CN106815601A - Hyperspectral image classification method based on recurrent neural network - Google Patents
Hyperspectral image classification method based on recurrent neural network Download PDFInfo
- Publication number
- CN106815601A CN106815601A CN201710014713.7A CN201710014713A CN106815601A CN 106815601 A CN106815601 A CN 106815601A CN 201710014713 A CN201710014713 A CN 201710014713A CN 106815601 A CN106815601 A CN 106815601A
- Authority
- CN
- China
- Prior art keywords
- sample
- feature
- spectrum image
- high spectrum
- recurrent neural
- 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.)
- Granted
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
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2136—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on sparsity criteria, e.g. with an overcomplete basis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/28—Determining representative reference patterns, e.g. by averaging or distorting; Generating dictionaries
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- General Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of hyperspectral image classification method based on recurrent neural network, mainly solution existing method input feature vector identification is weaker, and local spatial feature extracts insufficient problem, and implementation step includes:1. the spatial texture feature and rarefaction representation feature of high spectrum image are extracted, and to its stacked combination into low-level feature;2. sample local space sequence signature is extracted on low-level feature;3. recurrent neural networks model is built according to local space sequence signature, and using training sample local space sequence signature training recurrent neural networks model parameter;4. the recurrent neural networks model for the input of test sample local space sequence signature being trained, obtains the high-level semantics features of high abstraction, obtains the classification information of test sample.The present invention improves the accuracy of classification hyperspectral imagery using the method for deep learning, can be used for vegetation investigation, and disaster surveillance, cartography and information are obtained.
Description
Technical field
The invention belongs to technical field of image processing, it is related to a kind of hyperspectral image classification method, can be used for EO-1 hyperion distant
Feel the classification of image.
Background technology
At present, as the spectral resolution of remote sensor is improved constantly, people are to object spectrum attribute, the cognition of feature
Also deepen continuously, the characters of ground object that many is hidden in the range of narrow spectrum is gradually found, has greatly accelerated remote sensing technology
Development, high-spectrum remote-sensing is turned into one of most important research direction of 21 century remote sensing fields.
Different from multispectral remote sensing, spectral remote sensing technology using imaging spectrometer with nano level spectral resolution, with several
Ten or hundreds of wave bands earth's surface thing is imaged simultaneously, be obtained in that the continuous spectrum information of atural object, the spy with " collection of illustrative plates "
Property, all played an important role at aspects such as national economy, national defense construction, have been widely used for landmarks categorizations, target and visit
The fields such as survey, agricultural monitoring, mineral map plotting, environmental management and national defense construction.
The classification of high spectrum image is an important content of target in hyperspectral remotely sensed image treatment and application, and its final goal is
Unique classification logotype is assigned to each pixel in image.For the research and analysis of hyperspectral image data classification task,
Main is exactly that the characters of ground object in image is told by high-spectral data, i.e., believed by analyzing original spectrum or other features
Breath, different earth surface areas is marked off to come, such as meadow, farmland, waters, cities and towns, bridge, contributes to people to recognize and divide
Analysis earth's surface situation.
High in face of hyperspectral image data dimension, data dependence is strong, and redundancy is high, the features such as local space uniformity,
Conventional sorting technique is mainly started with from the following aspects:(1) rarefaction representation and dictionary learning, select a small amount of marked sample
As dictionary, gone to represent other samples with the linear combination of dictionary sample;(2) supporting vector machine SVM classifier and its kernel function
Correlating transforms, build new kernel function, such as polynomial kernel, adapt to the nonlinear Distribution of high-spectral data;(3) it is semi-supervised with
Active Learning, the classifying quality under lifting small sample;(4) proposition of new grader and the combination of multi-categorizer;(5) feature is carried
Take with combined transformation etc..Wherein conventional feature includes:1) itself spectral information characteristics of spectral signature, i.e. high spectrum image and its
Related derived character;2) space characteristics, including high spectrum image texture, shape, morphological feature;3) sky-spectrum signature, i.e., it is empty
Between feature and spectral signature be combined the feature for obtaining.
These features are started with from the inherent characteristic of high spectrum image, and the low layer spy that conversion is obtained is extracted by simple
Levy.With the burning hot rise of deep learning in recent years, how using deep learning model framework, fully extracting has more preferably expression
Property high spectrum image high-level characteristic, lifted classification hyperspectral imagery precision, increasingly become domestic and foreign scholars fall over each other research
Focus.
