CN108596163A - A kind of Coal-rock identification method based on CNN and VLAD - Google Patents
A kind of Coal-rock identification method based on CNN and VLAD Download PDFInfo
- Publication number
- CN108596163A CN108596163A CN201810748171.0A CN201810748171A CN108596163A CN 108596163 A CN108596163 A CN 108596163A CN 201810748171 A CN201810748171 A CN 201810748171A CN 108596163 A CN108596163 A CN 108596163A
- Authority
- CN
- China
- Prior art keywords
- coal
- image
- vlad
- feature vector
- cluster
- 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
- 238000000034 method Methods 0.000 title claims abstract description 80
- 239000011435 rock Substances 0.000 title claims abstract description 40
- 239000013598 vector Substances 0.000 claims abstract description 67
- 239000003245 coal Substances 0.000 claims abstract description 66
- 238000012549 training Methods 0.000 claims abstract description 32
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 26
- 238000000605 extraction Methods 0.000 claims abstract description 17
- 238000012706 support-vector machine Methods 0.000 claims abstract description 8
- 239000000284 extract Substances 0.000 claims abstract description 5
- 230000008569 process Effects 0.000 claims abstract description 3
- 230000009467 reduction Effects 0.000 claims description 11
- 238000012360 testing method Methods 0.000 claims description 9
- 238000012545 processing Methods 0.000 claims description 6
- 230000002708 enhancing effect Effects 0.000 claims description 4
- 230000002087 whitening effect Effects 0.000 claims description 4
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000000513 principal component analysis Methods 0.000 claims description 3
- 230000008859 change Effects 0.000 claims description 2
- 238000000205 computational method Methods 0.000 claims description 2
- 238000003475 lamination Methods 0.000 claims 1
- 238000010606 normalization Methods 0.000 claims 1
- 238000006116 polymerization reaction Methods 0.000 abstract description 3
- 230000000007 visual effect Effects 0.000 abstract description 3
- 244000287680 Garcinia dulcis Species 0.000 abstract description 2
- 230000004069 differentiation Effects 0.000 abstract 1
- 238000005065 mining Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 230000006870 function Effects 0.000 description 4
- 238000013528 artificial neural network Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 3
- 238000001514 detection method Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000003064 k means clustering Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000002776 aggregation Effects 0.000 description 1
- 238000004220 aggregation Methods 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 230000005251 gamma ray Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 238000004886 process control Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 230000006641 stabilisation Effects 0.000 description 1
- 238000011105 stabilization Methods 0.000 description 1
- 239000002699 waste material Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
-
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
-
- 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
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computational Linguistics (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of Coal-rock identification method based on CNN and VLAD, coal petrography image of this method by that will acquire is input in depth convolutional neural networks AlexNet by pretreatment, extracts image high-level characteristic;The characteristics of image of extraction is clustered using different initial methods, builds visual dictionary;Partial polymerization is carried out to characteristics of image by the method for residual sum between calculating feature, the form that each image is finally expressed as to single column vector is input to support vector machines, the training of device is identified;In identification process, coal petrography characteristics of image to be identified is extracted, inputs trained grader, realizes the differentiation to coal petrography image.This method, which is illuminated by the light factor, to be influenced low, and to the strong robustness of noise, correct recognition rata is high, and stability is good.
Description
Technical field
The present invention relates to a kind of Coal-rock identification methods based on CNN and VLAD, for the coal petrography figure in progress of coal mining
As identification, specifically belong to image processing and pattern recognition field.
Background technology
Coal is one of the mainstay of one of main source of China's national energy and national economy.In recent years, I
The coal production of state all occupy first place in the world, and coal consumption amount is huge.
