CN109740654A - A kind of tongue body automatic testing method based on deep learning - Google Patents
A kind of tongue body automatic testing method based on deep learning Download PDFInfo
- Publication number
- CN109740654A CN109740654A CN201811596872.3A CN201811596872A CN109740654A CN 109740654 A CN109740654 A CN 109740654A CN 201811596872 A CN201811596872 A CN 201811596872A CN 109740654 A CN109740654 A CN 109740654A
- Authority
- CN
- China
- Prior art keywords
- tongue
- image
- model
- rectangle frame
- training
- 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
Landscapes
- Image Analysis (AREA)
- Measurement Of The Respiration, Hearing Ability, Form, And Blood Characteristics Of Living Organisms (AREA)
Abstract
The invention discloses a kind of tongue body automatic testing method based on deep learning comprising step 1: choose include tongue facial image, marking the tongue body position in image is rectangle frame；Step 2: facial image and rectangle frame in step 1 obtain model using object detection network Yolo training as training sample；Step 3: choosing the facial image comprising tongue as test set, obtain tongue image using the model inspection of step 2；Step 4: choosing tongue image and non-tongue image as the positive negative sample of training；Step 5: using the sample training DenseNet model in step 4, for the tongue image in selecting step 3 as test set, automatic detection obtains the highest image of tongue confidence level.The invention has the benefit that detecting by using the method for deep learning the tongue body in tongue image automatically in tongue body detection process, tongue image is analyzed and calculated without using complicated algorithm, remove man-machine interactively in detection process from, improves the degree of automation.
Description
Technical field
The present invention relates to deep learning methods to be applied to tongue body detection field, it particularly relates to which a kind of be based on depth
The image classification and object detection of habit are applied in tongue body detection, and realization is classified and detected to tongue body automatically.
Background technique
The traditional method of research and analysis tongue body is to see tongue by doctor's naked eyes, by the disconnected disease of experience, limits tongue body research
With the succession and development of analysis, also to realize that it is certain that the objectifying of tongue body research and analysis, quantification and automation are brought
It is difficult.
With the development of computer, in conjunction with the image processing method of computer vision, sorts of systems is come into being.These tongues
The body research and analysis system first step is often to be acquired using complicated equipment to human body tongue, this step is not only to equipment
It is required that high, high operation requirements, it is also necessary to be collected object to collection in worksite, need to spend many equipment costs and time cost.
Second step be detection tongue body, collected data are analyzed using traditional algorithm, analytic process to acquisition data and
The dependence of artificial interaction is very strong, and influence of the error to result is very big, it is also possible to use complicated calculating, efficiency is relatively low.
Nearest deep neural network is grown rapidly, and the model based on deep neural network has stepped into maturation, answered
All trades and professions are used, and in the application of tongue body detection or a piece of blue sea.
Present mobile device has been widely applied in the daily work and life of people, if automatically detecting movement
Tongue body in the collected tongue picture of equipment, so that it may reduce the cost of man-machine interactively significantly.Detecting tongue image simultaneously is tongue body
Primary work and key technology in research and analysis system, simply and accurately obtain tongue body can in later step to tongue body into
Row diagnosis is laid a good foundation.
Summary of the invention
The object of the present invention is to provide a kind of tongue body automatic testing method based on deep learning, to overcome in conventional method
Obtain the deficiency of tongue body.
