CN109190638A - Classification method based on the online order limit learning machine of multiple dimensioned local receptor field - Google Patents
Classification method based on the online order limit learning machine of multiple dimensioned local receptor field Download PDFInfo
- Publication number
- CN109190638A CN109190638A CN201810902594.3A CN201810902594A CN109190638A CN 109190638 A CN109190638 A CN 109190638A CN 201810902594 A CN201810902594 A CN 201810902594A CN 109190638 A CN109190638 A CN 109190638A
- Authority
- CN
- China
- Prior art keywords
- multiple dimensioned
- online
- data set
- receptor field
- local receptor
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention discloses a kind of classification methods based on the online order limit learning machine of multiple dimensioned local receptor field, belong to field of image processing, it is mainly used in image classification, it mainly includes two parts, is respectively as follows: the initial learning period based on multiple dimensioned local receptor field and the on-line study stage based on multiple dimensioned local receptor field.The present invention is based on the algorithms (MSLRF+OSELM) of the online order limit learning machine of multiple dimensioned local receptor field for classifying, it not only can be used the image data generated online and carries out dynamic training, and the feature of the multiple dimensioned representative complex texture of local receptor field extraction height can be passed through, to substantially increase the precision of classification, there is actual use value well.
Description
Technical field
The invention belongs to field of image processings, and in particular to one kind is based on the online order limit of multiple dimensioned local receptor field
The classification method of habit machine.
Background technique
In recent years, the classification of surfacing causes the extensive concern of academia and industry.Traditional scholar utilizes static state
Data classify to surfacing.But in actual industrial production, data can not generate overnight, it is to connect
Continuous generation, mass data can be generated successively.And contemporary on-line study is the technology to grow up from the industrial revolution, to replace
Cheap labour and the workload for reducing worker.Huang GB et al. (Huang GB, Bai Z, Kasun LLC, et
al.Local receptive fields based extreme learning machine[J].IEEE
Computational Intelligence Magazine.2015;10 (2): 18-29.) one kind is proposed based on local experiences
The sorting algorithm of wild extreme learning machine, this method have good classification performance, but due to the unicity of local receptor field scale
Very complete characteristics of image cannot be obtained, the performance of classification is leveraged.Liang N Y et al. (Liang N Y, Huang G
B,Saratchandran P,et al.A fast and accurate online sequential learning
algorithm for feedforward networks.[J].IEEE Trans Neural Netw,2006,17(6):
1411-23.) propose a kind of fast and accurately online serial order learning algorithm of feedforward network.This method limit of utilization learning machine
The method that thought develops batch study, with good classification performance.But how to extract useful information from image is
One main problem.
Summary of the invention
Due to depending on property of the conventional images data classification to static image data will lead to classification accuracy it is low and point
Time-consuming for class, therefore proposes that the present invention is used to solve the problems, such as this, while special in single scale local receptor field method in order to make up
The limitation extracted is levied, a kind of method based on the online order limit learning machine of multiple dimensioned local receptor field is proposed.
The present invention is achieved by the following technical scheme:
A kind of classification method based on the online order limit learning machine of multiple dimensioned local receptor field mainly includes two portions
Point, the initial learning period respectively based on multiple dimensioned local receptor field and the on-line study based on multiple dimensioned local receptor field
Stage specifically comprises the following steps:
Step S1, the initial learning period based on multiple dimensioned local receptor field passes through initial data set DpIt calculates single hidden
The initial output weight of layer feedforward neural network, is arranged p=0.Steps are as follows:
Step S11), first on matlab using respective function by it is down-sampled to image progress RGB color triple channel
Separation, and R is obtained respectively, G, B single channel image vector.
Step S12), with matlab, be respectively randomly generated on three Color Channels in conjunction with MSLRF+OSELM algorithm
The initial weight of S scale, and singular value decomposition orthogonalization is carried out to initial weight, and with the initial weight after orthogonalization to list
The characteristic pattern of the input layer of hidden layer feedforward neural network is connected with the characteristic pattern of hidden layer, generates characteristics of image.