Coal petrography identification refers to making system automatic discrimination coal and rock by various technological means.Realize the coal of coal working face
Rock automatic identification is to study one of the important content of coal mining, is to unmanned comprehensive if it can effectively solve the problems, such as coal petrography identification
Mining face has stepped a vital step.Meanwhile coal petrography identification technology is also widely used for roller and mines, tunnels, fully mechanized mining
The production links such as Sub-Level Caving process control, bastard coal sorting.During coal mining, the key of coal petrography identification is will be on working face wall
Coal effectively distinguished with rock, a variety of Coal-rock identification methods occurred include radar detection system, gamma-ray detection method,
Infrared detecting method, memory cut method etc., these methods respectively have advantage and disadvantage, but are required for installing big quantity sensor and be identified, and one
The installation of aspect equipment is complicated, waste of manpower and material resources;On the other hand different coal mining environment are directed to, need to choose not simultaneous interpretation
Sensor, and sensor itself also easily breaks down and failure, and system is caused to be difficult to safeguard.
To solve the above problems, research precision is high, the Coal-rock identification method of strong robustness is most important.As computer regards
The continuous enhancing of feel technology and the development of depth learning technology, studying the Coal-rock identification method based on image has weight
Want meaning.Since there are larger differences in color, gloss, texture etc. for coal and rock, by excavating coal petrography digital picture
In visual information carry out coal petrography image classification have practical significance.But traditional coal petrography image-recognizing method computation complexity
Height, accuracy of identification is low, poor robustness, is affected by coal petrography sample image.Based on the Coal-rock identification method of image in stabilization
It need to be improved on property and discrimination.
A kind of coal petrography image-recognizing method efficiently, stable and high discrimination is needed, to solve existing coal petrography image recognition
Intrinsic problem in technology improves coal petrography recognition performance.
Invention content
Therefore, the purpose of the present invention is to provide a kind of coal petrography image-recognizing method based on CNN and VLAD algorithms, the party
Method using depth convolutional neural networks as feature extractor, can phenogram as high-level characteristic, then the feature extracted is led to
It crosses different initial methods to be clustered, obtains visual dictionary, feature is integrated using VLAD algorithms, to be divided
Class.Discrimination is high, and strong robustness, real-time is good, is conducive to improve coal petrography recognition correct rate, ensures downhole safety operation.
Coal-rock identification method of the present invention adopts the following technical scheme that realization, including sample training stage and coal petrography are known
It in the other stage, is as follows:
A. in the training stage of sample, respectively this image of samples of coal pulled and rock sample image several, constitute coal petrography image
Sample set is denoted as D, and wherein coal sample image and rock sample image respectively accounts for half, and image size is uniformly set as 224 × 224, point
Do not take 2/3 coal sample image and 2/3 rock sample image as training sample set, remaining image is as test sample collection;
B. feature extraction is carried out to the coal sample image and rock sample image of sample training concentration, steps are as follows:
(1) sample image in training set is input in the AlexNet depth convolutional neural networks by pre-training, it is right
Each image extracts the characteristic pattern that the 5th layer of 256 size of network are 13 × 13;
(2) to 256 characteristic patterns, in every width figure the characteristic value of same position be ranked sequentially the feature tieed up for one 256 to
Amount constitutes 13 × 13 feature vectors;
(3) cluster operation, the cluster fortune are carried out to the feature vector of 256 dimension with the different initial method of C kinds
In calculation, the quantity for giving cluster is k, wherein C=1,2,3 ...;
(4) each feature vector is distributed to the cluster center nearest apart from it, obtains the different cluster form of C groups, every group
Including k cluster;
(5) residual sum for calculating all feature vectors and cluster center in each cluster, using each residual sum as element, C kinds
Initial method respectively obtains a VLAD feature vector and indicates;
(6) the VLAD feature vectors are indicated to carry out symbol root and normalized, it is further that C groups is different
VLAD features are smooth to turn to a feature vector;
(7) dimensionality reduction and whitening processing are carried out to the feature vector obtained in step (6), obtains the feature vector after dimensionality reduction;
C. the feature vector after dimensionality reduction is inputted into support vector machines, carries out the training of grader;
D. by test set sample coal sample image and rock sample image carry out according to the method for step A and step B it is special
Then the extraction for levying vector inputs the trained grader obtained by step C, test the accuracy of identification of coal petrography image;
E. for coal petrography image to be identified, by image preprocessing, extraction image feature vector (being acquired in step B) inputs
In (being acquired in step C) trained grader, result is exported according to grader and differentiates coal lithotypes.