Realizing the specific technical solution of the object of the invention is:
A kind of tongue body automatic testing method based on deep learning, this method comprising the following specific steps
Step 1: choosing the facial image comprising tongue, the artificial tongue body position in rectangle frame data mark image；Rectangle frame
For (x1,y1,x2,y2) format, wherein x1,y1Indicate the position in the upper left corner of tongue on the image, x2,y2Indicate tongue in image
On the lower right corner position, it is D that 70%-80% image taggeds are chosen from the image for being labeled with tongue body position1；It is remaining
The image tagged for being labeled with tongue body position is D2, all non-tongue images are labeled as D3；
Step 2: the D in step 11The rectangle frame data manually marked is instructed as training set using object detection network Yolo
Get hundreds of models；A smallest model of loss value is chosen from model as test model；
Step 3: by D in step 12As test set, tested using the test model of step 2, test model is image
Resize extracts feature by the CNN in model and is predicted, utilize non-maximum restraining (NMS) algorithm at the size of 255*255
Extra i.e. overlapping window is eliminated, optimum position is found；Model prediction go out tongue in tongue image rectangle frame coordinate and
The confidence level of tongue, a threshold value is arranged in confidence level, and in the image of prediction, confidence level is more than the rectangle frame data of threshold value just can quilt
Output；Tongue image finally is cut using rectangle frame data, obtains several images for detecting test set, is labeled as D4, D4In
It include the image of the i.e. non-tongue of tongue body image and noise image；
Step 4: using the tongue image D in step 11As positive sample, all non-tongue image D3As negative sample；To positive and negative
Sample carries out 90 degree, 180 degree, 270 degree of rotations respectively, obtains 4 times of original sample quantity of data enhancing, constitutes DenseNet
The training set of network；
Step 5: using training set obtained in step 5, being input to training in DenseNet network, extract sample characteristics, obtain
DenseNet model；
Step 6: by data set D obtained in step 34Test is extracted using the DenseNet model of step 6 as test set
Collect feature, obtain the confidence level of tongue in image, takes the highest image of confidence level to be used as and eventually detect tongue image.
The invention has the benefit that detecting tongue figure automatically by using the method for deep learning in tongue body detection process
Tongue body as in does not need that tongue image is analyzed and calculated using complicated algorithm, removes man-machine interactively in detection process from,
Improve the degree of automation.
Detailed description of the invention
Fig. 1 is flow chart of the present invention.
Specific embodiment
Below in conjunction in the embodiment of the present invention and attached drawing, technical solution of the present invention is clearly and completely described,
Obviously, described embodiments are only a part of the embodiments of the present invention, instead of all the embodiments.Based in the present invention
Embodiment, those of ordinary skill in the art's every other embodiment obtained, shall fall within the protection scope of the present invention.
Refering to fig. 1, a kind of tongue body detection method based on deep learning of the invention, comprising the following steps:
Step 1: choosing hundreds of facial images comprising tongue, marking the tongue body position in image is rectangle frame data, and will
Facial image and rectangle frame data are D according to 8: 2 or 7: 3 ratio cut partition data set1And D2, non-tongue image is labeled as D3。
Step 2: it detects to obtain the tongue image in image using object detection network Yolo, specifically includes:
2-1: D in step 11As training sample, training pattern is obtained using object detection network Yolo；
2-2: D in step 12As test sample, tongue image is obtained in conjunction with the obtained model inspection of step 2-1, is labeled as D4。
Step 3: for tongue image D obtained in step 2-24, tongue is filtered out using neural network model DenseNet
Head image, specific steps include:
3-1: tongue image D is chosen1With non-tongue image D3Data enhancing is carried out, obtained image is as the positive negative sample of training；
3-2: bis- disaggregated model of sample training DenseNet in step 2-1 is utilized；
3-3: the tongue image D in selecting step 2-24As test set, by bis- disaggregated model of DenseNet, detection obtains tongue
The head highest image of confidence level.
Embodiment
Given facial image (tongue image) of a batch comprising tongue and a batch do not include facial image (the non-tongue figure of tongue
Picture).As shown in Figure 1, specifically includes the following steps:
1) tongue position is labelled with rectangle frame to all tongue images, such as (x1,y1,x2,y2) format, wherein x1,y1Indicate tongue
The position in the upper left corner on the image, x2,y2The position for indicating the lower right corner of tongue on the image, chooses from tongue image
70%-80% images are labeled as D as training set1；Other tongue images are labeled as D2, all non-tongue images are labeled as
D3。
2) D in step 1)1With corresponding rectangle frame as training set, obtained using object detection network Yolo training
300 models therefrom choose a smallest model of loss value as test model.Wherein model is image Resize at 255*
255 size is extracted feature by CNN and is predicted, utilizes NMS(non-maximum restraining) by the way that threshold value is arranged come pre- to confidence level
Measure the coordinate of bounding box and the confidence level of tongue.