It is very uniform using the random weight distribution of singular value Orthogonal Decomposition, more complete feature can be extracted separation
Color vector, so that image data more Line independent and easy classification.
Step S13), Analysis On Multi-scale Features mapping: using the multiple dimensioned local receptor field generated by continuous probability, to step
S12) characteristics of image generated carries out convolution Feature Mapping.
Step S14), multiple dimensioned pond: to convolution Feature Mapping formed combined joint value carry out square root pond, and
Obtain the low-level image feature of colored subgraph.
Square root pond makes neural network have the advantages that translation invariance.
Step S15), full connection: the low-level image feature of all obtained colored subgraphs is grouped together, picture number is obtained
According to low-level image feature.
Step S16), calculate initial output weight.
Step S2, based on the on-line study stage of multiple dimensioned local receptor field, that is, single sample or sample data are utilized
Block updates the output weight learnt in the initial stage.Steps are as follows:
Step S21), the D based on multiple dimensioned local receptor fieldp+1The online Analysis On Multi-scale Features of data set map.Concrete operations
Step is identical as in step S13), wherein S scale channel, and the initial weight in each scale channel and biasing are equal to step S13)
In setting.
Step S22), Dp+1The online multiple dimensioned pond of data set, wherein pond size is equal to step S14) in parameter set
It sets.
Step S23), Dp+1The low-level image feature of all obtained colored subgraphs is combined to one by the full connection of data set
It rises, obtains the low-level image feature of image data.
Step S24), calculate Dp+1The output weight of data set.
Step S25), setting p=p+1, determine Dp+1Whether data set is the last one online data collection, if so, stopping
On-line study, otherwise repeat step S21)-step S24), until the last one block that data integration is on-line training data set
Data set.
Step S26), calculate the output weight of the last one online data collection.
The present invention as the important ring in image classification field, for the image data problem largely generated online have compared with
Good classifying quality and processing capability in real time.
The method of the present invention has the beneficial effect that:
1, the deficiency that multiple dimensioned local receptor field (MSLRF) compensates for the simple single scale of local receptor field method is introduced,
High representative feature can be extracted from the complicated image of variation.
2, MSLRF is combined with OS-ELM, integrates feature extraction and classification, it can be to avoid manual intervention.The party
Method can either extract complicated characteristics of image, can also inherit that OS-ELM method batch training network is fast, good excellent of classification performance
Gesture.
3, method proposed by the present invention can not only classify to static image data, but also can be to dynamic image
Data are effectively classified.
The present invention has rational design, proposes a kind of algorithm based on the online order limit learning machine of multiple dimensioned local receptor field
(MSLRF+OSELM) for classifying, it not only can be used the image data generated online and carries out dynamic training, but also can lead to
The feature of the multiple dimensioned representative complex texture of local receptor field extraction height is crossed, to substantially increase the precision of classification, is had
There is actual use value well.
Detailed description of the invention
Fig. 1 shows the structure charts of the present invention based on the online order limit learning machine method of multiple dimensioned local receptor field.
Fig. 2 indicates the existing structure chart based on multiple dimensioned local receptor field extreme learning machine.
Fig. 3 a indicates that the present invention and other methods are applied to the nicety of grading on ALOT data set.
Fig. 3 b indicates that the present invention and other methods are applied to the nicety of grading on MNIST data set.
Specific embodiment
Specific embodiments of the present invention are described in detail with reference to the accompanying drawing.
A method of based on the online order limit learning machine of multiple dimensioned local receptor field, in multiple dimensioned local receptor field
On the basis of introduce online order limit learning machine method so that this method has good classification to dynamic image data
Can, it is as shown in Figure 1 this method structure chart, which is made of p+1 MSLRF-NET.The data set generated online connects
The continuous corresponding network of input is to update output weight.The wherein specific structure of first MSLRF-NET such as Fig. 2 is in the present invention
Complete entire training process step 1: the initial learning period based on multiple dimensioned local receptor field, that is, pass through initial data set
Calculate the initial output weight of Single hidden layer feedforward neural networks.Specific step is as follows:
Step S11, color triple channel separates.Image is subjected to RGB triple channel color separated by down-sampled, and respectively
Obtain R, G, B single channel image vector.
