CN104063687B - A kind of data component of degree n n extracting method based on neutral net - Google Patents
A kind of data component of degree n n extracting method based on neutral net Download PDFInfo
- Publication number
- CN104063687B CN104063687B CN201410273098.8A CN201410273098A CN104063687B CN 104063687 B CN104063687 B CN 104063687B CN 201410273098 A CN201410273098 A CN 201410273098A CN 104063687 B CN104063687 B CN 104063687B
- Authority
- CN
- China
- Prior art keywords
- data
- degree
- component
- neutral net
- vector
- 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.)
- Expired - Fee Related
Links
Abstract
The invention discloses a kind of data component of degree n n extracting method of neutral net, the method carries out the pretreatment of data first, such as the data of an image, using the method for image array piecemeal, the gray value data in each image block is changed a column vector；Build component of degree n n analysis neutral net；Initialization weight vector and k values；The input value that a column vector analyzes neutral net as component of degree n n is randomly selected in column vector is obtained；Calculate the output that component of degree n n analyzes neutral net；It is iterated calculating and updates weight vector value, whether check algorithm restrains, such as restrain, this vector for obtaining is exactly the characteristic vector corresponding to the minimal eigenvalue of the incidence matrix of the data set.The invention has the beneficial effects as follows the defect that prior art can be overcome to exist, extracts the weak feature in data, can be used to detecting cigarette in video, mist, the important information such as dust.
Description
Technical field
The invention belongs to technical field of data processing, is related to a kind of data component of degree n n extracting method based on neutral net.
Background technology
The extraction of data component of degree n n is for the letter such as Wave beam forming, straight line/surface fitting, the cigarette found in image, mist, dust
Breath has important effect.The component of degree n n of data, be data correlation matrix minimal eigenvalue corresponding to characteristic vector.?
In the component of degree n n extraction process of input data, conventional method is matrix method.It first asks for input data and (is typically expressed as arranging
Vector form) incidence matrix, then solve an eigenvalue problem of incidence matrix to obtain component of degree n n, this method cannot be located
The large-scale data of reason higher-dimension, and the component of degree n n extraction of online data can not be carried out.For example：If data are the row of 10000 dimensions
Vector, then its incidence matrix is the square formation of a 10000x10000, processes memory space and calculating that the incidence matrix needs
Resource is all very big.
Content of the invention
It is an object of the invention to provide a kind of data component of degree n n extracting method based on neutral net, solves existing
Method cannot process the large-scale data of higher-dimension, and can not carry out the problem that the component of degree n n of online data is extracted.
The technical solution adopted in the present invention is to follow the steps below：
Step 1：Carry out the pretreatment of data, such as the data of an image, using the method for image array piecemeal,
Gray value data in each image block is changed a column vector；
Step 2：Component of degree n n analysis neutral net is built, its input/output relation is y (k)=w (k)Tx(k)；
Step 3：It is 0 to initialize weight vector and make k values；
Step 4：From step 1 obtain randomly selecting in column vector set a column vector as neutral net input to
Amount；
Step 5：Calculate the output that component of degree n n analyzes neutral net；
Step 6：It is iterated calculating and updates weight vector value, iterative formula is：
Step 7：Whether check algorithm restrains, and the condition of convergence is：
＜ ∈, wherein, ∈ is 0.01 to | | w (k+1)-w (k) | |；If the condition of convergence meets,
Then think that algorithm is restrained, then weight vector w (k+1) is exactly the incidence matrix of the data set
Minimal eigenvalue corresponding to characteristic vector, i.e. the component of degree n n of data set, algorithm terminate；
Otherwise, k=k+1 is taken, iterative steps add 1, returns execution step 4.
The invention has the beneficial effects as follows the defect that prior art can be overcome to exist, extracts the component of degree n n in data, can use
The important informations such as cigarette, mist, dust in detection video.
Description of the drawings
Fig. 1 is that data component of degree n n of the present invention extracts flow chart；
Fig. 2 is Image semantic classification schematic diagram of the present invention；
Fig. 3 is component of degree n n analysis neutral net schematic diagram of the present invention.
Specific embodiment
The present invention is described in detail with reference to the accompanying drawings and detailed description.
It is illustrated in figure 1 the inventive method flow chart.The invention discloses a kind of data component of degree n n based on neutral net
Extracting method, analyzes the component of degree n n of neural net method On-line testing data including data prediction and using component of degree n n.
The present invention first carries out pretreatment to the data being input into and generates data matrix, then analyzes neutral net using component of degree n n
Algorithm for Solving component of degree n n, and then complete the extraction to weak feature.
1st, the pretreatment of view data：
For the data of an image, using the method for image array piecemeal, by the gray value number in each image block
According to one column vector of conversion；
During Image semantic classification, image is divided into m × n blocks first.As shown in Fig. 2 entire image above is divided
Into 8*8=64 blockage, these blockages from left to right, are numbered from top to bottom, and the image block of straight line indication is located at
5th row of whole image, the 7th row, it is thus evident that its numbering should be：(5-1)*8+7.One image is divided into 8 × 8 pieces.So
Afterwards, the image intensity value data in each image block are converted to a column vector.