Deep learning is a kind of Feature Extraction Technology, and it is developed from nerual network technique, layering by way of pair
Low-level feature carries out high abstraction, so as to obtain the more preferable method for expressing of feature.Conventional deep learning framework mainly has stack
Self-encoding encoder SAE, depth confidence network DBN, convolutional neural networks CNN, recurrent neural network RNN, they are widely used in
The fields such as natural language processing, computer vision, speech recognition, bioinformatics, and achieve extraordinary effect.
Scholar is had at present by stack self-encoding encoder SAE, depth confidence network DBN and convolutional neural networks CNN depth
Practise model and be incorporated into classification hyperspectral imagery.《Deep Learning-Based Classification of
Hyperspectral Data》In, high-spectral data is carried out PCA dimensionality reductions by Yushi Chen, chooses the pixel in rectangular window
Connect into and think characteristic vector as local spatial feature, then coupled together with original spectrum feature, be input into as low-level feature
Build stack self-encoding encoder SAE models.《Spectral-Spatial Classification of Hyperspectral
Data Based on Deep Belief Network》In, it is special that Yushi Chen build low layer using method same as above
Levy, and be input into the depth confidence network DBN model for building.The classifying quality of this method is general, and accuracy is not high, and
In the presence of many deficiencies, such as directly using spectral signature as input feature vector, comprising random noise too much, identification is weaker, nothing
Method obtains good classifying quality;And in local spatial feature extraction, simply pick all pixels of neighborhood and be not added with
Treatment, the then pixel for wherein being differed greatly with center pixel can badly influence nicety of grading.
The content of the invention
It is an object of the invention to be directed to above-mentioned the deficiencies in the prior art, a kind of bloom based on recurrent neural network is proposed
Spectrum image classification method, to build the degree of purity more more preferable low-level feature of high-class effect, while strengthening to picture in local space
The exploration of correlation between element, improves the effect of important pixel, reduces the influence of useless pixel, and low-level feature is abstract to sentence
Other property high-level semantics features higher, so as to more fully using the characteristic of high spectrum image, improve nicety of grading.
To achieve the above object, technical scheme includes as follows:
(1) be input into a panel height spectrum picture, the high spectrum image include K pixel, B EO-1 hyperion spectral coverage, c class atural objects,
Wherein K=K1×K2, K1Represent the length of high spectrum image, K2The width of high spectrum image is represented, each pixel of image is one
Sample, each sample represents that the intrinsic dimensionality of sample is B with a characteristic vector, and 10% sample is selected in every class atural object
This composition training sample set, is left 90% sample composition test sample collection;
(2) the principal component gray-scale map of high spectrum image is filtered using Gabor filter, obtains high spectrum image
Spatial texture featureWherein R represents real number field, and g is spatial texture feature vector dimension;
(3) rarefaction representation coefficient of each pixel in high spectrum image is calculated using the method for rarefaction representation, bloom is obtained
The rarefaction representation feature of spectrogram pictureWherein m is the dimension of rarefaction representation characteristic vector;
(4) by the spatial texture feature F of high spectrum image1With rarefaction representation feature F2Stacked combination is into high spectrum image
Low-level featureL is the dimension of low-level feature vector, l=g+m;
(5) in high spectrum image low-level feature matrix F, window is built centered on each sample, extracts the office of sample
Portion's space characteristics block, and the local space sequence signature of sample is built using the similitude between sample;
(6) recurrent neural networks model is built by time step long number of number of samples in window, and is input into training sample
Local space sequence signature and corresponding class label repetitive exercise recurrent neural networks model parameter, the recurrence for being trained
Neural network model;
(7) the local space sequence signature of test sample is input in the recurrent neural networks model for training, is obtained
Classification category, completes classification.
The invention has the advantages that:
(1) present invention gets up the spatial texture feature and rarefaction representation feature integration of high spectrum image, and this low layer is special
Levy both comprising high spectrum image pixel samples local spatial information, and comprising rarefaction representation of the pixel samples on other samples
Information, such low-level feature degree of purity and identification are higher, more preferable for classification task effect;
(2) present invention extracts the local space sequence of high spectrum image on the basis of high spectrum image local spatial feature
Row feature, not only obtains local spatial information, also explores the similarity information between each pixel samples in local space, carries
The effect of important pixel high, reduces the influence of useless pixel, improves classifying quality;
(3) present invention utilizes the recurrent neural networks model for being usually used in natural language processing field, by recurrent neural net
The temporal characteristicses of network combine with high spectrum image local space sequence information, can effectively integrate high spectrum image local empty
Between context relation, be extracted as high-level semantics features by low-level feature is abstract, take full advantage of the characteristic of EO-1 hyperion, improve
Classification accuracy rate.
