance is carried out to training image or diagnostic image, and require to turn into row format according to normalized images Change, while calculate the resolution ratio of training image or diagnostic image, and standardization figure of the resolution ratio less than given threshold is carried out Reacquire.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711376026.6A CN107862694A (en) | 2017-12-19 | 2017-12-19 | A kind of hand-foot-and-mouth disease detecting system based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711376026.6A CN107862694A (en) | 2017-12-19 | 2017-12-19 | A kind of hand-foot-and-mouth disease detecting system based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107862694A true CN107862694A (en) | 2018-03-30 |
Family
ID=61707000
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711376026.6A Pending CN107862694A (en) | 2017-12-19 | 2017-12-19 | A kind of hand-foot-and-mouth disease detecting system based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107862694A (en) |
Cited By (25)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108597602A (en) * | 2018-04-28 | 2018-09-28 | 中南大学 | A kind of label correction method towards dermatology data |
| CN108711172A (en) * | 2018-04-24 | 2018-10-26 | 中国海洋大学 | Unmanned plane identification based on fine grit classification and localization method |
| CN108734746A (en) * | 2018-05-07 | 2018-11-02 | 苏州苏相机器人智能装备有限公司 | Hand-foot-and-mouth disease detection device based on machine vision |
| CN108764329A (en) * | 2018-05-24 | 2018-11-06 | 复旦大学附属华山医院北院 | A kind of construction method of lung cancer pathology image data set |
| CN108765392A (en) * | 2018-05-20 | 2018-11-06 | 复旦大学 | A kind of digestive endoscope lesion detection and recognition methods based on sliding window |
| CN108846415A (en) * | 2018-05-22 | 2018-11-20 | 长沙理工大学 | The Target Identification Unit and method of industrial sorting machine people |
| CN108876765A (en) * | 2018-05-22 | 2018-11-23 | 塞伯睿机器人技术（长沙）有限公司 | The target locating set and method of industrial sorting machine people |
| CN108932502A (en) * | 2018-07-13 | 2018-12-04 | 希蓝科技（北京）有限公司 | A kind of electrocardiogram template classification model modification system and method for self study |
| CN108986913A (en) * | 2018-07-13 | 2018-12-11 | 希蓝科技（北京）有限公司 | A kind of optimization artificial intelligence cardiac diagnosis method and system |
| CN109243602A (en) * | 2018-08-23 | 2019-01-18 | 腾讯科技（深圳）有限公司 | Data identification method and device, storage medium and its corresponding electronic device |
| CN109523535A (en) * | 2018-11-15 | 2019-03-26 | 首都医科大学附属北京友谊医院 | A kind of preprocess method of lesion image |
| CN109615610A (en) * | 2018-11-13 | 2019-04-12 | 浙江师范大学 | A kind of medical band-aid flaw detection method based on YOLO v2-tiny |
| CN109934184A (en) * | 2019-03-19 | 2019-06-25 | 网易（杭州）网络有限公司 | Gesture identification method and device, storage medium, processor |
| CN110472581A (en) * | 2019-08-16 | 2019-11-19 | 电子科技大学 | A kind of cell image analysis method based on deep learning |
| CN110887336A (en) * | 2018-09-11 | 2020-03-17 | 东芝生活电器株式会社 | Article taking and placing management system of refrigerator and refrigerator |
| CN110974306A (en) * | 2019-12-17 | 2020-04-10 | 山东大学齐鲁医院 | System for discernment and location pancreas neuroendocrine tumour under ultrasonic endoscope |
| CN111161872A (en) * | 2019-12-03 | 2020-05-15 | 王洁 | Intelligent management system for child health |
| CN111281361A (en) * | 2020-03-09 | 2020-06-16 | 康瑞健康管理(杭州)有限公司 | Student health monitoring system based on big data |
| CN111353393A (en) * | 2020-02-19 | 2020-06-30 | 桂林电子科技大学 | Dog only detects and early warning system based on neural network |
| CN111598865A (en) * | 2020-05-14 | 2020-08-28 | 上海锘科智能科技有限公司 | Hand-foot-and-mouth disease detection method, device and system based on thermal infrared and RGB (red, green and blue) double-taking |
| CN112567385A (en) * | 2018-08-17 | 2021-03-26 | 孙永年 | Processing method, computer device and system for slicing or smear images |
| CN112786192A (en) * | 2021-01-18 | 2021-05-11 | 吾征智能技术(北京)有限公司 | Intelligent cognitive system, equipment and storage medium for hand-foot-and-mouth disease |
| CN112890767A (en) * | 2020-12-30 | 2021-06-04 | 浙江大学 | Automatic detection device and method for health state of mouth, hands and feet |