3) D in step 1)2As test set, is tested in conjunction with the test model that step 2 is chosen, finally detected
The i.e. non-tongue image of the n tongue image and noise image arrived is labeled as D4, in order to calculate the accuracy rate of tongue detection, simultaneously
The rectangle frame of the position of the tongue detected and non-tongue on picture can be also exported, format is equally (x1,y1,x2,y2).
4) it needs to find out tongue from the image that Yolo is detected, in order to improve network training effect, selecting in step 1)
For the tongue image taken as positive sample, all non-tongue images carry out data enhancing as negative sample, i.e., 90 degree, 180 degree,
270 degree of rotations obtain sample.
5) sample in step 4) carries out two classification based trainings using DenseNet network, obtains model, i.e., as training set
Classification judges whether contain tongue in picture.DenseNet is the disaggregated model based on residual error network, by the defeated of each layer network
Enter output all including all layer networks in front, improves the efficiency of transmission of information and gradient in neural network, every layer can
Gradient directly is taken from loss function, and directly obtains input signal, can thus train deeper network, while this net
There are also the effects of regularization for network structure.
6) the n image D that every tongue image detects in step 3)4As test set, trained in conjunction with DenseNet
Model obtains the confidence level that each image is tongue, for the n image that every tongue image detects, takes confidence level highest
Image is used as and detects tongue image.
7) for the accuracy rate of the tongue confirmly detected, tongue image phase in step 3) obtained in step 6) is found out
The rectangle frame answered, as candidate rectangle frame, with the rectangle frame manually marked, calculating hands over and than (IOU), judged according to artificial experience
One threshold value is set, hands over and the tongue image than being greater than threshold value is exactly last result.
8) it calculates and hands over and compare.It hands over and what is done than (IOU) function is to calculate the ratio between two bounding box intersections and union, this is general
Thought is in order to which whether evaluation object location algorithm is accurate, and formula is as follows.If the square of the tongue image obtained in step 7 by model
Shape frame data is A, and the rectangle frame data manually marked in step 1 is B, is calculated and is handed over and than (IOU) by formula.
It hands over and than function formula: IOU=(A ∩ B)/(A ∪ B)
9) for the recall rate of the tongue confirmly detected, judge that setting is handed over and than threshold value for 0.9 according to artificial experience, that is, hand over simultaneously
Tongue image than being greater than threshold value 0.9 is exactly last result.
The present invention completes a kind of method based on deep learning and detects and identify automatically the tongue body in tongue image, does not need
Tongue image is analyzed and calculated using complicated algorithm, removes man-machine interactively in detection process from, improves the degree of automation.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Within mind and principle, any modification, equivalent replacement, improvement and so on be should all be included in the protection scope of the present invention.
Claims (1)
1. a kind of tongue body automatic testing method based on deep learning, which is characterized in that this method comprising the following specific steps
Step 1: choosing the facial image comprising tongue, the artificial tongue body position in rectangle frame data mark image；Rectangle frame
For (x1,y1,x2,y2) format, wherein x1,y1Indicate the position in the upper left corner of tongue on the image, x2,y2Indicate tongue in image
On the lower right corner position, it is D that 70%-80% image taggeds are chosen from the image for being labeled with tongue body position1；It is remaining
The image tagged for being labeled with tongue body position is D2, all non-tongue images are labeled as D3；
Step 2: the D in step 11The rectangle frame data manually marked utilizes object detection network Yolo training as training set
Obtain hundreds of models；A smallest model of loss value is chosen from model as test model；
Step 3: by D in step 12As test set, tested using the test model of step 2, test model is image
Resize extracts feature by the CNN in model and is predicted, utilize non-maximum restraining (NMS) algorithm at the size of 255*255
Extra i.e. overlapping window is eliminated, optimum position is found；Model prediction go out tongue in tongue image rectangle frame coordinate and
The confidence level of tongue, a threshold value is arranged in confidence level, and in the image of prediction, confidence level is more than the rectangle frame data of threshold value just can quilt
Output；Tongue image finally is cut using rectangle frame data, obtains several images for detecting test set, is labeled as D4, D4In
It include the image of the i.e. non-tongue of tongue body image and noise image；
Step 4: using the tongue image D in step 11As positive sample, all non-tongue image D3As negative sample；To positive and negative
Sample carries out 90 degree, 180 degree, 270 degree of rotations respectively, obtains 4 times of original sample quantity of data enhancing, constitutes DenseNet
The training set of network；
Step 5: using training set obtained in step 4, being input to training in DenseNet network, extract sample characteristics, obtain
DenseNet model；
Step 6: by data set D obtained in step 34Test set is extracted using the DenseNet model of step 6 as test set
Feature obtains the confidence level of tongue in image, takes the highest image of confidence level to be used as and eventually detects tongue image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811596872.3A CN109740654A (en) | 2018-12-26 | 2018-12-26 | A kind of tongue body automatic testing method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811596872.3A CN109740654A (en) | 2018-12-26 | 2018-12-26 | A kind of tongue body automatic testing method based on deep learning |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109740654A true CN109740654A (en) | 2019-05-10 |
Family
ID=66359981
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811596872.3A Pending CN109740654A (en) | 2018-12-26 | 2018-12-26 | A kind of tongue body automatic testing method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109740654A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766665A (en) * | 2019-09-30 | 2020-02-07 | 天聚星信息科技(深圳)有限公司 | Tongue picture data analysis method based on strong supervision algorithm and deep learning network |
| CN111242038A (en) * | 2020-01-15 | 2020-06-05 | 北京工业大学 | Dynamic tongue tremor detection method based on frame prediction network |
| CN111881906A (en) * | 2020-06-18 | 2020-11-03 | 广州万维创新科技有限公司 | LOGO identification method based on attention mechanism image retrieval |
| CN112200091A (en) * | 2020-10-13 | 2021-01-08 | 深圳市悦动天下科技有限公司 | Tongue region identification method and device and computer storage medium |
| CN112949668A (en) * | 2019-12-10 | 2021-06-11 | 东北大学秦皇岛分校 | Garbage detection system based on deep learning |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107527009A (en) * | 2017-07-11 | 2017-12-29 | 浙江汉凡软件科技有限公司 | A kind of remnant object detection method based on YOLO target detections |
| CN108062764A (en) * | 2017-11-30 | 2018-05-22 | 极翼机器人(上海)有限公司 | A kind of object tracking methods of view-based access control model |
| CN108509859A (en) * | 2018-03-09 | 2018-09-07 | 南京邮电大学 | A kind of non-overlapping region pedestrian tracting method based on deep neural network |
| CN108875595A (en) * | 2018-05-29 | 2018-11-23 | 重庆大学 | A kind of Driving Scene object detection method merged based on deep learning and multilayer feature |
| CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
| CN109087313A (en) * | 2018-08-03 | 2018-12-25 | 厦门大学 | A kind of intelligent tongue body dividing method based on deep learning |
-
2018
- 2018-12-26 CN CN201811596872.3A patent/CN109740654A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107527009A (en) * | 2017-07-11 | 2017-12-29 | 浙江汉凡软件科技有限公司 | A kind of remnant object detection method based on YOLO target detections |
| CN108062764A (en) * | 2017-11-30 | 2018-05-22 | 极翼机器人(上海)有限公司 | A kind of object tracking methods of view-based access control model |
| CN108509859A (en) * | 2018-03-09 | 2018-09-07 | 南京邮电大学 | A kind of non-overlapping region pedestrian tracting method based on deep neural network |
| CN108875595A (en) * | 2018-05-29 | 2018-11-23 | 重庆大学 | A kind of Driving Scene object detection method merged based on deep learning and multilayer feature |
| CN109087313A (en) * | 2018-08-03 | 2018-12-25 | 厦门大学 | A kind of intelligent tongue body dividing method based on deep learning |
| CN109064461A (en) * | 2018-08-06 | 2018-12-21 | 长沙理工大学 | A kind of detection method of surface flaw of steel rail based on deep learning network |
Non-Patent Citations (1)
| Title |
|---|
| 王贺璋: "基于深度学习的交通对象检测与识别", 《中国优秀硕士学位论文全文数据库 工程科技II辑》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110766665A (en) * | 2019-09-30 | 2020-02-07 | 天聚星信息科技(深圳)有限公司 | Tongue picture data analysis method based on strong supervision algorithm and deep learning network |
| CN112949668A (en) * | 2019-12-10 | 2021-06-11 | 东北大学秦皇岛分校 | Garbage detection system based on deep learning |
| CN111242038A (en) * | 2020-01-15 | 2020-06-05 | 北京工业大学 | Dynamic tongue tremor detection method based on frame prediction network |
| CN111242038B (en) * | 2020-01-15 | 2024-06-07 | 北京工业大学 | Dynamic tongue fibrillation detection method based on frame prediction network |
| CN111881906A (en) * | 2020-06-18 | 2020-11-03 | 广州万维创新科技有限公司 | LOGO identification method based on attention mechanism image retrieval |
| CN112200091A (en) * | 2020-10-13 | 2021-01-08 | 深圳市悦动天下科技有限公司 | Tongue region identification method and device and computer storage medium |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109740654A (en) | A kind of tongue body automatic testing method based on deep learning | |
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN108596046A (en) | A kind of cell detection method of counting and system based on deep learning | |
| CN109977780A (en) | A kind of detection and recognition methods of the diatom based on deep learning algorithm | |
| CN105096307B (en) | The method of detection object in paired stereo-picture | |
| CN109598224A (en) | Recommend white blood cell detection method in the Sections of Bone Marrow of convolutional neural networks based on region | |
| CN109829467A (en) | Image labeling method, electronic device and non-transient computer-readable storage medium | |
| CN107909010A (en) | A kind of road barricade object detecting method and device | |
| CN103324937A (en) | Method and device for labeling targets | |
| CN108647625A (en) | A kind of expression recognition method and device | |
| CN105989347A (en) | Intelligent marking method and system of objective questions | |
| CN106056597B (en) | Object visible detection method and device | |
| CN109086708A (en) | A kind of parking space detection method and system based on deep learning | |
| CN109886928A (en) | A kind of target cell labeling method, device, storage medium and terminal device | |
| CN112149512A (en) | Helmet wearing identification method based on two-stage deep learning | |
| CN101984453B (en) | Human eye recognition system and method | |
| CN104063677A (en) | Equipment used for estimating human body posture and method thereof | |
| CN107016353B (en) | A kind of integrated method and system of variable resolution target detection and identification | |
| CN108564085A (en) | A kind of method of automatic reading pointer type instrument reading | |
| CN110738132B (en) | Target detection quality blind evaluation method with discriminant perception capability | |
| CN105718954B (en) | A kind of recognition methods of objective attribute target attribute and classification based on view tactile fusion | |
| CN109919135A (en) | Behavioral value method, apparatus based on deep learning | |
| CN101930533A (en) | Device and method for performing sky detection in image collecting device | |
| CN104463240B (en) | A kind of instrument localization method and device | |
| CN105354549B (en) | Based on the rapid pedestrian detection method estimated like physical property |
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 | ||
| WD01 | Invention patent application deemed withdrawn after publication | ||
| WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20190510 |