Brief description of the drawings
Fig. 1 is schematic flow sheet of the invention;
Fig. 2 is recurrent neural networks model schematic diagram;
Fig. 3 is the Indian Pines images that present invention emulation is used;
Fig. 4 is the classification results comparison diagram of the present invention and existing method to Indian Pines images.
Specific embodiment
Reference picture 1, specific implementation step of the invention includes:
Step 1, is input into high spectrum image.
It is input into a width three-dimensional matrice high spectrum image, the high spectrum image includes K pixel samples, B EO-1 hyperion spectral coverage,
C class atural objects, wherein K=K1×K2, K1Represent the length of high spectrum image, K2The width of high spectrum image is represented, is selected in every class atural object
10% sample is selected as training sample, is left 90% sample as test sample.
Step 2, obtains the spatial texture feature F of high spectrum image1。
2a) enter line translation to high spectrum image using PCA, k=10 principal component gray-scale map before extracting;
43 kinds of direction Gabor filters of yardstick 2b) are set, that is, 4 different Gabor kernel functions directions and 3 are set
Individual different sinusoidal plane wave wavelength, obtains 12 Gabor filters, and the kernel function of each Gabor filter is as follows:
Wherein, x'=xcos θ+ysin θ, y'=-xcos θ+ysin θ, x and y denotation coordination positional informations, λ represents sinusoidal
The wavelength of plane wave, θ represents the direction of Gabor kernel functions,Phase deviation is represented, σ represents the standard deviation of Gaussian envelope, γ tables
Show space aspect ratio;
2c) using 12 Gabor filters for setting respectively to each principal component ash in preceding k principal component gray-scale map
Degree figure carries out Gabor filtering, obtains 12 filtered images of each principal component gray-scale map;
2d) by 12 × k filtered image stack together, the spatial texture feature of high spectrum image is obtainedWherein R represents real number field, and g=12 × k is spatial texture characteristic vector length.
Step 3, obtains the rarefaction representation feature F of high spectrum image2。
3a) 1% sample is randomly selected from the c class training samples of high spectrum image per class build son as dictionary atom
Dictionary, wherein the sub- dictionary of the i-th class isI=1,2 ..., c,Represent the i-th class
J-th dictionary atom, j=1,2 ..., mi, miRepresent the dictionary atom number of the i-th class;
3b) c sub- dictionary is lined up, an entirety is merged into, total structuring dictionary D=[D are obtained1 ... Di
... Dc], D ∈ RB×mIt is a two-dimensional matrix, represents total structuring dictionary that the sub- dictionary of all classes is constituted, m=m1+…+
mi+…+mcRepresent the sub- dictionary atom number sum of all categories;
The rarefaction representation vector of each pixel 3c) is solved using orthogonal matching pursuit algorithm, i.e., by orthogonal matching pursuit
Algorithm optimization equation below, obtains rarefaction representation vector αs of each pixel x on structuring dictionary D in high spectrum image:
S.t.D α=x
Wherein | | α | |0Expression takes 0 norm to α;
3d) by the rarefaction representation vector α of all samples in high spectrum image, constituted according to raw image data correspondence position
One rarefaction representation eigenmatrix of three-dimensional
Step 4, by spatial texture featureWith rarefaction representation featureStacked combination is got up, and is obtained
To the low-level feature matrix of high spectrum imageWherein l=g+m is low-level feature vector length.
Step 5, builds the local space sequence signature of sample.
5a) in high spectrum image low-level feature matrix F, centered on each sample x, size is built by the length of side of w=9
It is the window of w × w, extracts the local spatial feature block of x, then the local spatial feature block of x includes w2=81 samples, wherein often
Individual sample is a low-level feature vector of a length of l;
Each sample is big with the similarity of central sample x in 5b) calculating local spatial feature block according to Euclidean distance formula
Small, the smaller similarity of Euclidean distance is bigger, and each sample is lined up from big to small according to similarity, obtains sample x's
Local space sequence signature
Step 6, using training sample set and its correspondence class label training recurrent neural networks model.