| CN114566273A (en) * | 2022-01-05 | 2022-05-31 | 中山大学孙逸仙纪念医院 | Method for constructing onychomycosis diagnosis model, diagnosis model and diagnosis device |
| CN116596861A (en) * | 2023-04-28 | 2023-08-15 | 中山大学 | Dental lesion recognition method, system, equipment and storage medium |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407699A (en) * | 2016-09-28 | 2017-02-15 | 湖南老码信息科技有限责任公司 | Coronary heart disease prediction method and prediction system based on incremental neural network model |
| CN107274406A (en) * | 2017-08-07 | 2017-10-20 | 北京深睿博联科技有限责任公司 | A kind of method and device of detection sensitizing range |
-
2017
- 2017-12-19 CN CN201711376026.6A patent/CN107862694A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106407699A (en) * | 2016-09-28 | 2017-02-15 | 湖南老码信息科技有限责任公司 | Coronary heart disease prediction method and prediction system based on incremental neural network model |
| CN107274406A (en) * | 2017-08-07 | 2017-10-20 | 北京深睿博联科技有限责任公司 | A kind of method and device of detection sensitizing range |
Non-Patent Citations (1)
| Title |
|---|
| 魏湧明 等: "基于YOLO V2的无人机航拍图像定位研究", 《激光与光电子学进展》 * |
Cited By (35)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108711172A (en) * | 2018-04-24 | 2018-10-26 | 中国海洋大学 | Unmanned plane identification based on fine grit classification and localization method |
| CN108711172B (en) * | 2018-04-24 | 2020-07-03 | 中国海洋大学 | Unmanned aerial vehicle identification and positioning method based on fine-grained classification |
| CN108597602A (en) * | 2018-04-28 | 2018-09-28 | 中南大学 | A kind of label correction method towards dermatology data |
| CN108597602B (en) * | 2018-04-28 | 2021-11-05 | 中南大学 | Label error correction method for skin medical data |
| CN108734746A (en) * | 2018-05-07 | 2018-11-02 | 苏州苏相机器人智能装备有限公司 | Hand-foot-and-mouth disease detection device based on machine vision |
| CN108734746B (en) * | 2018-05-07 | 2021-11-12 | 苏州苏相机器人智能装备有限公司 | Hand-foot-and-mouth disease detection device based on machine vision |
| CN108765392B (en) * | 2018-05-20 | 2022-03-18 | 复旦大学 | Digestive tract endoscope lesion detection and identification method based on sliding window |
| CN108765392A (en) * | 2018-05-20 | 2018-11-06 | 复旦大学 | A kind of digestive endoscope lesion detection and recognition methods based on sliding window |
| CN108846415A (en) * | 2018-05-22 | 2018-11-20 | 长沙理工大学 | The Target Identification Unit and method of industrial sorting machine people |
| CN108876765A (en) * | 2018-05-22 | 2018-11-23 | 塞伯睿机器人技术（长沙）有限公司 | The target locating set and method of industrial sorting machine people |
| CN108764329A (en) * | 2018-05-24 | 2018-11-06 | 复旦大学附属华山医院北院 | A kind of construction method of lung cancer pathology image data set |
| CN108986913A (en) * | 2018-07-13 | 2018-12-11 | 希蓝科技（北京）有限公司 | A kind of optimization artificial intelligence cardiac diagnosis method and system |
| CN108932502A (en) * | 2018-07-13 | 2018-12-04 | 希蓝科技（北京）有限公司 | A kind of electrocardiogram template classification model modification system and method for self study |
| CN112567385A (en) * | 2018-08-17 | 2021-03-26 | 孙永年 | Processing method, computer device and system for slicing or smear images |
| CN109243602A (en) * | 2018-08-23 | 2019-01-18 | 腾讯科技（深圳）有限公司 | Data identification method and device, storage medium and its corresponding electronic device |
| CN110887336B (en) * | 2018-09-11 | 2021-10-29 | 东芝生活电器株式会社 | Article taking and placing management system of refrigerator and refrigerator |
| CN110887336A (en) * | 2018-09-11 | 2020-03-17 | 东芝生活电器株式会社 | Article taking and placing management system of refrigerator and refrigerator |
| CN109615610A (en) * | 2018-11-13 | 2019-04-12 | 浙江师范大学 | A kind of medical band-aid flaw detection method based on YOLO v2-tiny |
| CN109523535B (en) * | 2018-11-15 | 2023-11-17 | 首都医科大学附属北京友谊医院 | Pretreatment method of lesion image |
| CN109523535A (en) * | 2018-11-15 | 2019-03-26 | 首都医科大学附属北京友谊医院 | A kind of preprocess method of lesion image |
| CN109934184A (en) * | 2019-03-19 | 2019-06-25 | 网易（杭州）网络有限公司 | Gesture identification method and device, storage medium, processor |
| CN110472581A (en) * | 2019-08-16 | 2019-11-19 | 电子科技大学 | A kind of cell image analysis method based on deep learning |
| CN111161872B (en) * | 2019-12-03 | 2022-08-23 | 王洁 | Intelligent management system for child health |
| CN111161872A (en) * | 2019-12-03 | 2020-05-15 | 王洁 | Intelligent management system for child health |