Further, the AlexNet depth convolutional neural networks by pre-training are in ILSVR2012 data sets
Upper trained depth network model, including 5 convolutional layers and 3 full articulamentums have given up AlexNet nets in the present invention
Three layers of full articulamentum after network, remaining 5 convolutional layers carry out the feature extraction of coal petrography image directly as feature extractor.
Further, in step (3), the initial method different with C kinds carries out cluster operation, initialization side
Method is random initializtion, and the selection at cluster center, which meets, to be uniformly distributed, and operation is clustered using K-means.
Further, in step (4), the set expression of the cluster is Sc,m, i.e.,
Sc,m={ fi,j| m=argminp‖fI, j-μC, p‖}
Indicate the set for each feature vector being assigned to the cluster center nearest apart from it, wherein fI, jIt indicates in characteristic pattern
Corresponding 256 dimensional feature vector of characteristic point of i-th row jth row, 1≤i≤13,1≤j≤13, m expression clusters center, m=1,
2 ... k }, μC, pIt indicates in c kinds initialization clustering method, the corresponding feature vector in cluster center of pth cluster, c={ 1,2 ...
C }, ‖ ‖ are L2Norm, for calculating feature vector fi,jWith cluster center μc,pThe distance between.
Further, in step (5), the computational methods of the residual sum are:
Wherein, vc,mIndicate residual sum;fi,jIndicate characteristic pattern in the i-th row jth arrange corresponding 256 dimensional feature of characteristic point to
Amount;Sc,mIt indicates in c kinds initialization clustering method, the set of m clusters;μc,mIt indicates in c kinds initialization clustering method, m
The corresponding feature vector in cluster center of cluster.
Further, the VLAD feature vectors are expressed as, Fc=[vc,1,vc,2…vc,k], to F in step (6)cIn it is every
One element vc,iThe method for executing symbol root and normalized is:
Further, in step 6, the VLAD features of the C class are smooth to turn to a feature vector
Further, in the step (7), using the method for principal component analysis to feature vectorCarry out dimensionality reduction and white
Change is handled.
Further, it in the step C, selects radial basis function as grader kernel function, it is non-thread to carry out two classification
The training of property support vector machines.
Further, in the step D, by adjusting parameter, ensure that the accuracy of identification of grader reaches 95% or more,
If accuracy of identification is relatively low, by increasing training samples number, carry out data enhancing to sample image, carry out grader into one
Step training.
The present invention extracts feature from the middle layer of neural network, then handles these features, in conjunction with tradition
Clustering method realize coal petrography identification.K-means clustering method effect quality and its initial method have much relations, so
The present invention is influenced using different initial methods to reduce.Then use VLAD method (mainly to the feature after cluster into
Row partial polymerization) ask residual error to acquire the single feature vector expression of each image.This method be illuminated by the light factor influence it is low, to noise
Strong robustness, correct recognition rata is high, and stability is good.
Description of the drawings
Fig. 1 is the basic flow chart of Coal-rock identification method of the present invention.
Fig. 2 is coal petrography image characteristics extraction process schematic of the present invention.
Specific implementation mode
The invention discloses one kind based on CNN (convolutional neural network, convolutional neural networks) and
The coal petrography identification side of VLAD (vector of locally aggregated descriptors, local feature Aggregation Descriptor)
Method, attached drawing 1 are the basic flow charts of the Coal-rock identification method based on CNN and VLAD, and attached drawing 2 is coal petrography identification side of the present invention
The characteristic extraction procedure schematic diagram of method, is specifically described referring to attached drawing.
Referring to Fig.1, coal petrography image-recognizing method overall procedure of the invention is:
A. in the training stage of sample, respectively this image of samples of coal pulled and rock sample image several, constitute coal petrography image
Sample set is denoted as D, wherein coal sample image D1 and rock sample image D2 and respectively accounts for half, D1=D2, and image size is uniformly set as
224 × 224, take 2/3 coal sample image and 2/3 rock sample image as training sample set respectively, remaining image is as test
Sample set carries out category label to all images in sample set D, and category label collection is set as L, is used for the training and survey of grader
Examination;
B. feature extraction is carried out to the coal sample image and rock sample image of sample training concentration;
C. the feature vector after dimensionality reduction is inputted into support vector machines, carries out the training of grader.
The present embodiment selects radial basis function as grader kernel function, carries out the instruction of two classification Nonlinear Support Vector Machines
Practice.Support vector machines is selected to carry out tagsort, it is advantageous that, image feature maps to higher dimensional space can be solved special
The linearly inseparable problem of sign is conducive to classification.
D. by test set sample coal sample image and rock sample image carry out according to the method for step A and step B it is special
Then the extraction for levying vector inputs the trained grader obtained by step C, test the accuracy of identification of coal petrography image.
It can ensure that the accuracy of identification of grader reaches 95% or more by adjusting parameter, if accuracy of identification is relatively low, lead to
It crosses increase training samples number, data enhancing is carried out to sample image, carry out the further training of grader.
E. for coal petrography image to be identified, by image preprocessing, extraction image feature vector (being acquired in step B) inputs
In (being acquired in step C) trained grader, result is exported according to grader and differentiates coal lithotypes.
Wherein, the detailed process of step B progress feature extraction is as follows:
(1) sample image in training set is input in the AlexNet depth convolutional neural networks by pre-training, it is right
Each image extracts the characteristic pattern that the 5th layer of 256 size of network are 13 × 13
In the present embodiment, the AlexNet depth convolutional neural networks by pre-training are instructed on ILSVR2012 data sets
The depth network model perfected, including 5 convolutional layers and 3 full articulamentums, in the present invention, after having given up AlexNet networks
Three layers of full articulamentum, remaining 5 convolutional layers carry out the feature extraction of coal petrography image directly as feature extractor.
(2) to 256 characteristic patterns, the characteristic value of same position is according to the suitable of the 5th layer of convolutional layer output of network in every width figure
Sequence is arranged as the feature vector of one 256 dimension, constitutes 13 × 13 feature vectors;
(3) cluster operation, the cluster fortune are carried out to the feature vector of 256 dimension with the different initial method of C kinds
In calculation, the quantity for giving cluster is k, wherein C=1,2,3 ....
The initial methods different with C kinds carry out cluster operation, and initial method is random initializtion, and
The selection obedience of feature vector as cluster center is uniformly distributed, while clustering operation using K-means.
(4) each feature vector is distributed to the cluster center nearest apart from it, obtains the different cluster form of C groups, every group
Including k cluster.
The set expression of cluster is Sc,m, i.e.,
SC, m={ fi,j| m=arg minp‖fi,j-μc,p‖}
Indicate the set for each feature vector being assigned to the cluster center nearest apart from it, wherein fi,jIt indicates in characteristic pattern
Corresponding 256 dimensional feature vector of characteristic point of i-th row jth row, fi,j∈R256, 1≤i≤13,1≤j≤13, m expression clusters center,
M={ 1,2 ... k }, μC, pIt indicates in c kinds initialization clustering method, the corresponding feature vector in cluster center of pth cluster, c=1,
2 ... C }, ‖ ‖ are L2Norm, for calculating feature vector fi,jWith cluster center μc,pThe distance between.
Feature vector fi,jIt is expressed as fi,j=(x1,x2,…,x169), μc,pIt is expressed as μc,p=(y1,y2,…,y169), feature
Vector fi,jWith cluster center μc,pThe distance between calculation formula be:
(5) all feature vector f in each cluster are calculatedI, jWith the residual sum at cluster center, i.e.,:
Then each different initial method respectively obtains a VLAD feature vector and indicates, is denoted as Fc:
Fc=[vc,1,vc,2…vc,k]
Wherein, vc,mIndicate residual sum;fi,jIndicate characteristic pattern in the i-th row jth arrange corresponding 256 dimensional feature of characteristic point to
Amount;Sc,mIt indicates in c kinds initialization clustering method, the set of m clusters;μc,mIt indicates in c kinds initialization clustering method, m
The corresponding feature vector in cluster center of cluster.
(6) F is indicated to VLAD feature vectorsc(c={ 1,2 ... C }) carries out symbol root and normalized, i.e., to Fc
In each element vc,iExecute operation:
Further a vector is turned to by the VLAD features of C class are smooth
So far, each image is represented as a vectorForm;
(7) it uses the method for principal component analysis to the feature vector obtained in step (6), carries out dimensionality reduction and whitening processing,
Obtain the feature vector after dimensionality reduction
General features extraction is with some traditional algorithms, or the training image Jing Guo neural network, directly by god
Classification results are obtained through network.The present invention is to extract feature from the middle layer of neural network, then to these features at
Reason realizes coal petrography identification in conjunction with traditional clustering method.K-means clustering method effect quality and its initial method have
Much relations, so the present invention is influenced using different initial methods to reduce.Then use the method for VLAD (mainly to poly-
Feature after class carries out partial polymerization) ask residual error to acquire the single feature vector expression of each image.
Claims (10)
1. a kind of Coal-rock identification method based on CNN and VLAD, which is characterized in that include the following steps:
A. in the training stage of sample, respectively this image of samples of coal pulled and rock sample image several, wherein coal sample image and
Rock sample image respectively accounts for half, takes 2/3 coal sample image and 2/3 rock sample image as training sample set, remaining figure respectively
As being used as test sample collection;
B. feature extraction is carried out to the coal sample image and rock sample image of sample training concentration, steps are as follows:
(1) sample image in training set is input in the AlexNet depth convolutional neural networks by pre-training, to every width
Sample image extracts the characteristic pattern that the 5th layer of 256 size of network are 13 × 13;
(2) to 256 characteristic patterns, the characteristic value of same position is ranked sequentially the feature tieed up for one 256 in every width sample image
Vector constitutes 13 × 13 feature vectors;
(3) cluster operation is carried out to the feature vector of 256 dimension with the different initial method of C kinds, in the cluster operation,
The quantity of given cluster is k, wherein C=1,2,3 ...;
(4) each feature vector is distributed to the cluster center nearest apart from it, obtains the different cluster form of C groups, every group includes k
A cluster;
(5) residual sum of all feature vectors and cluster center in each cluster is calculated, using each residual sum as element, C kinds are initial
Change method respectively obtains a VLAD feature vector and indicates;
(6) the VLAD feature vectors are indicated to carry out symbol root and normalized, further by the different VLAD of C groups
Feature is smooth to turn to a feature vector;
(7) dimensionality reduction and whitening processing are carried out to the feature vector obtained in step (6), obtains the feature vector after dimensionality reduction;
C. the feature vector after dimensionality reduction is inputted into support vector machines, carries out the training of grader;
D. by test set sample coal sample image and rock sample image according to the method for step A and step B carry out feature to
Then the extraction of amount inputs the trained grader obtained by step C, test the accuracy of identification of coal petrography image;
E. for coal petrography image to be identified, by image preprocessing, extraction inputs root according to the image feature vector that step B is obtained
In the trained grader obtained according to step C, result is exported according to grader and differentiates coal lithotypes.
2. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that the process is pre-
Trained AlexNet depth convolutional neural networks are the trained depth network models on ILSVR2012 data sets, including 5
A convolutional layer and 3 full articulamentums have given up three layers of full articulamentum after AlexNet networks, remaining 5 volumes in the present invention
Lamination carries out the feature extraction of coal petrography image directly as feature extractor.
3. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that in step (3), C kinds
Different initial methods is random initializtion, while clustering operation using K-means.
4. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that in step (4), cluster
Set expression be SC, m, i.e.,
SC, m={ fI, j| m=argminp‖fI, j-μC, p‖}
Indicate the set for each feature vector being assigned to the cluster center nearest apart from it, wherein fi,jIt indicates i-th in characteristic pattern
Corresponding 256 dimensional feature vector of characteristic point of row jth row, 1≤i≤13,1≤j≤13, m indicate cluster center, m={ 1,2 ...