For example, for an image block,
Wherein, i=1,2 ..., m × n.By 9 data in the image block on to
Under, it is x ' after being from left to right sequentially converted into a column vectori=[x11, x21, x31, x12, x22, x32, x13, x23, x33]T.
Finally, the column vector that the data conversion by all image blocks is obtained is constituted a data acquisition system, is entirely schemed
The data acquisition system of picture is X=[x '1, x '2..., x 'm×n].
2nd, component of degree n n analysis neutral net is built, as shown in Figure 3：
Network inputs/the output relation of structure is：Y (k)=w (k)Tx(k).
3rd, initialization weight vector w (k) is w (0)=[w1(0) w2(0)...wm(0)]T, by the value of each of which element all
It is set as the initial value of random generation.Initialization k=0.
4th, randomly select in the column vector data acquisition system X that step 1 is obtained data (i.e. in X matrix arrange to
Amount) as neutral net in step 2 input value x (k)=[x1(k), x2(k) ..., xm(k)]T.
5th, output y (k)=w (k) that component of degree n n analyzes neutral net is calculatedTX (k), wherein, x (k)=[x1(k), x2
(k) ..., xm(k)]T, w (k)=[w1(k), w2(k) ..., wm(k)]T.
6th, the value of weight vector is updated by the learning algorithm of following iteration.
Iterative formula is：
Wherein, multiplying is represented.
7th, whether check algorithm restrains.The condition of convergence is：| | w (k+1)-w (k) | | ＜ ∈, wherein, ∈ is given for one
Very little number, be usually taken to be 0.01.If above-mentioned condition has met, then it is assumed that algorithm is restrained, then output is extracted
Component of degree n n w (k+1), algorithm terminates, and this vector for obtaining is exactly corresponding to the minimal eigenvalue of the incidence matrix of the data set
Characteristic vector.
Otherwise, k=k+1, iterative steps add 1；
Return execution step 4.
Beneficial effects of the present invention：
1st, online the data being input into can be carried out with component of degree n n extraction, the component of degree n n of the newest data of energy synchronization gain.
As can be seen that being input into data vector x (k) every time to update the value of component of degree n n w (k+1) from formula (1).(note：With online
The data that mode is obtained as input data x (k) of neutral net, and can carry out the extraction of component of degree n n by said method,
Complete data set is not needed.And matrix method needs complete data set, thus off-line data can only be processed).
2nd, without calculating and the correlation matrix of data storage, computation complexity and data space are greatly reduced.
(note：The calculating of vector sum scalar is pertained only in formula 1, and our method can be in the situation of the incidence matrix for disregarding the evidence that counts
Under, obtain the characteristic vector corresponding to the minimal eigenvalue of the matrix).
3rd, adaptive Learning Step, without setting Learning Step, enhances its availability in actual applications.(note：
Without unknown parameter in formula 1).
The above is only to presently preferred embodiments of the present invention, not makees any pro forma restriction to the present invention,
Every technical spirit according to the present invention is belonged to any simple modification made for any of the above embodiments, equivalent variations and modification
In the range of technical solution of the present invention.
Claims (1)
1. the data component of degree n n extracting method of a kind of neutral net, it is characterised in that follow the steps below：
Step 1：The pretreatment of view data is carried out, for the data of an image, using the method for image array piecemeal, will be per
Gray value data in one image block changes a column vector；
Step 2：Build component of degree n n analysis neutral net y (k)=w (k)Tx(k)；
Step 3：Initialization weight vector and k values；
Step 4：Obtain in column vector, randomly selecting the input that a column vector analyzes neutral net as component of degree n n in step 1
Value；
Step 5：Calculate the output that component of degree n n analyzes neutral net；
Step 6：It is iterated calculating and updates weight vector value, iterative formula is：
Step 7：Whether check algorithm restrains, and the condition of convergence is：＜ ∈, wherein, ∈ is 0.01 to | | w (k+1)-w (k) | |；Such as
The fruit condition of convergence meets, then it is assumed that algorithm is restrained, then export the component of degree n n w (k+1) for extracting, and this vector for obtaining is exactly
Characteristic vector corresponding to the minimal eigenvalue of the incidence matrix of the data set, algorithm terminate；Otherwise, k=k+1, iteration step are taken
Number Jia 1, returns execution step 4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410273098.8A CN104063687B (en) | 2014-06-18 | 2014-06-18 | A kind of data component of degree n n extracting method based on neutral net |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410273098.8A CN104063687B (en) | 2014-06-18 | 2014-06-18 | A kind of data component of degree n n extracting method based on neutral net |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN104063687A CN104063687A (en) | 2014-09-24 |
| CN104063687B true CN104063687B (en) | 2017-03-15 |
Family
ID=51551392
