CN109508732A - A kind of heating ablation model data processing method based on support vector regression - Google Patents
A kind of heating ablation model data processing method based on support vector regression Download PDFInfo
- Publication number
- CN109508732A CN109508732A CN201811197853.3A CN201811197853A CN109508732A CN 109508732 A CN109508732 A CN 109508732A CN 201811197853 A CN201811197853 A CN 201811197853A CN 109508732 A CN109508732 A CN 109508732A
- Authority
- CN
- China
- Prior art keywords
- data
- sample
- image
- identified
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/03—Recognition of patterns in medical or anatomical images
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Processing (AREA)
Abstract
A kind of heating ablation model data processing method based on support vector regression, is related to support vector machines field.It includes that data set training obtains module, image data difference processing module, sample selection module, disaggregated model production module and data identification module；Wherein, data set training obtains module for obtaining data sample training set, includes the data sample set of multiple classifications in data sample training set, includes multiple data samples in the data sample set of each classification；Image data difference processing module is used to carry out interpolation calculation processing to the image data in data training set；Sample chooses module for choosing similar sample and foreign peoples's sample；Disaggregated model production module is used to obtain similitude judgment models using support vector machines training to set for difference sample, and obtains disaggregated model according to similitude judgment models；Data identification module includes data sample acquiring unit to be identified, for obtaining data sample to be identified.
Description
Technical field
The present invention relates to support vector machines field more particularly to a kind of heating ablation model datas based on support vector regression
Processing method.
Background technique
The method of radiofrequency ablation therapy tumor of spine has obtained clinical approval, but the multiplicity based on tumor of spine at present
Property and scrambling, treatment parameter (voltage and time) setting be current difficult point and emphasis.In operation decision process, with
By means of the experience of clinical many years, voltage and the time of rough judgement treatment are carried out according to the size and shape of tumor of spine.The method
Although application have the effect of certain, with very big hidden danger and unstability, can not only cause ablation of excess or not complete
It totally disappeared the problems such as melting, or even can excessive damage's normal tissue.Therefore in order to reduce human error and unsafe factor, largely
Isolated experiment or the data of numerical simulation more accurately can provide reference data for clinical practice, to guarantee that radio frequency, which is arranged, to disappear
Melt the science of voltage and time, it is ensured that the correctness of pre-operative surgical decision.
Since the isolated experiment of RF ablation tumor of spine has certain difficulty, thus this research is using numerical simulation
Mode simulates its process, and calculate its ablation areas as a result, analysis discuss temperature field feature, provide reference for clinical practice
Foundation.
Support vector machines is a kind of new machine learning method to grow up on the basis of Statistical Learning Theory, it is
It establishes in the VC dimension theory and structural risk minimization of Statistical Learning Theory, avoids local minimum point, and can have
Effect ground solves the problems, such as overfitting, has good popularization performance and preferable classification accuracy.Support vector machines is solving sample
Originally, many distinctive advantages shown in non-linear and high dimensional pattern identification problem make it a kind of outstanding engineering
Practise algorithm.Support vector machines has been widely used for each neck such as pattern-recognition, regression estimates, Multilayer networks at present
Domain.Moreover, the appearance of support vector machines has pushed kernel-based learning algorithms method (Kernel-based
LearningMethods rapid development), this method enable researcher efficiently to analyze non-linear relation, and this
Originally only linear algorithm can just obtain high efficiency.
The numerical simulation process of RF ablation is divided into two classes: monopolar electrode and mostly son according to the different emulation of electrode structure
The numerical simulation of pin electrode.
Numerical simulation process mainly has studied the temperature field characteristics under different voltages and time.Simultaneously according to point in temperature field
Cloth, selects the ablation areas of tumor of spine, and calculates transverse diameter, vertical diameter, depth, volume and the least radius of ablation areas.
The numerical simulation of 1.1 single electrodes
This research carries out the numerical simulation of RF ablation using monopolar electrode.Studies have shown that radiofrequency ablation therapy backbone is swollen
The voltage range of tumor is between 5-30V, and ablation time need to be higher than 1min, usually 2-20min.But due to the process of RF ablation
In, if the maximum temperature at center is higher than 120 DEG C, excessive tissue can be caused to be carbonized and damage electrode；And tumor of spine complete inactivation
Minimum temperature is 54 DEG C, therefore this research can select the treatment being more suitable for join according to the limitation of maximum temperature and minimum temperature
Amount.It can be visually seen from figure, when ablation time is 20min, voltage 5V, maximum temperature is 46.1 DEG C；Voltage is 7V,
Its maximum temperature is 54.9 DEG C.Therefore the numerical value of voltage should be higher than the minimum temperature that 7V just can reach tumour inactivation.Therefore originally
The voltage range of research is that should be higher than that 8V, and the range of ablation time is 2-20min.
According to Fig.3, the ablation areas under monopolar electrode is rendered as class elliposoidal or spherical, thus preferably to retouch
The feature of ablation areas is stated, this research is measured and calculated to its transverse diameter, vertical diameter, depth and volume, design parameter such as Fig. 3
It is shown.
Thermo parameters method under 1.2 different voltages
This research has carried out the research discussion of thermo parameters method in the case where being directed to voltage as 9V, 11V, 13V and 15V.When
When ablation time is 5min, the thermo parameters method of ablation areas (temperature be higher than 54 DEG C of region) as shown in figure 4, this figure be along
The sectional view in vertical diameter direction.With the increase of voltage, maximum temperature is constantly increased, and maximum temperature is followed successively by 61.9 DEG C, 74.2
DEG C, 88.9 DEG C, 106.2 DEG C, simultaneously because 54 DEG C of thermoisopleth constantly expands outward, ablation areas also constantly increases.From figure
In can find, the increase of voltage, transverse diameter is higher than the increase rate of vertical diameter, so that ablation areas expands as class ball by class elliposoidal
Shape is more conducive to clinical treatment.
Detailed data is as shown in the table:
| Voltage (V) | Transverse diameter (mm) | Vertical diameter (mm) | Depth (mm) | Volume (mm) |
| 9 | 5.00 | 7.12 | 5.80 | 119.65 |
| 11 | 9.51 | 10.25 | 9.73 | 360.01 |
| 13 | 11.55 | 12.76 | 11.69 | 683.82 |
| 15 | 13.41 | 13.37 | 13.70 | 1175.13 |
In conclusion problem of the existing technology is: support vector machines needs to handle mass data, however to picture number
Not accurate enough according to handling, precision is not high, ineffective；It is not high simultaneously for the recognition performance of set of metadata of similar data.
Summary of the invention
Therefore, to solve the above-mentioned problems, the present invention provides a kind of heating ablation model data based on support vector regression
Processing method carries out data identification using disaggregated model, can greatly improve support vector machines recognition performance；Image data simultaneously
Difference processing it is obvious improve interpolation result image accuracy have and widely answer suitable for the integer zooming ratio of image
Use prospect.
To achieve the above object, the present invention adopts the following technical scheme that.
A kind of heating ablation model data processing method based on support vector regression, which is characterized in that assemble for training including data
Practice obtain module (1), image data difference processing module (2), sample choose module (3), disaggregated model production module (4) and
Data identification module (5)；
Wherein, data set training obtains module (1) for obtaining data sample training set, in data sample training set
Include the data sample set of multiple classifications, includes multiple data samples in the data sample set of each classification；The figure
Interpolation calculation processing is carried out to the image data in data training set as data difference processing module (2) are used for；The sample choosing
Modulus block (3) is for choosing similar sample and foreign peoples's sample；Disaggregated model production module (4) is used for for difference sample to collection
It closes and similitude judgment models is obtained using support vector machines training, and obtain disaggregated model according to similitude judgment models；It is described
Data identification module (5) includes data sample acquiring unit to be identified, for obtaining data sample to be identified.
Preferably, image data difference processing module carries out n times wavelet transformation to image first and retains wavelet transformation
The subgraph of low frequency part afterwards, the width and height of subgraph are original imageSubgraph area is original image areaIt is matched in subgraph, using normalized crosscorrelation measure, the traversal search on subgraph, in search process
Retain the higher match point of several similarities, after search, the higher match point of similarity is mapped to original figure to be matched
In the search subregion of picture, the formula of normalized crosscorrelation measure is as follows:
Wherein, S (x, y) indicates gray value of the coordinate position for pixel at (x, y), T (u, v) in original image to be matched
Indicate that coordinate position is the gray value of pixel at (u, v) in template image, template image size is m × n, and m, n are positive integer.
Preferably, sample chooses module for the step of choosing similar sample and foreign peoples's sample: data sample is instructed
Practice each of collection data sample, randomly selects k and belong to the other data sample of same class as same with the data sample
Class sample randomly selects a data sample to belong to a different category with the data sample of k as foreign peoples's sample, and sample chooses module
The false-alarm probability of the malicious attack mode computation overall situation according to malicious node；
The first step is the secondary user's Cri, i=that each participates in cooperative sensing according to the signal-to-noise ratio γ i of each node
1 ... k designs a weightThen linear weighted function is carried out to the signal energy statistic Ui that collection obtains
Obtain the statistic of final signal energy
Second step, analysis false-alarm malicious attack mode influenced caused by frequency spectrum perception, obtain global false-alarm probability Pf and
The function expression attacked between Probability p a, attack strength Δ is as follows:
Preferably, data identification module includes data sample acquiring unit to be identified, difference sample to be identified to collection symphysis
At unit, similarity probabilities computing unit and classification determination unit；
Wherein, data sample acquiring unit to be identified is for obtaining data sample to be identified, and difference sample to be identified is to set
Generation unit generates 2k for randomly selecting k number from the sample set of each classification of data training set respectively according to sample
Difference sample pair to be identified obtains difference sample to be identified to set, and similarity probabilities computing unit using disaggregated model for being treated
Identification difference sample analyzes set, obtains the similarity probabilities of every one kind in data sample to be identified and data training set,
Classification determination unit is used to determine the classification of data sample ownership to be identified according to similarity probabilities.