K }, μc,pIt indicates in c kinds initialization clustering method, the corresponding feature vector in cluster center of pth cluster, c={ 1,2 ... C }, ‖ ‖
For L2Norm, for calculating feature vector fi,jWith cluster center μc,pThe distance between.
5. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that residual in step (5)
The computational methods of poor sum are:
Wherein, vC, mIndicate residual sum;fi,jIndicate corresponding 256 dimensional feature vector of characteristic point that the i-th row jth arranges in characteristic pattern;
Sc,mIt indicates in c kinds initialization clustering method, the set of m clusters;μc,mIt indicates in c kinds initialization clustering method, m clusters
The corresponding feature vector in cluster center.
6. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that the VLAD features
Vector is expressed as, Fc=[vc,1,vc,2…vc,k], to F in step (6)cIn each element vc,iExecute symbol root and normalization
The method of processing is:
7. the Coal-rock identification method according to claim 6 based on CNN and VLAD, which is characterized in that the VLAD of C class is special
It levies and smooth turns to a feature vector
8. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that in step (7), make
With the method for principal component analysis to feature vectorCarry out dimensionality reduction and whitening processing.
9. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that in step C, selection
Radial basis function carries out the training of two classification Nonlinear Support Vector Machines as grader kernel function.
10. the Coal-rock identification method according to claim 1 based on CNN and VLAD, which is characterized in that in step D, pass through
Adjusting parameter ensures that the accuracy of identification of grader reaches 95% or more, if accuracy of identification is relatively low, by increasing number of training
Amount carries out data enhancing to sample image, carries out the further training of grader.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810748171.0A CN108596163A (en) | 2018-07-10 | 2018-07-10 | A kind of Coal-rock identification method based on CNN and VLAD |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810748171.0A CN108596163A (en) | 2018-07-10 | 2018-07-10 | A kind of Coal-rock identification method based on CNN and VLAD |
Publications (1)
Publication Number | Publication Date |
---|---|
CN108596163A true CN108596163A (en) | 2018-09-28 |
Family
ID=63617808
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810748171.0A Pending CN108596163A (en) | 2018-07-10 | 2018-07-10 | A kind of Coal-rock identification method based on CNN and VLAD |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN108596163A (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615024A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | A kind of Rock Species intelligence Division identification and localization method |
CN110135460A (en) * | 2019-04-16 | 2019-08-16 | 广东工业大学 | Image information intensifying method based on VLAD convolution module |
CN110209859A (en) * | 2019-05-10 | 2019-09-06 | 腾讯科技(深圳)有限公司 | The method and apparatus and electronic equipment of place identification and its model training |
CN110245702A (en) * | 2019-06-12 | 2019-09-17 | 深圳大学 | Mechanical wear particle recognition method, apparatus, electronic equipment and storage medium |
CN111126488A (en) * | 2019-12-24 | 2020-05-08 | 威创集团股份有限公司 | Image identification method based on double attention |
CN111738348A (en) * | 2020-06-29 | 2020-10-02 | 南方电网科学研究院有限责任公司 | Power data anomaly detection method and device |
CN111968136A (en) * | 2020-08-18 | 2020-11-20 | 华院数据技术(上海)有限公司 | Coal rock microscopic image analysis method and analysis system |
CN112348100A (en) * | 2020-11-13 | 2021-02-09 | 成都理工大学 | Rock recognition method, device, equipment and storage medium |
Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103473748A (en) * | 2013-09-23 | 2013-12-25 | 中国矿业大学(北京) | Method for enhancing underground coal mine image |