It is 6a) time step long number T with number of samples in window, builds input layer and hidden node number is the recurrence of l
Neural network model, wherein T=w × w=81;
The local space sequence signature and corresponding class label of training sample 6b) are input to recurrent neural network mould
Type, each low-level feature vector that will be in sample local space sequence signature is input into each corresponding time step, tool
Body method is by t-th low-level feature x in the local space sequence signature of any one training sample xtInput recurrent neural net
T-th time step of network, by the input x of the time steptWith the hidden layer state s of step-length at previous i.e. the t-1t-1Altogether
With the hidden layer state s for constituting the time stept:
st=σ (Uxt+Wst-1),
Wherein U represents input layer to the weight matrix of hidden layer, and W represents hidden layer to the weight matrix of hidden layer, and σ represents non-linear
Activation primitive, σ of the invention selects ReLU functions, then last output o81By the hidden layer state s of last time step81
Determine, i.e.,:
Wherein V represents hidden layer to the weight matrix of output layer,Nonlinear activation function is represented, this exampleSelection
Softmax functions, then using the parameter in the back-propagation method repetitive exercise recurrent neural networks model by the time, repeatedly
Generation 200 times after stop, the recurrent neural networks model for being trained.
Step 7, the local space sequence signature of test sample is input in the recurrent neural networks model for training, and is obtained
To classification category, classification is completed.
Effect of the invention can be further illustrated by following emulation experiment:
1. simulated conditions:
Emulation experiment uses Indian Pines images, the image to be pushed away by NASA's NASA jets in June, 1992
Unloaded visible ray/Infrared Imaging Spectrometer the AVIRIS for entering laboratory is obtained in the Indiana northwestward, as shown in figure 3, image
Size is 145 × 145, totally 220 wave bands, and the wave band that removal noise and air and waters absorb also has 200 wave bands, totally 16
Class terrestrial object information, as shown in table 1.
Emulation experiment is Intel Core i5-4210, dominant frequency 2.90GHz in CPU, and the Windows 10 for inside saving as 8G is
Carried out with Python on system.
16 class data in table 1Indian Pines images
Classification | Item name | Number of samples | Classification | Item name | Number of samples |
1 | Alfalfa | 46 | 9 | Oats | 20 |
2 | Corn-notill | 1428 | 10 | Soybean-notill | 972 |
3 | Corn-mintill | 830 | 11 | Soybean-mintill | 2455 |
4 | Corn | 237 | 12 | Soybean-clean | 593 |
5 | Grass-pasture | 483 | 13 | Wheat | 205 |
6 | Grass-trees | 730 | 14 | Woods | 1265 |
7 | Grass-pasture-mowed | 28 | 15 | Buildings-Grass-Trees-Drives | 386 |
8 | Hay-windrowed | 478 | 16 | Stone-Steal-Towers | 93 |
2. simulation parameter:
10% as training sample, remaining 90% used as test sample for the unification selection of above emulation experiment, SVM punishment because
Son is set to 491;In SRC methods, dictionary directly is constituted with training sample, degree of rarefication is set to 10;In SOMP methods, directly use
Training sample constitutes dictionary, and window size w is set to 9, and degree of rarefication is set to 30;In the present invention, PCA conversion retains preceding 10 masters
Composition, frequency selection { 0.25,0.5,0.75 } 3 kinds of yardsticks of Gabor filter, set direction4 kinds of directions
Totally 12 wave filters, rarefaction representation builds dictionary per class sample random selection 1%, and degree of rarefication is set to 30, and window size w is set
It is 9, timing node number T is set to 81.
3. emulation content and result:
High spectrum image Indian Pine are classified with existing three kinds of common methods using the present invention, it is conventional
Three kinds of methods are respectively:Based on supporting vector machine svm classifier method, based on the sorting technique of rarefaction representation SRC, based on partial zones
The rarefaction representation SOMP sorting techniques in domain.
Indian Pines images are classified with above-mentioned three kinds of common methods with the present invention, as a result as shown in figure 4, its
Middle Fig. 4 a are the classification results figures for using SVM methods, and Fig. 4 b are the classification results figures for using rarefaction representation SRC methods, and Fig. 4 c are to use
The classification results figure of SOMP methods, Fig. 4 d are classification results figures of the invention.Figure 4, it is seen that compared to conventional three
The method of kind, result figure of the invention becomes apparent from totally, and local space uniformity and edge uniformity are all better than existing method,
Nicety of grading is higher.
The present invention respectively carries out 10 emulation experiments with other various methods, and the average value for taking classification results divides as final
Class accuracy, including overall accuracy (OA), per class average accuracy (AA) and Kappa coefficient (Kappa), as a result such as the institute of table 2
Show.
The classification accuracy rate of the present invention of table 2 and other method
Method | OA (%) | AA (%) | Kappa |
SVM | 81.24 | 74.06 | 0.79 |
SRC | 68.53 | 64.23 | 0.64 |
SOMP | 95.27 | 83.48 | 0.95 |
The present invention | 97.17 | 95.84 | 0.97 |
As seen from Table 2, SOMP methods of the invention and existing are due to containing local spatial information, compared to only with single picture
The SVM methods and SRC methods of prime information, classification accuracy rate is substantially higher, and the present invention had both incorporated spatial texture feature
With the information of rarefaction representation feature, and the information between local space pixel is fully excavated, and will with recurrent neural networks model
It is the identification more preferable high-level semantics features of representative higher that low-level feature is extracted, and can obtain classification accuracy rate higher,
All well other three kinds of methods on overall accuracy, average accuracy, Kappa coefficient.
To sum up, the spatial texture feature and rarefaction representation feature integration of high spectrum image are low-level feature by the present invention, and
Local space sequence signature is extracted on the basis of local spatial feature, and using the recurrent neural net in deep learning framework
Network model is classified to high spectrum image, has both improve the degree of purity and identification of low-level feature, and high-spectrum is explored again
Similarity information in the local space of picture between each pixel samples, improves the effect of important pixel, reduces useless pixel
Influence, while the temporal characteristicses of recurrent neural network are combined with high spectrum image local space sequence information, can be with
The context relation of high spectrum image local space is effectively integrated, high-level semantics features is extracted as by low-level feature is abstract, fully
The characteristic of EO-1 hyperion is make use of, discrimination higher is obtained, compared with the existing methods with obvious advantage.
Claims (5)
1. a kind of hyperspectral image classification method based on recurrent neural network, including:
(1) a panel height spectrum picture is input into, the high spectrum image includes K pixel, B EO-1 hyperion spectral coverage, c class atural objects, wherein K
=K1×K2, K1Represent the length of high spectrum image, K2The width of high spectrum image is represented, each pixel of image is a sample,
Each sample represents that the intrinsic dimensionality of sample is B with a characteristic vector, and 10% sample composition is selected in every class atural object
Training sample set, is left 90% sample composition test sample collection;
(2) the principal component gray-scale map of high spectrum image is filtered using Gabor filter, obtains the space of high spectrum image
Textural characteristicsWherein R represents real number field, and g is spatial texture feature vector dimension;
(3) rarefaction representation coefficient of each pixel in high spectrum image is calculated using the method for rarefaction representation, high-spectrum is obtained
The rarefaction representation feature of pictureWherein m is the dimension of rarefaction representation characteristic vector;
(4) by the spatial texture feature F of high spectrum image1With rarefaction representation feature F2Low layer of the stacked combination into high spectrum image
FeatureL is the dimension of low-level feature vector, l=g+m;
(5) in high spectrum image low-level feature matrix F, window is built centered on each sample, extracts the local empty of sample
Between characteristic block, and the local space sequence signature of sample is built using the similitude between sample;
(6) recurrent neural networks model is built by time step long number of number of samples in window, and is input into the part of training sample
Spatial sequence feature and corresponding class label repetitive exercise recurrent neural networks model parameter, the recurrent neural for being trained
Network model;
(7) the local space sequence signature of test sample is input in the recurrent neural networks model for training, is classified
Category, completes classification.
2. method according to claim 1, using Gabor filter to the principal component of high spectrum image wherein in step 2
Gray-scale map is filtered, and carries out as follows:
2a) enter line translation to high spectrum image using PCA, k=10 principal component gray-scale map before extracting;
4 different Gabor kernel functions directions sinusoidal plane wave wavelength different with 3 2b) is set, 12 Gabor filters are obtained
Ripple device, the kernel function of each Gabor filter is as follows:
Wherein, x'=xcos θ+ysin θ, y'=-xcos θ+ysin θ, x and y denotation coordination positional informations, λ represents sinusoidal plane
The wavelength of ripple, θ represents the direction of Gabor kernel functions,Phase deviation is represented, σ represents the standard deviation of Gaussian envelope, and γ represents empty
Between aspect ratio;
2c) each the principal component gray-scale map in k principal component gray-scale map is entered respectively using 12 Gabor filters for setting
Row Gabor is filtered, and obtains 12 filtered images of each principal component gray-scale map;
2d) by 12 × k filtered image stack together, the spatial texture feature of high spectrum image is obtainedG=12 × k representation space texture feature vector length.
3. method according to claim 1, calculates every in high spectrum image wherein in step 3 using the method for rarefaction representation
The rarefaction representation coefficient of individual pixel, is carried out as follows:
3a) 1% sample is randomly selected from the c class training samples of high spectrum image per class build sub- word as dictionary atom
Allusion quotation, wherein the sub- dictionary of the i-th class isI=1,2 ..., c,Represent the i-th class
J-th dictionary atom, j=1,2 ..., mi, miRepresent the dictionary atom number of the i-th class;
3b) c sub- dictionary is lined up, an entirety is merged into, total structuring dictionary D=[D are obtained1 ... Di ...
Dc], D ∈ RB×mIt is a two-dimensional matrix, represents total structuring dictionary that the sub- dictionary of all classes is constituted, m=m1+…+mi+…
+mcRepresent the sub- dictionary atom number sum of all categories;
The rarefaction representation vector of each pixel 3c) is solved using orthogonal matching pursuit algorithm, i.e., by orthogonal matching pursuit algorithm
Optimization equation below, obtains rarefaction representation vector αs of each pixel x on structuring dictionary D in high spectrum image:
S.t. D α=x
Wherein | | α | |0Expression takes 0 norm to α;
3d) by the rarefaction representation vector α of all samples in high spectrum image, one is constituted according to raw image data correspondence position
Three-dimensional rarefaction representation eigenmatrix
4. method according to claim 1, builds the local space sequence signature of sample wherein in step 5, by following step
Suddenly carry out:
5a) in high spectrum image low-level feature matrix F, centered on any one sample x, built by the window length of side of w=9
Size is w × w rectangular windows, extract the three-dimensional matrice that the local spatial feature block of x, i.e. size are w × w × l, then x
Local spatial feature block includes w2=81 pixel samples, wherein each sample are a low-level feature vector of a length of l, l=g
+ m, g representation space texture feature vector length, m represent rarefaction representation characteristic vector length;
5b) calculate the similarity size of each pixel samples and center pixel sample x in local spatial feature block, and by each picture
Element is lined up from big to small according to similarity, obtains the local space sequence signature of pixel samples x
5. method according to claim 1, builds recurrent neural networks model and training pattern parameter wherein in step 6,
Carry out as follows:
It is that T, input layer and hidden node number are the recurrent neural networks model of l 6a) to build time step, wherein T=w ×
W=81, w represent the window length of side, and l=g+m, g representation space texture feature vector length, m represent that rarefaction representation characteristic vector is long
Degree;
The local space sequence signature of training sample 6b) is input to recurrent neural networks model, will sample local space sequence
Each low-level feature vector in row feature is input into each corresponding time step, and using the backpropagation by the time
Parameter in method repetitive exercise recurrent neural networks model, the recurrent neural networks model for being trained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710014713.7A CN106815601B (en) | 2017-01-10 | 2017-01-10 | Hyperspectral image classification method based on recurrent neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710014713.7A CN106815601B (en) | 2017-01-10 | 2017-01-10 | Hyperspectral image classification method based on recurrent neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106815601A true CN106815601A (en) | 2017-06-09 |
CN106815601B CN106815601B (en) | 2019-10-11 |
Family
ID=59110109
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710014713.7A Active CN106815601B (en) | 2017-01-10 | 2017-01-10 | Hyperspectral image classification method based on recurrent neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106815601B (en) |
Cited By (30)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107169535A (en) * | 2017-07-06 | 2017-09-15 | 谈宜勇 | The deep learning sorting technique and device of biological multispectral image |
CN107194437A (en) * | 2017-06-22 | 2017-09-22 | 重庆大学 | Image classification method based on Gist feature extractions Yu conceptual machine recurrent neural network |
CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技(深圳)有限公司 | Image classification method, device and computer-readable recording medium |
CN107798348A (en) * | 2017-10-27 | 2018-03-13 | 广东省智能制造研究所 | Hyperspectral image classification method based on neighborhood information deep learning |
CN107844751A (en) * | 2017-10-19 | 2018-03-27 | 陕西师范大学 | The sorting technique of guiding filtering length Memory Neural Networks high-spectrum remote sensing |
CN108171270A (en) * | 2018-01-05 | 2018-06-15 | 大连海事大学 | A kind of hyperspectral image classification method based on Hash study |
CN108256454A (en) * | 2018-01-08 | 2018-07-06 | 浙江大华技术股份有限公司 | A kind of training method based on CNN models, human face posture estimating and measuring method and device |
CN108460342A (en) * | 2018-02-05 | 2018-08-28 | 西安电子科技大学 | Hyperspectral image classification method based on convolution net and Recognition with Recurrent Neural Network |
CN108764303A (en) * | 2018-05-10 | 2018-11-06 | 电子科技大学 | A kind of remote sensing images spatial term method based on attention mechanism |
CN109002771A (en) * | 2018-06-26 | 2018-12-14 | 中国科学院遥感与数字地球研究所 | A kind of Classifying Method in Remote Sensing Image based on recurrent neural network |
CN109460471A (en) * | 2018-11-01 | 2019-03-12 | 信融源大数据科技(北京)有限公司 | A method of the mode based on self study establishes kinds of fibers spectrum library |
CN109615008A (en) * | 2018-12-11 | 2019-04-12 | 华中师范大学 | Hyperspectral image classification method and system based on stack width learning |
CN109670042A (en) * | 2018-12-04 | 2019-04-23 | 广东宜教通教育有限公司 | A kind of examination question classification and grade of difficulty method based on recurrent neural network |
CN109711466A (en) * | 2018-12-26 | 2019-05-03 | 陕西师范大学 | A kind of CNN hyperspectral image classification method retaining filtering based on edge |
CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技(深圳)有限公司 | Image classification method, device and computer readable storage medium |
CN109816002A (en) * | 2019-01-11 | 2019-05-28 | 广东工业大学 | The single sparse self-encoding encoder detection method of small target migrated certainly based on feature |
CN109978041A (en) * | 2019-03-19 | 2019-07-05 | 上海理工大学 | A kind of hyperspectral image classification method based on alternately update convolutional neural networks |
CN110163293A (en) * | 2019-05-28 | 2019-08-23 | 武汉轻工大学 | Red meat classification method, device, equipment and storage medium based on deep learning |
CN110188794A (en) * | 2019-04-23 | 2019-08-30 | 深圳大学 | A kind of training method, device, equipment and the storage medium of deep learning model |
CN110363078A (en) * | 2019-06-05 | 2019-10-22 | 汕头大学 | A kind of hyperspectral image classification method and device based on ADMM-Net |
CN110852451A (en) * | 2019-11-27 | 2020-02-28 | 电子科技大学 | Recursive kernel self-adaptive filtering method based on kernel function |
CN110866439A (en) * | 2019-09-25 | 2020-03-06 | 南京航空航天大学 | Hyperspectral image joint classification method based on multi-feature learning and superpixel kernel sparse representation |
US10643092B2 (en) | 2018-06-21 | 2020-05-05 | International Business Machines Corporation | Segmenting irregular shapes in images using deep region growing with an image pyramid |
CN111582330A (en) * | 2020-04-22 | 2020-08-25 | 北方民族大学 | Integrated ResNet-NRC method for dividing sample space based on lung tumor image |
US10776923B2 (en) | 2018-06-21 | 2020-09-15 | International Business Machines Corporation | Segmenting irregular shapes in images using deep region growing |
CN111860654A (en) * | 2020-07-22 | 2020-10-30 | 河南大学 | Hyperspectral image classification method based on recurrent neural network |
CN113128669A (en) * | 2021-04-08 | 2021-07-16 | 中国科学院计算技术研究所 | Neural network model for semi-supervised learning and semi-supervised learning method |
CN113139532A (en) * | 2021-06-22 | 2021-07-20 | 中国地质大学(武汉) | Classification method based on multi-output classification model, computer equipment and medium |
CN113887656A (en) * | 2021-10-21 | 2022-01-04 | 江南大学 | Hyperspectral image classification method combining deep learning and sparse representation |
CN117649943A (en) * | 2024-01-30 | 2024-03-05 | 吉林大学 | Shaping data intelligent analysis system and method based on machine learning |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103514456A (en) * | 2013-06-30 | 2014-01-15 | 安科智慧城市技术(中国)有限公司 | Image classification method and device based on compressed sensing multi-core learning |