Step S12, on three Color Channels, the initial weight of S scale is respectively randomly generated, and it is carried out orthogonal
Change operation.Isolated color vector is inputted corresponding Color Channel respectively, each Color Channel there are S different scales
Channel is adapted to complicated texture variations.It adequately indicates to obtain input picture more, is used in each scale channel
K different input weights, therefore in the characteristic pattern of the available K inequality in each scale channel of hidden layer, each color is logical
Road obtains S × K inequality characteristic pattern.Equipped with N number of training sample, input picture size is d × d, the size of s-th of scale LRFs
For rs×rs, then the size of characteristic pattern is (d-rs+1)×(d-rs+1)。
Step S13, make initial weight matrix orthogonalization to obtain orthogonal matrix using SVD method, and with orthogonalization after
Initial weight is connected to the characteristic pattern of the input layer of Single hidden layer feedforward neural networks with the characteristic pattern of hidden layer, generates image
Feature.
Step S14, Analysis On Multi-scale Features map.As shown in Fig. 2, MSLRF-ELM has S scale channel r1,r2,…rS, whole
A network can generate 3 × S × K inequality characteristic pattern at random, carry out convolution feature to the characteristics of image that step S13) is generated and reflect
It penetrates.
Step S15, multiple dimensioned average pond.Square root pond is carried out to the value for the combined joint that convolution Feature Mapping is formed
Change, and obtains the low-level image feature of subgraph.
Step S16, full connection.The low-level image feature of subgraph is grouped together, the low-level image feature of image is obtained.
Step S17, initial output weight is calculated.
There is new data set to enter below and then corresponds to subsequent MSLRF-NET.It is in the present invention entire training process
Step 2: the on-line study stage based on multiple dimensioned local receptor field, i.e., updated using single sample or sample data block
The output weight that initial stage learns.Specific step is as follows:
Step S21, the online Analysis On Multi-scale Features mapping of new data set.Concrete operation step is identical as in step S14,
Middle S scale channel, the initial weight in each scale channel and biasing are equal to the setting in step S14.
Step S22, the online multiple dimensioned pond of new data set, wherein pond size is equal to the setting of parameter in step S15.
Step S23, the full connection of new data set.The low-level image feature of all obtained colored subgraphs is grouped together,
Obtain the low-level image feature of image data.
Step S24, the output weight of new data set is calculated.
Step S25, determine whether the data set is the last one online data collection, if so, stop on-line study, it is on the contrary
Repeat step S21-step S24.
Step S26, the output weight of the last one online data collection is calculated.
Technical effect of the invention is further confirmed below by way of experiment:
The experimental situation of specific embodiment is matlab2016 in the present invention, is based on 64 8 operating system PC of windows,
Hardware configuration CPU Intel (R) Core (TM) i3-3110GM@2.4GHz, memory 4GB 1600MHz.Program code is based on
Matlab programming language is write, and wherein image procossing has used the processing function of matlab.
1, the acquisition and processing of data set
In order to verify the validity of MSLRF-OSELM, online experiment is carried out using ALOT data set.Entire ALOT data set
It is made of 25,000 images.It wherein include 250 kinds of natural materials, such as wood-fibred, rice, sugar, onion and wool felt etc..Data
Collection consists of two parts: color image and gray level image.The size of each image is 384 × 256.Firstly, for each classification with
Machine generates 20 samples as test sample, other samples are as training sample.In an experiment, it is contemplated that calculator memory and fortune
Scanning frequency degree, by ALOT data set sample it is down-sampled be 32 × 32.Different condition is come from view of data images, uses null component point
Analysis (ZCA) carries out whitening transformation to all samples to reduce the interference information of image.Muscle-setting exercise sample is converted to dynamic to increase
Training sample is measured with the online network of training.
There is good Generalization Capability for verifying MSLRF-OSELM, carry out another experiment using MNIST data set.Choosing
10000 samples are selected as test sample, other samples are as training sample.This process on-line training network is similar to ALOT
The processing of training dataset.
2, evaluation method
The performance of judgment method of the present invention is evaluated by testing the comparison of resulting measuring accuracy and testing time.
3, experimental data and analysis
1), the average test precision on ALOT data set
Average test precision on 1 ALOT data set of table
Four different experiments are carried out to multiple dimensioned and local receptor field ALOT data set, wherein LRF distinguishes
With 2 scales, 4 (d) (4 different scales) a scales, 4 (s) (4 same scales) and 8 scales.Three are shown in table 1
10 times of cross validation Average Accuracies of the combined effect of a important parameter, the scale including LRF, the quantity and block number of characteristic pattern
According to the quantity of collection.
From result, it is known that 4 kinds of difference LRF scale ratios have 4 kinds of same scale LRF in method proposed by the present invention
Traditional LRFELM nicety of grading it is higher.Thus, it can be observed that influence of the LRF scale to online network class precision: this hair
Point of the bright MSLRF-OSELM method to the classification performance of the LRF of 4 kinds of different scales better than the LRF of 2 kinds of scales and 8 kinds of scales
Class performance, multiple dimensioned LRF can extract the classification for being conducive to subject material with highly representative feature-rich.In addition,
It is also observed from table 1, when the quantity of block data set is 40, enough samples can make network have more differentiation features
And help correctly to classify to sample.
2), the average test precision on MNIST data set
Average test precision on 2 MNIST data set of table
Four different experiments are performed in an identical manner using announced MNIST data set.Three are shown in table 2
10 times of cross validation Average Accuracies of the combined effect of a important parameter.From table 2, it can be observed that working as each ratio channel
The quantity of characteristic pattern when being 20, optimal classification may be implemented using the method for the LRF of 8 kinds of different scales.
3) on ALOT data set and MNIST data set distinct methods nicety of grading
Fig. 3 a indicates that the present invention and other methods are applied to the nicety of grading on ALOT data set.Fig. 3 b indicate the present invention and
Other methods are applied to the nicety of grading on MNIST data set.The method of the present invention has better classification performance as shown in Figure 3.
4), on ALOT data set training and the test of distinct methods time
The time of training and the test of distinct methods on 4 ALOT data set of table
Method | MFS | WMFS | PLS | MSLRF-OSELM | MSLRF-ELM |
Training time (s) | 720.32 | 738.05 | 822.19 | 1290.45 | 238.09 |
Testing time (s) | 0.94 | 1.15 | 1.57 | 10.36 | 0.65 |
5), on MNIST data set training and the test of distinct methods time
The time of training and the test of distinct methods on 5 ALOT data set of table
Method | SVM | RELM | HOG+ELM | MSLRF-OSELM | MSLRF-ELM |
Training time (s) | 11.34 | 3.23 | 69.82 | 715.99 | 90.92 |
Testing time (s) | 5.62 | 0.94 | 0.93 | 19.25 | 1.34 |
It can be observed that the training and testing time cost maximum of proposition method of the present invention from table 4 and table 5, but from Fig. 3
In can be seen that the classification results of this method can be close to or handle online data more than the classification performance of above-mentioned other methods.
In short, the classification method of the present invention based on the online order limit learning machine of multiple dimensioned local receptor field, main
To be applied to image classification, include two parts, be respectively as follows: initial learning period and base based on multiple dimensioned local receptor field
In the on-line study stage of multiple dimensioned local receptor field.The present invention integrates feature extraction and real-time grading, avoids artificial
Intervene, has preferable classification performance and processing capability in real time for the online processing for generating image data.
The above is only the preferred embodiments of the present invention, and is not intended to limit the present invention in any form,
Any simple modification made to the above embodiment, equivalent variations belong to the present invention according to the technical essence of the invention
In the range of technical solution.
Claims (1)
1. a kind of classification method based on the online order limit learning machine of multiple dimensioned local receptor field, it is characterised in that: including such as
Lower step:
Step S1, the initial learning period based on multiple dimensioned local receptor field passes through initial data set DpCalculate single hidden layer feedforward
P=0 is arranged in the initial output weight of neural network;Specific step is as follows:
Step S11), first on matlab using respective function by it is down-sampled to image progress the separation of RGB color triple channel,
And R is obtained respectively, G, B single channel image vector;
Step S12), with matlab, in conjunction with MSLRF+OSELM algorithm, on three Color Channels, S is respectively randomly generated
The initial weight of scale, and singular value decomposition orthogonalization is carried out to initial weight, and with the initial weight after orthogonalization to single hidden
The characteristic pattern of input layer of layer feedforward neural network is connected with the characteristic pattern of hidden layer, generates characteristics of image;
Step S13), Analysis On Multi-scale Features mapping: to step S12) generate characteristics of image carry out convolution Feature Mapping;
Step S14), multiple dimensioned pond: square root pond is carried out to the value for the combined joint that convolution Feature Mapping is formed, and is obtained
The low-level image feature of colored subgraph;
Step S15), full connection: the low-level image feature of all obtained colored subgraphs is grouped together, image data is obtained
Low-level image feature;
Step S16), calculate initial output weight;
Step S2, based on the on-line study stage of multiple dimensioned local receptor field, i.e., more using single sample or sample data block
The output weight newly learnt in the initial stage;Specific step is as follows:
Step S21), the D based on multiple dimensioned local receptor fieldp+1The online Analysis On Multi-scale Features of data set map, concrete operation step
Identical as in step S13), wherein S scale channel, the initial weight in each scale channel and biasing are equal to step S13) in
Setting;
Step S22), Dp+1The online multiple dimensioned pond of data set, wherein setting of the pond size equal to parameter in step S14);
Step S23), Dp+1The low-level image feature of all obtained colored subgraphs is grouped together, obtains by the full connection of data set
Obtain the low-level image feature of image data;
Step S24), calculate Dp+1The output weight of data set;
Step S25), setting p=p+1, determine Dp+1Whether data set is the last one online data collection, if so, stopping online
Study, otherwise repeat step S21)-step S24);
Step S26), calculate the output weight of the last one online data collection.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810902594.3A CN109190638A (en) | 2018-08-09 | 2018-08-09 | Classification method based on the online order limit learning machine of multiple dimensioned local receptor field |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810902594.3A CN109190638A (en) | 2018-08-09 | 2018-08-09 | Classification method based on the online order limit learning machine of multiple dimensioned local receptor field |
Publications (1)
Publication Number | Publication Date |
---|---|
CN109190638A true CN109190638A (en) | 2019-01-11 |
Family
ID=64921268
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810902594.3A Pending CN109190638A (en) | 2018-08-09 | 2018-08-09 | Classification method based on the online order limit learning machine of multiple dimensioned local receptor field |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109190638A (en) |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942461A (en) * | 2014-05-14 | 2014-07-23 | 浙江师范大学 | Water quality parameter prediction method based on online sequential extreme learning machine |
CN105069479A (en) * | 2015-08-19 | 2015-11-18 | 西安电子科技大学 | Polarized SAR image classification method based on online sequence limit learning machine |
CN106874961A (en) * | 2017-03-03 | 2017-06-20 | 北京奥开信息科技有限公司 | A kind of indoor scene recognition methods using the very fast learning machine based on local receptor field |
CN107179077A (en) * | 2017-05-15 | 2017-09-19 | 北京航空航天大学 | A kind of self-adaptive visual air navigation aid based on ELM LRF |
CN107463952A (en) * | 2017-07-21 | 2017-12-12 | 清华大学 | A kind of object material sorting technique based on multi-modal fusion deep learning |
CN108038507A (en) * | 2017-12-13 | 2018-05-15 | 太原理工大学 | Local receptor field extreme learning machine image classification method based on particle group optimizing |
-
2018
- 2018-08-09 CN CN201810902594.3A patent/CN109190638A/en active Pending
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103942461A (en) * | 2014-05-14 | 2014-07-23 | 浙江师范大学 | Water quality parameter prediction method based on online sequential extreme learning machine |
CN105069479A (en) * | 2015-08-19 | 2015-11-18 | 西安电子科技大学 | Polarized SAR image classification method based on online sequence limit learning machine |
CN106874961A (en) * | 2017-03-03 | 2017-06-20 | 北京奥开信息科技有限公司 | A kind of indoor scene recognition methods using the very fast learning machine based on local receptor field |
CN107179077A (en) * | 2017-05-15 | 2017-09-19 | 北京航空航天大学 | A kind of self-adaptive visual air navigation aid based on ELM LRF |
CN107463952A (en) * | 2017-07-21 | 2017-12-12 | 清华大学 | A kind of object material sorting technique based on multi-modal fusion deep learning |
CN108038507A (en) * | 2017-12-13 | 2018-05-15 | 太原理工大学 | Local receptor field extreme learning machine image classification method based on particle group optimizing |
Non-Patent Citations (5)
Title |
---|
GUANG-BIN HUANG等: "Extreme learning machines: a survey", 《INT. J. MACH. LEARN. & CYBER. (2011)》 * |
HUAPING LIU等: "Surface Material Recognition Using Active Multi-modal Extreme Learning Machine", 《COGNITIVE COMPUTATION》 * |
JING FANG等: "Local receptive fild based extreme learning machine with three channels for histopathological image classifiation", 《INTERNATIONAL JOURNAL OF MACHINE LEARNING AND CYBERNETICS (2019)》 * |
JINGHONG HUANG等: "Extreme learning machine with multi-scale local receptive fields for texture classification", 《MULTIDIM SYST SIGN PROCESS (2017)》 * |
兰明等: "在线极限学习机在岩爆预测中的应用", 《安全与环境学报》 * |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109543203B (en) | Building cold and heat load prediction method based on random forest | |
WO2019179403A1 (en) | Fraud transaction detection method based on sequence width depth learning | |
CN108009509A (en) | Vehicle target detection method | |
CN103605711B (en) | Construction method and device, classification method and device of support vector machine | |
CN107230113A (en) | A kind of house property appraisal procedure of multi-model fusion | |
CN108388925A (en) | The anti-pattern collapse robust image generation method for generating network is fought based on New Conditions | |
CN108564094A (en) | A kind of Material Identification method based on convolutional neural networks and classifiers combination | |
CN111553114B (en) | Intelligent color matching method for textile printing and dyeing based on data driving | |
CN101276420A (en) | Classification method for syncretizing optical spectrum information and multi-point simulation space information | |
CN111311702B (en) | Image generation and identification module and method based on BlockGAN | |
CN104809476B (en) | A kind of multi-target evolution Fuzzy Rule Classification method based on decomposition | |
KR101680055B1 (en) | Method for developing the artificial neural network model using a conjunctive clustering method and an ensemble modeling technique | |
CN108596274A (en) | Image classification method based on convolutional neural networks | |
CN108108762A (en) | A kind of random forest classification method based on core extreme learning machine and parallelization for the classification of coronary heart disease data | |
CN108492298A (en) | Based on the multispectral image change detecting method for generating confrontation network | |
CN109271427A (en) | A kind of clustering method based on neighbour's density and manifold distance | |
CN108830312A (en) | A kind of integrated learning approach adaptively expanded based on sample | |
CN105095913A (en) | Remote sensing image classification method and system based on neighbor regular joint sparse representation | |
CN111275114A (en) | Network qualification image identification method based on ensemble learning under SDN architecture | |
CN105740917B (en) | The semi-supervised multiple view feature selection approach of remote sensing images with label study | |
CN105138975A (en) | Human body complexion area segmentation method based on deep belief network | |
CN109447118A (en) | A kind of uneven learning method based on Adaboost and lack sampling | |
CN109523514A (en) | To the batch imaging quality assessment method of Inverse Synthetic Aperture Radar ISAR | |
Qin et al. | A wireless sensor network location algorithm based on insufficient fingerprint information | |
CN111737110A (en) | Test input selection method for deep learning model |
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 |
Application publication date: 20190111 |
|
RJ01 | Rejection of invention patent application after publication |