Preferably, specific step is as follows for the anti-RSD attack blind Detecting digital fingerprinting method of data identification module:
Step 1: the generation of finger print data frame encrypts finger print information using grouping displacement scrambling algorithm, using frame
Coding techniques handles finger print information, obtains finger print data frame；
Step 2: it is embedded in finger print data frame in DCT domain, fragment is carried out to carrier image, it is embedding respectively in each complete fragment
Enter finger print data frame, forms multiple redundancy versions of finger print data frame；Carrier image is divided into the fragment that size is S × S first,
Wherein S=2k, then select m 8 × 8 block of pixels as embedded block in each fragment, be finally respectively embedded into embedded block
The data of n bit in finger print data frame should meet L=m × n if the binary length of finger print data frame is L；
Step 3: constructing the Differential Characteristics dot grid with constant spacing, fixed difference value in airspace, be embedded with number
In the carrier image of word fingerprint, with certain pixel (i0, j0) it is starting point, by line-spacing and column away from embedding in the picture in the way of being D
Enter Differential Characteristics point, to form a rectangular mesh in whole image；
The acquisition of step 4:RSD attack parameter；The doubtful mesh point on airspace is obtained first, then passes through setting seed
The mode combination " parallelogram law " of point, candidate point determines mesh approximation parallelogram, finally by side continuation approximation
Grid parallelogram carries out continuation to greatest extent, provides accurate parallelogram for attack type judgement and parameter calculating；
Step 5: the image rectification that digital finger-print extracts carries out school to mask image according to obtained RSD attack parameter
Just, it is specifically divided into: if α ≠ 0, by mask image rotation alpha angle counterclockwise, ifSo by mask image water
Flat distortionAngle；
Step 6: image synchronization positioning and fingerprint extraction and recovery, embedded block are B=(Iij)8×8, F=(Fuv)8×8For B
DCT coefficient, wherein IijA pixel value, F are arranged for the i-th row j of DCT embedded block BuvFor the u row v column in DCT coefficient matrix
Pixel value, is located at the finger print data of embedded block insertion n-bit, and fingerprint bit to be embedded is denoted as wi respectively.
Preferably, before image synchronization positioning and fingerprint extraction and recovery, to the image rectification of digital fingerprint extraction
Image information afterwards carries out image procossing, is two-dimensional function f (x, y) by the image definition after correction, and wherein x, y are that space is sat
Mark carries out image denoising processing, image of the image after denoising to image f (x, y) using image denoising unit first
Two-dimensional function is p (x, y), wherein
It reuses image smoothing unit to be smoothed above-mentioned image p (x, y), the image after picture smooth treatment
Two-dimensional function is h (x, y), and wherein smooth function is g (x, y),
(x, y) , ﹡ are convolution symbol to h (x, y)=p (x, y) * g, and σ is customized adjustable constant, and smooth effect is to pass through σ
Come what is controlled,
Image enhancement processing finally is carried out to above-mentioned image h (x, y) using image enhancing unit, by image enhancement processing
Two-dimensional image function is u (x, y) afterwards, wherein
U (x, y)=h (x, y)-h (x-1, y)+h (x+1, y)+h (x, y-1)-h (x, y+1)+h (x+1, y+1).
Preferably, the heating ablation model data processing method based on above-mentioned support vector regression, method include following step
It is rapid:
Step 1: data training set is obtained: for obtaining data sample training set；It include more in data sample training set
The data sample set of a classification includes multiple data samples in the data sample set of each classification；
Step 2: interpolation calculation processing is carried out to the image data in data training set: to the image in data training set
Data carry out interpolation calculation processing；
Step 3: sample is chosen: for executing the step of choosing similar sample and foreign peoples's sample；For data sample training
Each of collection data sample randomly selects k and belongs to the other data sample of same class as similar with the data sample
Sample randomly selects a data sample to belong to a different category with the data sample of k as foreign peoples's sample；
Step 4: disaggregated model produces module: for, to set, obtaining phase using support vector machines training for difference sample
Like property judgment models, disaggregated model is obtained according to similitude judgment models；
Kernel function is used to obtain similitude judgment models for the training of the support vector machines of Gaussian radial basis function；It can basis
One similitude judgment models obtains disaggregated model, can also obtain disaggregated model according to multiple similitude judgment models；Work as root
When obtaining disaggregated model according to a similitude judgment models, disaggregated model can be identical as similitude judgment models；When according to more
When a similitude judgment models obtain disaggregated model, disaggregated model can be the set of multiple similitude judgment models；
If x, z ∈ X, X belong to the space R (n), nonlinear function Φ realizes the mapping of input space X to feature space F,
Middle F belongs to R (m), n < < m；Had according to kernel function technology:
K (x, z)=<Φ (x), Φ (z)>；
Wherein:<,>it is inner product, K (x, z) is kernel function；From formula as can be seen that kernel function ties up m in higher dimensional space
The kernel function that product operation is converted into the n dimension low-dimensional input space calculates, to solve the dimension calculated in high-dimensional feature space
Disaster；
Gaussian radial basis function in this step refers to the gaussian kernel function in radial basis function, and radial basis function is edge
The scalar function of radial symmetric；It is normally defined the monotonic function of Euclidean distance between any point x to a certain center xc in space,
It can be denoted as k (| | x-xc | |)；
Step 5: data identification is carried out using disaggregated model, data sample acquiring unit to be identified is to be identified for obtaining
Data sample；Difference sample to be identified is to set generation unit, for respectively from the sample set of each classification of data training set
K number is randomly selected according to sample, 2k difference samples pair to be identified is generated, obtains difference sample to be identified to set；Similarity probabilities
Computing unit analyzes set for treating identification difference sample using disaggregated model, obtains data sample and number to be identified
According to the similarity probabilities of one kind every in training set；Classification determination unit, for determining data sample to be identified according to similarity probabilities
The classification of this ownership.
Compared with prior art, the present invention have it is following the utility model has the advantages that
Heating ablation model data processing method provided by the invention based on support vector regression, is carried out using disaggregated model
Data identification, can greatly improve support vector machines recognition performance, at the same the processing of image data difference it is obvious improve interpolation
The accuracy of result images has a wide range of applications suitable for the integer zooming ratio of image.
Detailed description of the invention
Fig. 1 is a kind of schematic diagram of heating ablation model data processing method based on support vector regression of the invention.
Appended drawing reference:
The training of 1- data set obtains module；2- image data difference processing module；3- sample chooses module；4- disaggregated model
Produce module；5- data identification module.
Liver neoplasm thermo parameters method when Fig. 2 applies voltage.
The measurement parameter of Fig. 3 ablation areas.
Thermo parameters method under Fig. 4 different voltages.
Specific embodiment
With reference to the accompanying drawings and examples at the heating ablation model data to provided by the invention based on support vector regression
Reason method is described in detail.
Embodiment 1
As shown in Figure 1, a kind of heating ablation model data processing method packet based on support vector regression provided by the invention
It includes data set training and obtains module, image data difference processing module, sample selection module, disaggregated model production module and number
According to identification module；
Data training set obtains module 1: for obtaining data sample training set；It include multiple in data sample training set
The data sample set of classification includes multiple data samples in the data sample set of each classification, such as selected from monopolar electrode into
The numerical simulation of row RF ablation, including voltage, tumor temperature field distribution when applying different voltages, ablation areas transverse diameter, vertical diameter,
Depth and volume, etc..
Image data difference processing module 2: interpolation calculation processing is carried out to the image data in data training set, improves figure
As the accuracy of data and the precision of picture, to obtain more preferably image property.
Sample chooses module 3: for executing the step of choosing similar sample and foreign peoples's sample: for data sample training set
Each of data sample, randomly select k with the data sample and belong to the other data sample of same class as similar sample
This, randomly selects a data sample to belong to a different category with the data sample of k as foreign peoples's sample.
Disaggregated model produces module 4: for difference sample to set, to be obtained similitude using support vector machines training and sentenced
Disconnected model, obtains disaggregated model according to similitude judgment models.
Data identification module 5 includes: data sample acquiring unit to be identified, for obtaining data sample to be identified；Wait know
Other difference sample is to set generation unit, for randomly selecting k number from the sample set of each classification of data training set respectively
According to sample, 2k difference samples pair to be identified are generated, obtain difference sample to be identified to set；Similarity probabilities computing unit, is used for
The difference sample to be identified analyzes set using the disaggregated model, obtains data sample to be identified and data training
Concentrate the similarity probabilities of every one kind；Classification determination unit, for determining data sample ownership to be identified according to similarity probabilities
Classification.
Image data difference processing module carries out n times wavelet transformation to image and retains the son of low frequency part after wavelet transformation
Image, the width and height of subgraph are original imageSubgraph area is original image areaIn subgraph into
Row matching, using normalized crosscorrelation measure, the traversal search on subgraph, retain in search process several similarities compared with
After search, the higher match point of similarity is mapped in the search subregion of original image to be matched for high match point,
The formula of normalized crosscorrelation measure is as follows:
S (x, y) indicates gray value of the coordinate position for pixel at (x, y), T (u, v) expression in original image to be matched
Coordinate position is the gray value of pixel at (u, v) in template image, and template image size is m × n；
Sample chooses 3 data fusion center of module and carries out data fusion to the perception information being collected into, and saves according to malice
The false-alarm probability of the malicious attack mode computation overall situation of point；
The first step is the secondary user's Cri, i=that each participates in cooperative sensing according to the signal-to-noise ratio γ i of each node
1 ... k designs a weightThen linear weighted function is carried out to the signal energy statistic Ui that collection obtains
Obtain the statistic of final signal energy
Second step, analysis false-alarm malicious attack mode influenced caused by frequency spectrum perception, obtain global false-alarm probability Pf and
The function expression attacked between Probability p a, attack strength Δ is as follows:
Specific step is as follows for the anti-RSD attack blind Detecting digital fingerprinting method of the data identification module 5:
The generation of finger print data frame；Finger print information is encrypted using grouping displacement scrambling algorithm；Skill is encoded using frame
Art handles finger print information, obtains finger print data frame；
It is embedded in finger print data frame in DCT domain, fragment is carried out to carrier image, is respectively embedded into fingerprint in each complete fragment
Data frame forms multiple redundancy versions of finger print data frame；Carrier image is divided into the fragment that size is S × S first, wherein S
=2k, then select m 8 × 8 block of pixels as embedded block in each fragment, fingerprint number be finally respectively embedded into embedded block
L=m × n should be met if the binary length of finger print data frame is L according to the data of n bit in frame；
The Differential Characteristics dot grid with constant spacing, fixed difference value is constructed in airspace, is being embedded with digital finger-print
Carrier image in, with certain pixel (i0, j0) it is starting point, by line-spacing and column away from being embedded in difference in the picture in the way of being D
Characteristic point, to form a rectangular mesh in whole image；
The acquisition of RSD attack parameter；The doubtful mesh point on airspace is obtained first, then passes through setting seed point, candidate
The mode of point determines mesh approximation parallelogram in conjunction with " parallelogram law ", parallel finally by side continuation mesh approximation
Quadrangle carries out continuation to greatest extent, provides accurate parallelogram for attack type judgement and parameter calculating；
The image rectification that digital finger-print extracts is corrected mask image according to obtained RSD attack parameter, has
Body is divided into: if α ≠ 0, by mask image rotation alpha angle counterclockwise, ifSo by mask image horizontal twistingAngle；
In above embodiment, image procossing is carried out to the image after correction, is two-dimentional letter by the image definition after correction
Number f (x, y), wherein x, y are space coordinates, carry out image denoising processing, image to image f (x, y) using image denoising unit
Two-dimensional image function after denoising is p (x, y), wherein
It reuses image smoothing unit to be smoothed above-mentioned image p (x, y), the image after picture smooth treatment
Two-dimensional function is h (x, y), and wherein smooth function is g (x, y),
(x, y) , ﹡ are convolution symbol to h (x, y)=p (x, y) * g, and σ is customized adjustable constant, and smooth effect is to pass through σ
Come what is controlled,
Image enhancement processing finally is carried out to above-mentioned image h (x, y) using image enhancing unit, by image enhancement processing
Two-dimensional image function is u (x, y) afterwards, wherein
U (x, y)=h (x, y)-h (x-1, y)+h (x+1, y)+h (x, y-1)-h (x, y+1)+h (x+1, y+1).
Image synchronization positioning and fingerprint extraction and recovery, embedded block are B=(Iij)8×8, F=(Fuv)8×8For the DCT system of B
It counts, wherein IijA pixel value, F are arranged for the i-th row j of DCT embedded block BuvA pixel value is arranged for the u row v in DCT coefficient matrix,
It is located at the finger print data of embedded block insertion n-bit, fingerprint bit to be embedded is denoted as wi respectively, and the embedded location of selection is uvi.
The present invention provide a kind of heating ablation model data processing method identification process based on support vector regression include with
Lower step:
Step S1: data training set is obtained；
For obtaining data sample training set；It include the set of data samples of multiple classifications in the data sample training set
It closes, includes multiple data samples in the data sample set of each classification；
Step S2: interpolation calculation processing is carried out to the image data in data training set；
S201: closest 6 known pixels regions around interpolation pixel are determined；
S202: it is supported vector machine training；
Each support vector machines is respectively trained in the number that support vector machines is determined according to the case where pixel to be inserted into, instruction
Each pixel in white silk in original image is the input sample of support vector machines, and input pattern includes adjacent 6 in selection area
The gray value of known pixels, and the local spaces characteristic such as average gray, gray scale difference of adjacent 6 known pixels.
S203: interpolation calculation is carried out to each pixel to be estimated using the support vector machines for completing training；
Support vector machines input pattern in calculating is identical with the input pattern of training, and the output of support vector machines is exactly to insert
It is worth result.
Step S3: sample is chosen；
For executing the step of choosing similar sample and foreign peoples's sample；For each of described data sample training set
Data sample randomly selects k and belongs to the other data sample of same class as similar sample with the data sample, randomly selects
The k data samples to belong to a different category with the data sample are as foreign peoples's sample.
Step S4: disaggregated model produces module；
For, to set, similitude judgment models being obtained using support vector machines training, according to institute for the difference sample
It states similitude judgment models and obtains disaggregated model；
Kernel function can be used to obtain the similitude judgment models for the training of the support vector machines of Gaussian radial basis function.
Disaggregated model can be obtained according to a similitude judgment models, classification mould can also be obtained according to multiple similitude judgment models
Type.When obtaining disaggregated model according to a similitude judgment models, the disaggregated model can judge mould with the similitude
Type is identical；When obtaining disaggregated model according to multiple similitude judgment models, the disaggregated model can be multiple similitudes and sentence
The set of disconnected model.
If x, z ∈ X, X belong to the space R (n), nonlinear function Φ realizes the mapping of input space X to feature space F,
Middle F belongs to R (m), n < < m；Had according to kernel function technology:
K (x, z)=<Φ (x), Φ (z)>；
Wherein:<,>it is inner product, K (x, z) is kernel function；The inner product operation of m dimension higher dimensional space is converted n dimension by kernel function
The kernel function of the low-dimensional input space calculates, to solve the dimension disaster calculated in high-dimensional feature space；
Gaussian radial basis function in this step refers to the gaussian kernel function in radial basis function, and radial basis function is edge
The scalar function of radial symmetric；It is normally defined the monotonic function of Euclidean distance between any point x to a certain center xc in space,
It can be denoted as k (| | x-xc | |), effect is often part, i.e. the function value very little when x is far from xc.
Most common radial basis function is gaussian kernel function, and form isWherein xc
For kernel function center, σ is the width parameter of function, controls the radial effect range of function.
Step S5: data identification is carried out using disaggregated model
S501: data sample to be identified is obtained.
S502: k number is randomly selected from the sample set of each classification of the data training set respectively according to sample, is generated
2k difference samples pair to be identified, obtain difference sample to be identified to set.
S503: the difference sample to be identified analyzes set using the disaggregated model, obtains described to be identified
The similarity probabilities of every one kind in data sample and the data training set.
S504: according to the similarity probabilities, the classification of the data sample ownership to be identified is determined.
Wherein, S503 analyzes set the difference sample to be identified using the disaggregated model, refers to institute
Difference sample to be identified is stated to as input value, the disaggregated model is substituted into and is calculated.The calculated result of disaggregated model indicates institute
State the similarity probabilities of every one kind in data sample to be identified and the data training set.It can be according to the big of similarity probabilities
It is small, by the corresponding classification of maximum similarity probabilities, it is determined as the classification of the data sample ownership to be identified.
It should be noted last that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting.Although ginseng
It is described the invention in detail according to embodiment, those skilled in the art should understand that, to technical side of the invention
Case is modified or replaced equivalently, and without departure from the spirit and scope of technical solution of the present invention, should all be covered in the present invention
Scope of the claims in.
Claims (7)
1. a kind of heating ablation model data processing method based on support vector regression, which is characterized in that it is described based on support to
The heating ablation model data processing method that amount returns includes that data set training obtains module (1), image data difference processing module
(2), sample chooses module (3), disaggregated model production module (4) and data identification module (5)；
Wherein, the data set training obtains module (1) for obtaining data sample training set, includes in data sample training set
There is the data sample set of multiple classifications, includes multiple data samples in the data sample set of each classification；Described image number
It is used to carry out interpolation calculation processing to the image data in data training set according to difference processing module (2)；The sample chooses mould
Block (3) is for choosing similar sample and foreign peoples's sample；Disaggregated model production module (4) is for adopting difference sample to set
Similitude judgment models are obtained with support vector machines training, and obtain disaggregated model according to similitude judgment models；The data
Identification module (5) includes data sample acquiring unit to be identified, for obtaining data sample to be identified.
2. the heating ablation model data processing method according to claim 1 based on support vector regression, which is characterized in that
Described image data difference processing module (2) carries out n times wavelet transformation to image and retains the son of low frequency part after wavelet transformation
Image, the width and height of subgraph are original imageSubgraph area is original image areaIn subgraph into
Row matching, using normalized crosscorrelation measure, the traversal search on subgraph, retain in search process several similarities compared with
After search, the higher match point of similarity is mapped in the search subregion of original image to be matched for high match point,
The formula of the normalized crosscorrelation measure is as follows:
Wherein, S (x, y) indicates gray value of the coordinate position for pixel at (x, y), T (u, v) expression in original image to be matched
Coordinate position is the gray value of pixel at (u, v) in template image, and template image size is m × n, and m, n are positive integer.
3. the heating ablation model data processing method according to claim 1 based on support vector regression, which is characterized in that
The sample chooses module (3) for the step of choosing similar sample and foreign peoples's sample: for every in data sample training set
One data sample randomly selects k and belongs to the other data sample of same class as similar sample, at random with the data sample
A data sample to belong to a different category with the data sample of k is chosen as foreign peoples's sample, the sample chooses module according to evil
The false-alarm probability of the malicious attack mode computation overall situation of meaning node；
The first step is secondary user's Cri, i=1 ... the k that each participates in cooperative sensing according to the signal-to-noise ratio γ i of each node,
Design a weightThen linear weighted function is carried out to the signal energy statistic Ui that collection obtains to obtain most
The statistic of whole signal energy
Second step, analysis false-alarm malicious attack mode are influenced caused by frequency spectrum perception, obtain global false-alarm probability PfIt is general with attack
Rate Pa, the function expression between attack strength Δ it is as follows:
4. the heating ablation model data processing method according to claim 1 based on support vector regression, which is characterized in that
The data identification module (5) includes data sample acquiring unit to be identified, difference sample to be identified to set generation unit, similar
Property probability calculation unit and classification determination unit；
Wherein, data sample acquiring unit to be identified generates set for obtaining data sample to be identified, difference sample to be identified
Unit generates 2k wait know for randomly selecting k number from the sample set of each classification of data training set respectively according to sample
Other difference sample pair obtains difference sample to be identified to set, and similarity probabilities computing unit is used for using disaggregated model to be identified
Difference sample analyzes set, obtains the similarity probabilities of every one kind in data sample to be identified and data training set, classification
Determination unit is used to determine the classification of the data sample ownership to be identified according to the similarity probabilities.
5. the heating ablation model data processing method according to claim 4 based on support vector regression, which is characterized in that
Specific step is as follows for the anti-RSD attack blind Detecting digital fingerprinting method of data identification module (5):
Step 1: the generation of finger print data frame is encrypted finger print information using grouping displacement scrambling algorithm, is encoded using frame
Technical treatment finger print information obtains finger print data frame；
Step 2: being embedded in finger print data frame in DCT domain, fragment is carried out to carrier image, is respectively embedded into finger in each complete fragment
Line data frame forms multiple redundancy versions of finger print data frame；Carrier image is divided into the fragment that size is S × S first, wherein
S=2k, then select m 8 × 8 block of pixels as embedded block in each fragment, fingerprint be finally respectively embedded into embedded block
The data of n bit in data frame should meet L=m × n if the binary length of finger print data frame is L；
Step 3: constructing the Differential Characteristics dot grid with constant spacing, fixed difference value in airspace, refer to embedded with number
In the carrier image of line, with certain pixel (i0, j0) it is starting point, by line-spacing and column away from being embedded in difference in the picture in the way of being D
Divide characteristic point, to form a rectangular mesh in whole image；
The acquisition of step 4:RSD attack parameter；The doubtful mesh point on airspace is obtained first, then passes through setting seed point, time
The mode of reconnaissance determines mesh approximation parallelogram in conjunction with " parallelogram law ", flat finally by side continuation mesh approximation
Row quadrangle carries out continuation to greatest extent, provides accurate parallelogram for attack type judgement and parameter calculating；
Step 5: the image rectification that digital finger-print extracts is corrected mask image according to obtained RSD attack parameter,
It is specifically divided into: if α ≠ 0, by mask image rotation alpha angle counterclockwise, ifSo mask image level is turned round
It is bentAngle；
Step 6: image synchronization positioning and fingerprint extraction and recovery, embedded block are B=(Iij)8×8, F=(Fuv)8×8For the DCT of B
Coefficient, wherein IijA pixel value, F are arranged for the i-th row j of DCT embedded block BuvA pixel is arranged for the u row v in DCT coefficient matrix
Value, is located at the finger print data of embedded block insertion n-bit, and fingerprint bit to be embedded is denoted as wi respectively.
6. the heating ablation model data processing method according to claim 5 based on support vector regression, which is characterized in that
Before the image synchronization positioning and fingerprint extraction and recovery of step 6, the image after the image rectification of digital fingerprint extraction is believed
Breath carries out image procossing, is two-dimensional function f (x, y) by the image definition after correction, wherein x, y are space coordinates, are used first
Image denoising unit carries out image denoising processing to image f (x, y), and two-dimensional image function of the image after denoising is p
(x, y), wherein
It reuses image smoothing unit to be smoothed above-mentioned image p (x, y), the two-dimensional image after picture smooth treatment
Function is h (x, y), and wherein smooth function is g (x, y),
(x, y) , ﹡ are convolution symbol to h (x, y)=p (x, y) * g, and σ is customized adjustable constant, and smooth effect is controlled by σ
System,
Image enhancement processing finally is carried out to above-mentioned image h (x, y) using image enhancing unit, is schemed after image enhancement processing
As two-dimensional function is u (x, y), wherein
U (x, y)=h (x, y)-h (x-1, y)+h (x+1, y)+h (x, y-1)-h (x, y+1)+h (x+1, y+1).
7. a kind of heating ablation model data using described in any claim in claim 1-6 based on support vector regression
The method of processing method, which is characterized in that the method the following steps are included:
Step 1: data training set is obtained: for obtaining data sample training set；It include more in the data sample training set
The data sample set of a classification includes multiple data samples in the data sample set of each classification；
Step 2: interpolation calculation processing is carried out to the image data in data training set: to the image data in data training set
Carry out interpolation calculation processing；
Step 3: sample is chosen: for executing the step of choosing similar sample and foreign peoples's sample；For data sample training
Each of collection data sample randomly selects k and belongs to the other data sample of same class as similar with the data sample
Sample randomly selects a data sample to belong to a different category with the data sample of k as foreign peoples's sample；
Step 4: disaggregated model produces module: for, to set, obtaining phase using support vector machines training for the difference sample
Like property judgment models, disaggregated model is obtained according to the similitude judgment models；
Kernel function is used to obtain the similitude judgment models for the training of the support vector machines of Gaussian radial basis function；It can basis
One similitude judgment models obtains disaggregated model, can also obtain disaggregated model according to multiple similitude judgment models；Work as root
When obtaining disaggregated model according to a similitude judgment models, the disaggregated model can be identical as the similitude judgment models；
When obtaining disaggregated model according to multiple similitude judgment models, the disaggregated model can be multiple similitude judgment models
Set；
If x, z ∈ X, X belong to the space R (n), nonlinear function Φ realizes input space X to the mapping of feature space F, and wherein F belongs to
In R (m), n < < m；Had according to kernel function technology:
K (x, z)=<Φ (x), Φ (z)>；
Wherein:<,>it is inner product, K (x, z) is kernel function；From formula as can be seen that kernel function transports the inner product of m dimension higher dimensional space
The kernel function calculating for being converted into the n dimension low-dimensional input space is calculated, to solve the dimension disaster calculated in high-dimensional feature space；
Gaussian radial basis function in this step refers to the gaussian kernel function in radial basis function, and radial basis function is radially
Symmetrical scalar function；It is normally defined the monotonic function of Euclidean distance between any point x to a certain center xc in space, can be remembered
Make k (| | x-xc | |)；
Step 5: data identification, data sample acquiring unit to be identified, for obtaining data to be identified are carried out using disaggregated model
Sample；Difference sample to be identified is to set generation unit, for respectively from the sample set of each classification of the data training set
K number is randomly selected according to sample, 2k difference samples pair to be identified is generated, obtains difference sample to be identified to set；Similarity probabilities
Computing unit is obtained described to be identified for being analyzed set using the disaggregated model the difference sample to be identified
The similarity probabilities of every one kind in data sample and the data training set；Classification determination unit, for according to the similitude
Probability determines the classification of the data sample ownership to be identified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811197853.3A CN109508732A (en) | 2018-10-15 | 2018-10-15 | A kind of heating ablation model data processing method based on support vector regression |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811197853.3A CN109508732A (en) | 2018-10-15 | 2018-10-15 | A kind of heating ablation model data processing method based on support vector regression |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109508732A true CN109508732A (en) | 2019-03-22 |
Family
ID=65746588
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811197853.3A Pending CN109508732A (en) | 2018-10-15 | 2018-10-15 | A kind of heating ablation model data processing method based on support vector regression |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109508732A (en) |
Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107688829A (en) * | 2017-08-29 | 2018-02-13 | 湖南财政经济学院 | A kind of identifying system and recognition methods based on SVMs |
-
2018
- 2018-10-15 CN CN201811197853.3A patent/CN109508732A/en active Pending
Patent Citations (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107688829A (en) * | 2017-08-29 | 2018-02-13 | 湖南财政经济学院 | A kind of identifying system and recognition methods based on SVMs |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109408653B (en) | Human body hairstyle generation method based on multi-feature retrieval and deformation | |
| JP4921858B2 (en) | Image processing apparatus and image processing program | |
| CN107633226B (en) | Human body motion tracking feature processing method | |
| CN109584251A (en) | A kind of tongue body image partition method based on single goal region segmentation | |
| CN107977952A (en) | Medical image cutting method and device | |
| CN107229917A (en) | A kind of several remote sensing image general character well-marked target detection methods clustered based on iteration | |
| CN106408039A (en) | Off-line handwritten Chinese character recognition method carrying out data expansion based on deformation method | |
| CN105931253A (en) | Image segmentation method combined with semi-supervised learning | |
| Ghose et al. | A random forest based classification approach to prostate segmentation in MRI | |
| CN107451594B (en) | Multi-view gait classification method based on multiple regression | |
| He et al. | Automatic magnetic resonance image prostate segmentation based on adaptive feature learning probability boosting tree initialization and CNN-ASM refinement | |
| CN110705565A (en) | Lymph node tumor region identification method and device | |
| CN102831614B (en) | Sequential medical image quick segmentation method based on interactive dictionary migration | |
| CN108629762B (en) | Image preprocessing method and system for reducing interference characteristics of bone age evaluation model | |
| CN108986114B (en) | Automatic abdominal CT sequence image liver segmentation method based on level set and shape descriptor | |
| Park et al. | Bayesian manifold learning: the locally linear latent variable model (LL-LVM) | |
| CN110263804A (en) | A kind of medical image dividing method based on safe semi-supervised clustering | |
| Wang et al. | A novel sparse boosting method for crater detection in the high resolution planetary image | |
| CN108596195A (en) | A kind of scene recognition method based on sparse coding feature extraction | |
| CN108090913B (en) | Image semantic segmentation method based on object-level Gauss-Markov random field | |
| CN116343048A (en) | Accurate land block boundary extraction method and system for plain crop type complex region | |
| CN103456017A (en) | Image segmentation method of semi-supervised weight kernel fuzzy clustering based on seed set | |
| CN102663728B (en) | Dictionary learning-based medical image interactive joint segmentation | |
| Dong et al. | An improved YOLOv5 network for lung nodule detection | |
| CN106909881A (en) | The method and system of corn breeding base ridge number are extracted based on unmanned aerial vehicle remote sensing images |
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: 20190322 |