CN103530621A (en) * | 2013-11-04 | 2014-01-22 | 中国矿业大学(北京) | Coal and rock image identification method based on back propagation (BP) neural network |
CN104318254A (en) * | 2014-11-13 | 2015-01-28 | 中国矿业大学(北京) | Quick coal and rock recognition method based on DCT low-frequency component characteristics |
CN105243401A (en) * | 2015-11-10 | 2016-01-13 | 中国矿业大学(北京) | Coal rock recognition method based on coal structure element study |
CN106649665A (en) * | 2016-12-14 | 2017-05-10 | 大连理工大学 | Object-level depth feature aggregation method for image retrieval |
CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
CN106845329A (en) * | 2016-11-11 | 2017-06-13 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of action identification method based on depth convolution feature multichannel pyramid pond |
WO2017096570A1 (en) * | 2015-12-10 | 2017-06-15 | Intel Corporation | Visual recognition using deep learning attributes |
CN107122712A (en) * | 2017-03-27 | 2017-09-01 | 大连大学 | It polymerize the palmprint image recognition methods of description vectors based on convolutional neural networks and two-way local feature |
CN107145903A (en) * | 2017-04-28 | 2017-09-08 | 武汉理工大学 | A kind of Ship Types recognition methods extracted based on convolutional neural networks picture feature |
CN107169415A (en) * | 2017-04-13 | 2017-09-15 | 西安电子科技大学 | Human motion recognition method based on convolutional neural networks feature coding |
CN108154183A (en) * | 2017-12-25 | 2018-06-12 | 深圳市唯特视科技有限公司 | A kind of objective classification method based on part and depth characteristic set |
-
2018
- 2018-07-10 CN CN201810748171.0A patent/CN108596163A/en active Pending
Patent Citations (12)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103473748A (en) * | 2013-09-23 | 2013-12-25 | 中国矿业大学(北京) | Method for enhancing underground coal mine image |
CN103530621A (en) * | 2013-11-04 | 2014-01-22 | 中国矿业大学(北京) | Coal and rock image identification method based on back propagation (BP) neural network |
CN104318254A (en) * | 2014-11-13 | 2015-01-28 | 中国矿业大学(北京) | Quick coal and rock recognition method based on DCT low-frequency component characteristics |
CN105243401A (en) * | 2015-11-10 | 2016-01-13 | 中国矿业大学(北京) | Coal rock recognition method based on coal structure element study |
WO2017096570A1 (en) * | 2015-12-10 | 2017-06-15 | Intel Corporation | Visual recognition using deep learning attributes |
CN106845329A (en) * | 2016-11-11 | 2017-06-13 | 广东顺德中山大学卡内基梅隆大学国际联合研究院 | A kind of action identification method based on depth convolution feature multichannel pyramid pond |
CN106649665A (en) * | 2016-12-14 | 2017-05-10 | 大连理工大学 | Object-level depth feature aggregation method for image retrieval |
CN106682233A (en) * | 2017-01-16 | 2017-05-17 | 华侨大学 | Method for Hash image retrieval based on deep learning and local feature fusion |
CN107122712A (en) * | 2017-03-27 | 2017-09-01 | 大连大学 | It polymerize the palmprint image recognition methods of description vectors based on convolutional neural networks and two-way local feature |
CN107169415A (en) * | 2017-04-13 | 2017-09-15 | 西安电子科技大学 | Human motion recognition method based on convolutional neural networks feature coding |
CN107145903A (en) * | 2017-04-28 | 2017-09-08 | 武汉理工大学 | A kind of Ship Types recognition methods extracted based on convolutional neural networks picture feature |
CN108154183A (en) * | 2017-12-25 | 2018-06-12 | 深圳市唯特视科技有限公司 | A kind of objective classification method based on part and depth characteristic set |
Non-Patent Citations (5)
Title |
---|
JOE YUE-HEI NG等: "Exploiting Local Features from Deep Networks for Image Retrieval" * |
ZHE WANG等: "CODEBOOK ENHANCEMENT OF VLAD REPRESENTATION FOR VISUAL RECOGNITION" * |
卞颖: "基于深度学习的图像检索技术研究" * |
王倩等: "基于改进VLAD算法的图像分类" * |
田子建等: "基于机器视觉的煤岩界面识别研究" * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109615024A (en) * | 2018-12-28 | 2019-04-12 | 东北大学 | A kind of Rock Species intelligence Division identification and localization method |
CN110135460A (en) * | 2019-04-16 | 2019-08-16 | 广东工业大学 | Image information intensifying method based on VLAD convolution module |
CN110209859A (en) * | 2019-05-10 | 2019-09-06 | 腾讯科技(深圳)有限公司 | The method and apparatus and electronic equipment of place identification and its model training |
CN110209859B (en) * | 2019-05-10 | 2022-12-27 | 腾讯科技(深圳)有限公司 | Method and device for recognizing places and training models of places and electronic equipment |
CN110245702A (en) * | 2019-06-12 | 2019-09-17 | 深圳大学 | Mechanical wear particle recognition method, apparatus, electronic equipment and storage medium |
CN111126488A (en) * | 2019-12-24 | 2020-05-08 | 威创集团股份有限公司 | Image identification method based on double attention |
CN111126488B (en) * | 2019-12-24 | 2023-08-18 | 威创集团股份有限公司 | Dual-attention-based image recognition method |
CN111738348A (en) * | 2020-06-29 | 2020-10-02 | 南方电网科学研究院有限责任公司 | Power data anomaly detection method and device |
CN111968136A (en) * | 2020-08-18 | 2020-11-20 | 华院数据技术(上海)有限公司 | Coal rock microscopic image analysis method and analysis system |
CN112348100A (en) * | 2020-11-13 | 2021-02-09 | 成都理工大学 | Rock recognition method, device, equipment and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108596163A (en) | A kind of Coal-rock identification method based on CNN and VLAD | |
Tian et al. | Detecting text in natural image with connectionist text proposal network | |
CN109002834B (en) | Fine-grained image classification method based on multi-modal representation | |
CN106897738B (en) | A kind of pedestrian detection method based on semi-supervised learning | |
CN108171184A (en) | Method for distinguishing is known based on Siamese networks again for pedestrian | |
CN111275688A (en) | Small target detection method based on context feature fusion screening of attention mechanism | |
CN108960073A (en) | Cross-module state image steganalysis method towards Biomedical literature | |
CN106650780A (en) | Data processing method, device, classifier training method and system | |
CN103488415B (en) | Personal letter word based on paper medium and pattern recognition system and recognition methods | |
CN108491077A (en) | A kind of surface electromyogram signal gesture identification method for convolutional neural networks of being divided and ruled based on multithread | |
CN108427740B (en) | Image emotion classification and retrieval algorithm based on depth metric learning | |
Zhang et al. | ImageSem at ImageCLEF 2018 caption task: Image retrieval and transfer learning | |
Liu et al. | Subtler mixed attention network on fine-grained image classification | |
CN110413791A (en) | File classification method based on CNN-SVM-KNN built-up pattern | |
CN109086794B (en) | Driving behavior pattern recognition method based on T-LDA topic model | |
CN104156690B (en) | A kind of gesture identification method based on image space pyramid feature bag | |
CN105938565A (en) | Multi-layer classifier and Internet image aided training-based color image emotion classification method | |
CN110348494A (en) | A kind of human motion recognition method based on binary channels residual error neural network | |
CN101551855A (en) | Auxiliary diagnostic system for tracing self-adaptive kernel matching and auxiliary diagnostic method thereof | |
CN109242047A (en) | Bank card number detection and recognition methods based on K-means++ cluster and residual error network class | |
CN110008828A (en) | Pairs of constraint ingredient assay measures optimization method based on difference regularization | |
CN109325513A (en) | A kind of image classification network training method based on magnanimity list class single image | |
CN110059765A (en) | A kind of mineral intelligent recognition categorizing system and method | |
CN108664986A (en) | Based on lpThe multi-task learning image classification method and system of norm regularization | |
CN110287985A (en) | A kind of deep neural network image-recognizing method based on the primary topology with Mutation Particle Swarm Optimizer |
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 |
Application publication date: 20180928 |
|
WD01 | Invention patent application deemed withdrawn after publication |