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410273098.8A Expired - Fee Related CN104063687B (en) | 2014-06-18 | 2014-06-18 | A kind of data component of degree n n extracting method based on neutral net |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN104063687B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101634709A (en) * | 2009-08-19 | 2010-01-27 | 西安电子科技大学 | Method for detecting changes of SAR images based on multi-scale product and principal component analysis |
| CN101908206A (en) * | 2010-07-01 | 2010-12-08 | 西北工业大学 | Morphological component analysis (MCA)-based synthetic aperture radar (SAR) image noise suppression method |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| FR2827060B1 (en) * | 2001-07-05 | 2003-09-19 | Eastman Kodak Co | METHOD FOR IDENTIFYING THE SKY IN AN IMAGE AND IMAGE OBTAINED THANKS TO THIS PROCESS |
| KR100669251B1 (en) * | 2005-11-25 | 2007-01-16 | 한국전자통신연구원 | Apparatus and method for automatically analyzing digital image quality |
-
2014
- 2014-06-18 CN CN201410273098.8A patent/CN104063687B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101634709A (en) * | 2009-08-19 | 2010-01-27 | 西安电子科技大学 | Method for detecting changes of SAR images based on multi-scale product and principal component analysis |
| CN101908206A (en) * | 2010-07-01 | 2010-12-08 | 西北工业大学 | Morphological component analysis (MCA)-based synthetic aperture radar (SAR) image noise suppression method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN104063687A (en) | 2014-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110377984B (en) | Industrial equipment residual effective life prediction method and system and electronic equipment | |
| US11544522B2 (en) | Methods, systems, and computer readable mediums for determining a system state of a power system using a convolutional neural network | |
| CN107463966B (en) | Radar range profile's target identification method based on dual-depth neural network | |
| CN106803071A (en) | Object detecting method and device in a kind of image | |
| CN108052881A (en) | The method and apparatus of multiclass entity object in a kind of real-time detection construction site image | |
| CN107169598A (en) | A kind of weather prediction method and system based on deep learning | |
| CN108448610A (en) | A kind of short-term wind power prediction method based on deep learning | |
| CN107766894A (en) | Remote sensing images spatial term method based on notice mechanism and deep learning | |
| CN105931264B (en) | A kind of sea infrared small target detection method | |
| CN106597574A (en) | Weather temperature prediction method and device based on time-varying cloud model | |
| CN109828552B (en) | Intermittent process fault monitoring and diagnosing method based on width learning system | |
| CN109740655B (en) | Article scoring prediction method based on matrix decomposition and neural collaborative filtering | |
| CN109543662A (en) | Object detection method, system, device and the storage medium proposed based on region | |
| CN107392128A (en) | The robust image recognition methods returned based on double low-rank representations and local constraint matrix | |
| CN106570522A (en) | Object recognition model establishment method and object recognition method | |
| CN105607631B (en) | The weak fault model control limit method for building up of batch process and weak fault monitoring method | |
| CN106649479A (en) | Probability graph-based transformer state association rule mining method | |
| CN112508190A (en) | Method, device and equipment for processing structured sparse parameters and storage medium | |
| CN106203628A (en) | A kind of optimization method strengthening degree of depth learning algorithm robustness and system | |
| Kim et al. | Untangling the contribution of input parameters to an artificial intelligence PM2. 5 forecast model using the layer-wise relevance propagation method | |
| CN105550712A (en) | Optimized convolution automatic encoding network-based auroral image sorting method | |
| He et al. | DTCWT-based zinc fast roughing working condition identification | |
| Dewandra et al. | Journal Unique Visitors Forecasting Based on Multivariate Attributes Using CNN | |
| Michelle et al. | A bibliometric analysis of generative design, algorithmic design, and parametric design in architecture | |
| CN104063687B (en) | A kind of data component of degree n n extracting method based on neutral net |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C41 | Transfer of patent application or patent right or utility model | ||
| CB03 | Change of inventor or designer information |
Inventor after: Peng Dezhong Inventor after: Zhang Lijun Inventor after: Lin Yi Inventor after: Liu Jie Inventor after: Liu Wen Inventor after: Yu Honghong Inventor before: Peng Dezhong Inventor before: Zhang Haixian Inventor before: Guo Jixiang Inventor before: Sang Yongsheng Inventor before: Mao Hua Inventor before: Zhen Liangli |
|
| COR | Change of bibliographic data | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20170112 Address after: 610041 Chengdu high tech Zone, Xiaojiahe, No. 11, Sichuan street, building 2, building 1 Applicant after: CHENGDU RUIBEI YINGTE INFORMATION TECHNOLOGY CO., LTD. Address before: 610065 Chenghua District, Chengdu, North Road, building No. two, No. 4, No. Applicant before: Peng Dezhong |
|
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20170315 Termination date: 20180618 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |