CN109583273A - A kind of analysis process system of magnanimity plantar pressure data - Google Patents
A kind of analysis process system of magnanimity plantar pressure data Download PDFInfo
- Publication number
- CN109583273A CN109583273A CN201710903880.7A CN201710903880A CN109583273A CN 109583273 A CN109583273 A CN 109583273A CN 201710903880 A CN201710903880 A CN 201710903880A CN 109583273 A CN109583273 A CN 109583273A
- Authority
- CN
- China
- Prior art keywords
- data
- image
- magnanimity
- training
- footprint
- 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
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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Human Computer Interaction (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of analysis process systems of magnanimity plantar pressure data, comprising: magnanimity footprint data acquisition module, comprising: the acquisition for the personal information that the acquisition of dynamic footprint data, the acquisition of static footprint data, user input；Data attribute unified modules, comprising: data type unified modules and data dimension unified modules；Label file makes module, carries out taxonomic revision to the magnanimity footprint data and user information got；Image pre-processing module, to barefoot or wearing sock print image data and pre-process；Data set make module, will complete it is pretreated barefoot or wear sock print image data be divided into training set and verifying collection；Deep learning network training and adjustment module, design are directed to the deep learning network of magnanimity footprint image, and circulation carries out deep learning network training and adjustment.The present invention realizes the processing of magnanimity footprint data, so that the processing mode specification of magnanimity footprint data is effective.
Description
Technical field
The present invention relates to a kind of Data Analysis Services system, the analysis of specifically a kind of magnanimity plantar pressure data is handled
System.
Background technique
The distribution of plantar pressure can reflect out the different degrees of posture of people, and plantar pressure can be applied in every field
In, for example, plantar nervous arch can help to distinguish by detection ulcer high risk zone diabetes patient and non-diabetic people, into
The pedopathy of row orthopaedics is examined, detection walking posture judges patient's recovery extent and provides for lower-limb ailments rehabilitation in rehabilitation training
Solution etc., while in criminal detection, judge all to be answered in the information such as people's height and weight and ground interactive game
With.But the analysis processing mode of current plantar pressure data is all undesirable.
Summary of the invention
This application provides a kind of analysis process systems of magnanimity plantar pressure data, for the magnanimity footprint got
Data are arranged, are analyzed, are utilized.
The first technical solution of the application is: a kind of analysis process system of magnanimity plantar pressure data, comprising:
Magnanimity footprint data acquisition module, comprising: the acquisition of dynamic footprint data, the acquisition of static footprint data, user
The acquisition of the personal information of input；
Data attribute unified modules, comprising: data type unified modules and data dimension unified modules；
Label file makes module, carries out taxonomic revision to the magnanimity footprint data and user information got；
Image pre-processing module, to barefoot or wearing sock print image data and pre-process；
Data set makes module, will complete pretreated barefoot or to wear sock print image data and be divided into training set and verifying
Collection；
Deep learning network training and adjustment module, design be directed to magnanimity footprint image deep learning network, recycle into
Row deep learning network training and adjustment.
Further, data type unified modules are divided into following two situation:
(1) real-time dynamic data needs static data to be converted into, in averaging process by the average treatment of certain time
In, the dynamic data that can not be obtained does feature extraction, directly forms one-dimensional data and does training or test；
It (2), be by Automatic Feature Extraction, with same place for the data of stride characteristics information in walking process can be obtained
The mode of track forms one-dimensional data, does training or test.
Further, data dimension unified modules are divided into following two situation:
(1) is constructed by virtual hard into trace object, is recorded three-dimension foot using the method for emulation collision for three dimensional point cloud
With the point set of object collision, record point set to the vertical range at trace object, to be highly that information forms height map；
(2) two dimension barefoot or wears sock print data, for the trace figure of background complexity can by following two mode into
Row processing,
A, generation or acquisition are a variety of containing barefoot or wearing the Background of sock print for training；
B, the mode for manually or automatically removing background proposes barefoot or wears sock print foreground picture；
(3) one-dimensional data is considered the amount extracted by initial characteristics, projects by dimensionality reduction, in conjunction with other data characteristicses
It uses.
Further, image pre-processing module, specifically:
1) automatic screening: by the training of batch data, to reject to barefoot or wearing sock print image and doing automatic screening
It is unsatisfactory for desired image；Image is inputted into trained CNN model and is made whether the differentiation met the requirements, CNN mould herein
Type by known quality great amount of images by whether meet requirement classify so that training obtain.Differentiate that result is to meet to want
That asks leaves, and the rejecting for being unsatisfactory for requiring is not involved in subsequent step；
2) footprint image resolution ratio is normalized；
3) footprint image after normalization is subjected to regional area segmentation.
Further, the normalization of footprint image resolution ratio there is into following two mode:
(1) manual type: needs mark barefoot or wear sock print same place position, same place position totally 4 points, i.e.,
Toes inner edge bump, toes outer rim bump, heel rear bump, longest toe leading edge bump；Every width figure is done according to the scale of 4 points
The normalizing in the direction x/y；
(2) automated manner: by the training of batch data, come to barefoot or wearing sock print image and do automatic same place
Label, every width figure do the normalizing in the direction x/y according to the scale of 4 points, trained feature include correlation between angle point, point,
Whole picture figure；Mode includes the full convolutional network based on deep learning, the registration of the same place based on Image geometry transform.
Further, regional area is divided specifically:
(1) all images are subjected to dimension normalization, normalized process includes extracting target footprint area, will extract area
Domain mends 0 into square, and the image scaling for meeting subsequent processing demand is carried out using interpolation method；
(2) footprint image after normalization is subjected to rectangular area according to toes area, vola pedis area, arch area and heel area
Segmentation, obtains local area image, specific segmentation such as the following figure:
(3) all obtained local area images of dividing are classified according to regional area classification, i.e., by all toes
Area's image is put together, and all vola pedis areas image is put together, and all arch area images are put together, and all heel area images are put
Together；Under each area classification again image is classified according to secondary user ID, that is, belong to same user toes area,
Vola pedis area, arch area, heel area image are individually put together.
As further, by pretreated data application in classification (such as gender, age, height, weight determine):
It can will pre-process obtained general image herein and each local area image is trained to obtain for different plantar regions respectively
Model；
(1) training data prepare: to completed pretreated training set and verifying collection two-dimensional image data, according to
Classification demand is respectively divided into N group；
(2) grouping carries out the training based on CNN network, is used herein as improved AlexNet network, network improvement is as follows:
A. initial network:
Network is constituted: 4 layers of convolutional layer, 2 layers of pond layer, 2 layers of full articulamentum.
Network connection:
conv1+pooling1+relu→conv2+pooling2+relu→conv3+relu→conv4+relu→fc5
→fc6
Wherein, conv represents convolutional layer, and pooling represents pond layer, and relu represents activation primitive, and fc represents full connection
Layer；
B. network is adjusted by trained and verification result:
Using initial network carry out it is primary complete after training, tested with verify data, it is assumed that sex determination's accuracy rate is not
Sufficient N%, then it is assumed that network structure needs to adjust；
C. the trained disaggregated model based on CNN network is obtained.
As further, the method for adjustment of network structure is as follows:
For any one width figure, every layer of convolutional layer result is exported, it is same with the result of every layer of convolution and other images
Layer convolution results do correlation ratio pair:
If i. the case where correlation obviously becomes larger occurs in certain level, the convolution kernel ruler of the level and its rear end is reduced
It is very little, or directly using the layer as full articulamentum re -training；
If ii. correlation is goed deep into level, gradually get higher, but correlation variation is little after certain level, then directly by the layer
As full articulamentum re -training, simplified model；
If the iii. equal very little of the correlation of every level, the convolution kernel size close with input level is improved, or increase
Convolution layer number, until there are other situations.
Circuit training and test are carried out, network structure and parameter are constantly adjusted, when gender determination rate of accuracy in test result
Greater than N%, then network adjustment terminates, and saves the network model that training obtains.
The beneficial effects of the present invention are: this patent is used to by providing a kind of analysis and processing method of magnanimity footprint data
Finishing analysis is carried out to the magnanimity footprint data of acquisition, carries out neural network for different problems using the method for deep learning
Adjustment and training.The present invention realizes the processing of magnanimity footprint data, so that the processing mode specification of magnanimity footprint data has
Effect, so that the data of processing are more suitable for the adjustment and training of deep learning neural network.
Detailed description of the invention
The present invention shares 3 width of attached drawing:
Fig. 1 is that scale normalizing refers to point diagram；
Fig. 2 is region segmentation schematic diagram；
Fig. 3 is this system work flow diagram.
Specific embodiment
To make the objectives, technical solutions, and advantages of the present invention clearer, right in the following with reference to the drawings and specific embodiments
The present invention is described in detail.
Embodiment 1
The present embodiment provides a kind of analysis process systems of magnanimity plantar pressure data, comprising:
1. magnanimity footprint data acquisition module:
For this module by acquisition user's footprint data, footprint data collected mainly include dynamic and static footprint
Data；Then the personal information of user's input, gender, age, height, weight including user etc. are obtained；Finally carry out magnanimity
Data acquisition.
2. data attribute unified modules；
1) data type is unified:
(1) real-time dynamic data needs static data to be converted into, in averaging process by the average treatment of certain time
In, the dynamic data that can not be obtained does feature extraction, directly forms one-dimensional data and does training or test；
It (2), be by Automatic Feature Extraction, with of the same name for the data of stride characteristics information in available walking process
The mode of the locus of points forms one-dimensional data, does training or test.
2) data dimension is unified:
(1) is constructed by virtual hard into trace object, is recorded three-dimension foot using the method for emulation collision for three dimensional point cloud
The point set of (having deformation, non-rigid) and object collision records point set to the vertical range at trace object, highly to be formed for information
Height map；
(2) two dimension barefoot or wears sock print data the trace figure of background complexity can be generated or acquire and is a variety of
Containing barefoot or the Background of sock print is worn for training, can also manually or automatically be removed background mode propose it is red
Foot wears sock print foreground picture；
(3) one-dimensional data is considered the amount extracted by initial characteristics, can project by dimensionality reduction, with other data characteristicses
It is used in combination.
3. label file makes module:
This module mainly carries out taxonomic revision to the magnanimity footprint data and user information got.It is first each completion
The user of footprint data acquisition distributes ID；Then by the footprint data of all acquisitions according to the ID of respective secondary user into
Row classification, i.e., all footprint data of the same user are individually put together (including left foot, right crus of diaphragm image and various differences
The data of angle difference weight)；Finally the userspersonal information of all acquisitions forms label file, file format according to User ID
Such as the following figure:
4. image pre-processing module:
1) automatic screening: by the training of batch data, to reject to barefoot or wearing sock print image and doing automatic screening
It is unsatisfactory for desired image；Image is inputted into trained CNN model and is made whether the differentiation met the requirements, CNN mould herein
Type by known quality great amount of images by whether meet requirement classify so that training obtain.Differentiate that result is to meet to want
That asks leaves, and the rejecting for being unsatisfactory for requiring is not involved in subsequent step；
2) resolution ratio normalizes:
Since acquisition mode that is each barefoot or wearing sock print is different, resolution ratio is also different, and embodiment is on the image
The footprint of the same person, on different figures not of uniform size, it is artificial/automatically process by way of to barefoot or wearing socks foot
Mark does resolution ratio normalizing (people barefoot or to wear sock print length and width ratio be not what equal proportion scaled, therefore even if picture size
Unanimously, barefoot or to wear sock print also different):
(1) manual type: needs mark barefoot or wear sock print same place position, and same place position and definition refer to
Fig. 1, totally 4 points, every width figure do the normalizing in the direction x/y according to the scale of 4 points；
(2) automated manner: by the training of batch data, come to barefoot or wearing sock print image and do automatic same place
Label, every width figure do the normalizing in the direction x/y according to the scale of 4 points, and trained feature includes but is not limited between angle point, point
Correlation in addition whole picture figure, mode include but is not limited to full convolutional network (FCN) based on deep learning, be based on image
The same place registration of geometric transformation (affine, perspective etc.).
3) regional area segmentation (is analyzed, the judgement including gender, age, weight, height, after processing for biological characteristic
Image be mainly used to classify):
(1) all images are subjected to dimension normalization, normalized process includes extraction target footprint area (can be by footprint
The minimum rectangular area completely framed), region will be extracted and mend 0 into square, carry out meeting subsequent processing demand using interpolation method
The image scaling Aspect Ratio of proleg mark (guarantee)；
(2) footprint image after normalization is subjected to rectangular area according to toes area, vola pedis area, arch area and heel area
Segmentation, obtains local area image, specific segmentation such as Fig. 2:
(3) all obtained local area images of dividing are classified according to regional area classification, i.e., by all toes
Area's image is put together, and all vola pedis areas image is put together, and all arch area images are put together, and all heel area images are put
Together；Image is classified again according to secondary user ID under each area classification, that is, belongs to the toes area of same user
(vola pedis area/arch area/heel area) image is individually put together.
5. data set makes:
1) it will complete pretreated barefoot or to wear sock print image data set and be defined as two parts:
(1) training set: for the training process of deep learning, each footprint data sampling barefoot is believed with subordinate gender
Breath, this gender information are exactly this barefoot or wear the label of sock print；
(2) verifying collection: for verifying the quality of deep learning result.It is each barefoot or to wear sock print data sampling and have
Subordinate gender information, but verify collection and be not involved in training, it is used only to measure the accuracy of sex determination.
2) wherein, requirement of each section to data:
(1) data dimension of verifying collection must not be higher than training set data dimension, and the data information amount of collection to be identified must not be high
In the information content of training set data；
(2) to guarantee that the information integrity of verify data, effective information data dimension must not be lower than the spies of identification data set
Levy dimension；
(3) for doing trained data, everyone data, more than at least 10 groups different types of (each 5 groups of left and right foot),
Trained individual amount is in ten thousand people grade.
6. deep learning network training and adjustment module:
This module makes data set using by pretreated magnanimity footprint image mainly according to practical application request
(including training dataset and validation data set)；Design is directed to the deep learning network of magnanimity footprint image；Circulation carries out depth
Learning network training and adjustment.
Embodiment 2
Feelings the present embodiment provides the data handled by embodiment 1 for the adjustment and training of deep learning neural network
Condition:
By pretreated data application in classification (such as gender, age, height, weight determine) herein: (will can locate in advance
It manages obtained general image and each local area image is trained to obtain the model for different plantar regions respectively)
(1) training data prepare: to completed pretreated training set and verifying collection two-dimensional image data, according to
Classification demand carries out artificially respectively being divided into N group；
(2) grouping carries out the training based on CNN network, is used herein as improved AlexNet network, network improvement is as follows:
A, initial network:
Network is constituted: 4 layers of convolutional layer, 2 layers of pond layer, 2 layers of full articulamentum.
Network connection:
conv1+pooling1+relu→conv2+pooling2+relu→conv3+relu→conv4+relu→fc5
→fc6
Wherein, conv represents convolutional layer, and pooling represents pond layer, and relu represents activation primitive, and fc represents full connection
Layer；
Each layer network parameter:
Conv1: convolution kernel size: 5*5；Convolution nuclear volume: 16；Convolution step-length: 2；Weights initialisation mode: xavier
Pooling1: core size: 2*2；Convolution step-length: 1
Conv2: convolution kernel size: 3*3；Convolution nuclear volume: 32；Convolution step-length: 1；Weights initialisation mode: xavier
Pooling2: core size: 2*2；Convolution step-length: 1
Conv3: convolution kernel size: 3*3；Convolution nuclear volume: 64；Convolution step-length: 1；Weights initialisation mode: xavier
Conv4: convolution kernel size: 3*3；Convolution nuclear volume: 128；Convolution step-length: 1；Weights initialisation mode: xavier
Fc5: neuron number: 1024；Weights initialisation mode: xavier
Fc6: output channel number: 2；Weights initialisation mode: xavier
B, network is adjusted by trained and verification result:
Using initial network carry out it is primary complete after training, tested with verify data, it is assumed that sex determination's accuracy rate is not
Foot 80%, then it is assumed that network structure needs to adjust, and method of adjustment is as follows:
For any one width figure, every layer of convolutional layer result is exported, it is same with the result of every layer of convolution and other images
Layer convolution results do correlation ratio pair:
If i. the case where correlation obviously becomes larger occurs in certain level, the convolution kernel ruler of the level and its rear end is reduced
It is very little, or directly using the layer as full articulamentum re -training；
If ii. correlation is goed deep into level, gradually get higher, but correlation variation is little after certain level, then directly by the layer
As full articulamentum re -training, simplified model；
If the iii. equal very little of the correlation of every level, the convolution kernel size close with input level is improved, or increase
Convolution layer number, until there are other situations.
Circuit training and test are carried out, network structure and parameter are constantly adjusted, when gender determination rate of accuracy in test result
Greater than 80% network adjustment terminates, and saves the network model that training obtains.
C, the trained disaggregated model based on CNN network is obtained.
The application provides a kind of analysis and processing method of magnanimity footprint data, for the magnanimity footprint data got into
Row is arranged, analysis, is utilized.It is wherein main comprising carrying out attribute unification and pretreatment to the magnanimity footprint data of acquisition, it will handle
Data afterwards carry out the adjustment and training of neural network using the method for deep learning for different problems, finally obtain and are based on
The disaggregated model of magnanimity footprint data.
The foregoing is only a preferred embodiment of the present invention, but scope of protection of the present invention is not limited thereto,
Anyone skilled in the art within the technical scope of the present disclosure, according to the technique and scheme of the present invention and its
Inventive concept is subject to equivalent substitution or change, should be covered by the protection scope of the present invention.
Claims (8)
1. a kind of analysis process system of magnanimity plantar pressure data characterized by comprising
Magnanimity footprint data acquisition module, comprising: the acquisition of dynamic footprint data, the acquisition of static footprint data, user's input
Personal information acquisition；
Data attribute unified modules, comprising: data type unified modules and data dimension unified modules；
Label file makes module, carries out taxonomic revision to the magnanimity footprint data and user information got；
Image pre-processing module, to barefoot or wearing sock print image data and pre-process；
Data set make module, will complete it is pretreated barefoot or wear sock print image data be divided into training set and verifying collection；
Deep learning network training and adjustment module, design are directed to the deep learning network of magnanimity footprint image, and circulation carries out deep
Spend learning network training and adjustment.
2. a kind of analysis process system of magnanimity plantar pressure data according to claim 1, which is characterized in that data type
Unified modules are divided into following two situation:
(1) real-time dynamic data needs to be converted into static data by the average treatment of certain time, in averaging process, nothing
The dynamic data that method obtains does feature extraction, directly forms one-dimensional data and does training or test；
It (2), be by Automatic Feature Extraction, with the locus of points of the same name for the data of stride characteristics information in walking process can be obtained
Mode form one-dimensional data, do training or test.
3. a kind of analysis process system of magnanimity plantar pressure data according to claim 1 or claim 2, which is characterized in that data
Dimension unified modules are divided into following two situation:
(1) is constructed by virtual hard into trace object, is recorded three-dimension foot and visitor using the method for emulation collision for three dimensional point cloud
The point set of body collision, record point set to the vertical range at trace object, to be highly that information forms height map；
(2) two dimension barefoot or wears sock print data, for the trace figure of background complexity can be by following two mode at
Reason,
A, generation or acquisition are a variety of containing barefoot or wearing the Background of sock print for training；
B, the mode for manually or automatically removing background proposes barefoot or wears sock print foreground picture；
(3) one-dimensional data is considered the amount extracted by initial characteristics, projects by dimensionality reduction, makes in conjunction with other data characteristicses
With.
4. a kind of analysis process system of magnanimity plantar pressure data according to claim 3, which is characterized in that image is located in advance
Module is managed, specifically:
1) automatic screening: by the training of batch data, image is inputted in trained CNN model to discriminate whether to meet and is wanted
It asks, is left what is met the requirements, the rejecting for being unsatisfactory for requiring is not involved in subsequent step；
2) footprint image resolution ratio is normalized；
3) footprint image after normalization is subjected to regional area segmentation.
5. a kind of analysis process system of magnanimity plantar pressure data according to claim 4, which is characterized in that by footprint figure
As resolution ratio normalization has following two mode:
(1) manual type: needs mark barefoot or wear sock print same place position, same place position totally 4 points, i.e. toes
Inner edge bump, toes outer rim bump, heel rear bump, longest toe leading edge bump；Every width figure is x/y according to the scale of 4 points
The normalizing in direction；
(2) automated manner: by the training of batch data, come to barefoot or wearing the mark that sock print image does automatic same place
Note, every width figure do the normalizing in the direction x/y according to the scale of 4 points, and trained feature includes correlation between angle point, point, whole
Width figure；Mode includes the full convolutional network based on deep learning, the registration of the same place based on Image geometry transform.
6. a kind of analysis process system of magnanimity plantar pressure data according to claim 4, which is characterized in that regional area
Segmentation specifically:
(1) all images are subjected to dimension normalization, normalized process includes extracting target footprint area, will extract region and mends
0 carries out the image scaling for meeting subsequent processing demand at square, using interpolation method；
(2) footprint image after normalization is subjected to rectangular area point according to toes area, vola pedis area, arch area and heel area
It cuts, obtains local area image；
(3) all obtained local area images of dividing are classified according to regional area classification, i.e., by all toes areas figure
As putting together, all vola pedis areas image is put together, and all arch area images are put together, and all heel area images are placed on one
It rises；Image is classified again according to secondary user ID under each area classification, that is, belongs to toes area, the vola pedis of same user
Area, arch area, heel area image are individually put together.
7. according to a kind of analysis process system of magnanimity plantar pressure data of claim 5 or 6, which is characterized in that pass through
Pretreated data application is in classification: can will pre-process obtained general image herein and each local area image is instructed respectively
Get the model for different plantar regions；
(1) training data prepares: to the two-dimensional image data for having completed pretreated training set and verifying collection, according to classification
Demand is respectively divided into N group；
(2) grouping carries out the training based on CNN network, is used herein as improved AlexNet network, network improvement is as follows:
A. initial network:
Network is constituted: 4 layers of convolutional layer, 2 layers of pond layer, 2 layers of full articulamentum.
Network connection:
conv1+pooling1+relu→conv2+pooling2+relu→conv3+relu→conv4+relu→fc5→
fc6
Wherein, conv represents convolutional layer, and pooling represents pond layer, and relu represents activation primitive, and fc represents full articulamentum；
B. network is adjusted by trained and verification result:
Using initial network carry out it is primary complete after training, tested with verify data, it is assumed that sex determination's accuracy rate is insufficient
N%, then it is assumed that network structure needs to adjust；
C. the trained disaggregated model based on CNN network is obtained.
8. a kind of analysis process system of magnanimity plantar pressure data according to claim 7, which is characterized in that network structure
Method of adjustment it is as follows:
For any one width figure, every layer of convolutional layer result is exported, is rolled up with the result of every layer of convolution and the same layer of other images
Product result does correlation ratio pair:
If i. the case where correlation obviously becomes larger occurs in certain level, the convolution kernel size of the level and its rear end is reduced, or
Person is directly using the layer as full articulamentum re -training；
If ii. correlation is goed deep into level, gradually get higher, but correlation variation is little after certain level, then directly by this layer of conduct
Full articulamentum re -training, simplified model；
If the iii. equal very little of the correlation of every level, the convolution kernel size close with input level is improved, or increase convolution
Layer number, until there are other situations.
Circuit training and test are carried out, network structure and parameter are constantly adjusted, when gender determination rate of accuracy is greater than in test result
Then network adjustment terminates N%, saves the network model that training obtains.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710903880.7A CN109583273A (en) | 2017-09-29 | 2017-09-29 | A kind of analysis process system of magnanimity plantar pressure data |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710903880.7A CN109583273A (en) | 2017-09-29 | 2017-09-29 | A kind of analysis process system of magnanimity plantar pressure data |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109583273A true CN109583273A (en) | 2019-04-05 |
Family
ID=65914629
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710903880.7A Pending CN109583273A (en) | 2017-09-29 | 2017-09-29 | A kind of analysis process system of magnanimity plantar pressure data |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109583273A (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110751200A (en) * | 2019-10-15 | 2020-02-04 | 辽宁师范大学 | Shoe print height estimation method based on multivariate gauss |
| CN111580411A (en) * | 2020-04-27 | 2020-08-25 | 珠海格力电器股份有限公司 | Control parameter optimization method, device and system |
| CN112766142A (en) * | 2021-01-15 | 2021-05-07 | 天津大学 | Plantar pressure image processing method, plantar pressure image identification method and gait analysis system |
| CN113724151A (en) * | 2021-07-30 | 2021-11-30 | 荣耀终端有限公司 | Image enhancement method, electronic equipment and computer readable storage medium |
| CN113838117A (en) * | 2021-08-06 | 2021-12-24 | 公安部物证鉴定中心 | Height estimation method based on plantar pressure |
| US11574421B2 (en) | 2019-08-28 | 2023-02-07 | Visualize K.K. | Methods and systems for predicting pressure maps of 3D objects from 2D photos using deep learning |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SG177884A1 (en) * | 2010-08-03 | 2012-02-28 | Archright M Sdn Bhd | A method for producing an orthotic insole with fine-tuned contour |
| CN104574426A (en) * | 2015-02-03 | 2015-04-29 | 大连恒锐科技股份有限公司 | Method and device for human body feature analysis and based on barefoot or stocking-wearing footprint images |
| CN105534526A (en) * | 2015-12-16 | 2016-05-04 | 哈尔滨工业大学深圳研究生院 | Method for measuring plantar pressure |
-
2017
- 2017-09-29 CN CN201710903880.7A patent/CN109583273A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SG177884A1 (en) * | 2010-08-03 | 2012-02-28 | Archright M Sdn Bhd | A method for producing an orthotic insole with fine-tuned contour |
| CN104574426A (en) * | 2015-02-03 | 2015-04-29 | 大连恒锐科技股份有限公司 | Method and device for human body feature analysis and based on barefoot or stocking-wearing footprint images |
| CN105534526A (en) * | 2015-12-16 | 2016-05-04 | 哈尔滨工业大学深圳研究生院 | Method for measuring plantar pressure |
Non-Patent Citations (2)
| Title |
|---|
| 南通大学教务处: "《学海图南南通大学优秀毕业设计（论文）集2015届》", 31 March 2016, 苏州大学出版社 * |
| 王欣: "基于卷积神经网络与足底压力信息的步态识别", 《中国优秀硕士学位论文全文数据库》 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US11574421B2 (en) | 2019-08-28 | 2023-02-07 | Visualize K.K. | Methods and systems for predicting pressure maps of 3D objects from 2D photos using deep learning |
| CN110751200A (en) * | 2019-10-15 | 2020-02-04 | 辽宁师范大学 | Shoe print height estimation method based on multivariate gauss |
| CN110751200B (en) * | 2019-10-15 | 2023-09-29 | 辽宁师范大学 | Shoe stamp height estimation method based on multiple gauss |
| CN111580411A (en) * | 2020-04-27 | 2020-08-25 | 珠海格力电器股份有限公司 | Control parameter optimization method, device and system |
| CN112766142A (en) * | 2021-01-15 | 2021-05-07 | 天津大学 | Plantar pressure image processing method, plantar pressure image identification method and gait analysis system |
| CN113724151A (en) * | 2021-07-30 | 2021-11-30 | 荣耀终端有限公司 | Image enhancement method, electronic equipment and computer readable storage medium |
| CN113838117A (en) * | 2021-08-06 | 2021-12-24 | 公安部物证鉴定中心 | Height estimation method based on plantar pressure |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109583273A (en) | A kind of analysis process system of magnanimity plantar pressure data | |
| CN104850825B (en) | A kind of facial image face value calculating method based on convolutional neural networks | |
| Dehshibi et al. | A new algorithm for age recognition from facial images | |
| CN109584251A (en) | A kind of tongue body image partition method based on single goal region segmentation | |
| CN109034045A (en) | A kind of leucocyte automatic identifying method based on convolutional neural networks | |
| CN104166842B (en) | It is a kind of based on block statistics feature and combine represent three-dimensional palm print recognition methods | |
| CN105678235B (en) | Three-dimensional face expression recognition methods based on representative region various dimensions feature | |
| CN105005765A (en) | Facial expression identification method based on Gabor wavelet and gray-level co-occurrence matrix | |
| CN112396573A (en) | Facial skin analysis method and system based on image recognition | |
| CN106991417A (en) | A kind of visual projection's interactive system and exchange method based on pattern-recognition | |
| CN106529504B (en) | A kind of bimodal video feeling recognition methods of compound space-time characteristic | |
| KR101700818B1 (en) | Method and apparatus for estimating age or gender using face image | |
| CN109583452B (en) | Human identity identification method and system based on barefoot footprints | |
| Hassan et al. | Deep learning analysis and age prediction from shoeprints | |
| Sajid et al. | The role of facial asymmetry in recognizing age-separated face images | |
| CN104573722A (en) | Three-dimensional face race classifying device and method based on three-dimensional point cloud | |
| CN104376312B (en) | Face identification method based on bag of words compressed sensing feature extraction | |
| Hitimana et al. | Automatic estimation of live coffee leaf infection based on image processing techniques | |
| CN104036291A (en) | Race classification based multi-feature gender judgment method | |
| Lanz et al. | Automated classification of therapeutic face exercises using the Kinect | |
| CN109523484B (en) | Fractal feature-based finger vein network repair method | |
| CN104573628A (en) | Three-dimensional face recognition method | |
| CN106886754A (en) | Object identification method and system under a kind of three-dimensional scenic based on tri patch | |
| CN106846399A (en) | A kind of method and device of the vision center of gravity for obtaining image | |
| Gornale | Fingerprint based gender classification for biometric security: A state-of-the-art technique |
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: 20190405 |