| CN110974306A (en) * | 2019-12-17 | 2020-04-10 | 山东大学齐鲁医院 | System for discernment and location pancreas neuroendocrine tumour under ultrasonic endoscope |
| CN111353393A (en) * | 2020-02-19 | 2020-06-30 | 桂林电子科技大学 | Dog only detects and early warning system based on neural network |
| CN111281361A (en) * | 2020-03-09 | 2020-06-16 | 康瑞健康管理(杭州)有限公司 | Student health monitoring system based on big data |
| CN111281361B (en) * | 2020-03-09 | 2024-09-10 | 明瞳健康管理(杭州)有限公司 | Student health monitoring system based on big data |
| CN111598865A (en) * | 2020-05-14 | 2020-08-28 | 上海锘科智能科技有限公司 | Hand-foot-and-mouth disease detection method, device and system based on thermal infrared and RGB (red, green and blue) double-taking |
| CN111598865B (en) * | 2020-05-14 | 2023-05-16 | 上海锘科智能科技有限公司 | Hand-foot-mouth disease detection method, device and system based on thermal infrared and RGB double-shooting |
| CN112890767A (en) * | 2020-12-30 | 2021-06-04 | 浙江大学 | Automatic detection device and method for health state of mouth, hands and feet |
| CN112786192A (en) * | 2021-01-18 | 2021-05-11 | 吾征智能技术(北京)有限公司 | Intelligent cognitive system, equipment and storage medium for hand-foot-and-mouth disease |
| CN114566273A (en) * | 2022-01-05 | 2022-05-31 | 中山大学孙逸仙纪念医院 | Method for constructing onychomycosis diagnosis model, diagnosis model and diagnosis device |
| CN116596861A (en) * | 2023-04-28 | 2023-08-15 | 中山大学 | Dental lesion recognition method, system, equipment and storage medium |
| CN116596861B (en) * | 2023-04-28 | 2024-02-23 | 中山大学 | Dental lesion recognition method, system, equipment and storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107862694A (en) | A kind of hand-foot-and-mouth disease detecting system based on deep learning | |
| CN111476292B (en) | Small sample element learning training method for medical image classification processing artificial intelligence | |
| CN107610087B (en) | Tongue coating automatic segmentation method based on deep learning | |
| Al-Bander et al. | Multiscale sequential convolutional neural networks for simultaneous detection of fovea and optic disc | |
| CN108364006B (en) | Medical image classification device based on multi-mode deep learning and construction method thereof | |
| CN109670510A (en) | A kind of gastroscopic biopsy pathological data screening system and method based on deep learning | |
| US20190311478A1 (en) | System and Method for Automatic Detection, Localization, and Semantic Segmentation of Anatomical Objects | |
| CN109858540B (en) | Medical image recognition system and method based on multi-mode fusion | |
| CN109584251A (en) | A kind of tongue body image partition method based on single goal region segmentation | |
| CN110705425A (en) | Tongue picture multi-label classification learning method based on graph convolution network | |
| CN109033978B (en) | Error correction strategy-based CNN-SVM hybrid model gesture recognition method | |
| CN110059656B (en) | Method and system for classifying white blood cells based on convolution countermeasure generation neural network | |
| CN114842238B (en) | Identification method of embedded breast ultrasonic image | |
| CN108053398A (en) | A kind of melanoma automatic testing method of semi-supervised feature learning | |
| CN110910377B (en) | Cerebral infarction MRI image identification method based on neural network | |
| CN106780453A (en) | A kind of method realized based on depth trust network to brain tumor segmentation | |
| CN114663426B (en) | Bone age assessment method based on key bone region positioning | |
| CN109767459A (en) | Novel ocular base map method for registering | |
| CN117975170B (en) | Medical information processing method and system based on big data | |
| CN111127423A (en) | Rice pest and disease identification method based on CNN-BP neural network algorithm | |
| CN111402231B (en) | Automatic evaluation system and method for lung CT image quality | |
| Bhimavarapu et al. | Analysis and characterization of plant diseases using transfer learning | |
| CN112036298A (en) | Cell detection method based on double-segment block convolutional neural network | |
| Zhao et al. | Pulmonary nodule detection based on multiscale feature fusion | |
| Cao et al. | 3D convolutional neural networks fusion model for lung nodule detection onclinical CT scans |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180330 |