CN104036289A (en) * | 2014-06-05 | 2014-09-10 | 哈尔滨工程大学 | Hyperspectral image classification method based on spatial and spectral features and sparse representation |
CN104091151A (en) * | 2014-06-30 | 2014-10-08 | 南京信息工程大学 | Vehicle identification method based on Gabor feature extraction and sparse representation |
CN104298999A (en) * | 2014-09-30 | 2015-01-21 | 西安电子科技大学 | Hyperspectral feature leaning method based on recursion automatic coding |
US9152881B2 (en) * | 2012-09-13 | 2015-10-06 | Los Alamos National Security, Llc | Image fusion using sparse overcomplete feature dictionaries |
WO2016075081A1 (en) * | 2014-11-10 | 2016-05-19 | Lego A/S | System and method for toy recognition |
-
2017
- 2017-01-10 CN CN201710014713.7A patent/CN106815601B/en active Active
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9152881B2 (en) * | 2012-09-13 | 2015-10-06 | Los Alamos National Security, Llc | Image fusion using sparse overcomplete feature dictionaries |
CN103514456A (en) * | 2013-06-30 | 2014-01-15 | 安科智慧城市技术(中国)有限公司 | Image classification method and device based on compressed sensing multi-core learning |
CN104036289A (en) * | 2014-06-05 | 2014-09-10 | 哈尔滨工程大学 | Hyperspectral image classification method based on spatial and spectral features and sparse representation |
CN104091151A (en) * | 2014-06-30 | 2014-10-08 | 南京信息工程大学 | Vehicle identification method based on Gabor feature extraction and sparse representation |
CN104298999A (en) * | 2014-09-30 | 2015-01-21 | 西安电子科技大学 | Hyperspectral feature leaning method based on recursion automatic coding |
WO2016075081A1 (en) * | 2014-11-10 | 2016-05-19 | Lego A/S | System and method for toy recognition |
Non-Patent Citations (2)
Title |
---|
MENGQI JI ETAL.: "Preprocessing-free surface material classification using convolutional neural networks pretrained by sparse Autoencoder", 《2015 IEEE 25TH INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING (MLSP)》 * |
朱可: "基于稀疏表示的人脸表情识别", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (50)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107194437A (en) * | 2017-06-22 | 2017-09-22 | 重庆大学 | Image classification method based on Gist feature extractions Yu conceptual machine recurrent neural network |
CN107194437B (en) * | 2017-06-22 | 2020-04-07 | 重庆大学 | Image classification method based on Gist feature extraction and concept machine recurrent neural network |
CN107169535B (en) * | 2017-07-06 | 2023-11-03 | 谈宜勇 | Deep learning classification method and device for biological multispectral image |
CN107169535A (en) * | 2017-07-06 | 2017-09-15 | 谈宜勇 | The deep learning sorting technique and device of biological multispectral image |
CN107844751A (en) * | 2017-10-19 | 2018-03-27 | 陕西师范大学 | The sorting technique of guiding filtering length Memory Neural Networks high-spectrum remote sensing |
CN107844751B (en) * | 2017-10-19 | 2021-08-27 | 陕西师范大学 | Method for classifying hyperspectral remote sensing images of guide filtering long and short memory neural network |
CN107798348A (en) * | 2017-10-27 | 2018-03-13 | 广东省智能制造研究所 | Hyperspectral image classification method based on neighborhood information deep learning |
CN107798348B (en) * | 2017-10-27 | 2020-02-18 | 广东省智能制造研究所 | Hyperspectral image classification method based on neighborhood information deep learning |
CN107679525A (en) * | 2017-11-01 | 2018-02-09 | 腾讯科技(深圳)有限公司 | Image classification method, device and computer-readable recording medium |
CN107679525B (en) * | 2017-11-01 | 2022-11-29 | 腾讯科技(深圳)有限公司 | Image classification method and device and computer readable storage medium |
CN109753978B (en) * | 2017-11-01 | 2023-02-17 | 腾讯科技(深圳)有限公司 | Image classification method, device and computer readable storage medium |
CN109753978A (en) * | 2017-11-01 | 2019-05-14 | 腾讯科技(深圳)有限公司 | Image classification method, device and computer readable storage medium |
CN108171270A (en) * | 2018-01-05 | 2018-06-15 | 大连海事大学 | A kind of hyperspectral image classification method based on Hash study |
CN108171270B (en) * | 2018-01-05 | 2021-08-27 | 大连海事大学 | Hyperspectral image classification method based on Hash learning |
CN108256454B (en) * | 2018-01-08 | 2020-08-14 | 浙江大华技术股份有限公司 | Training method based on CNN model, and face posture estimation method and device |
CN108256454A (en) * | 2018-01-08 | 2018-07-06 | 浙江大华技术股份有限公司 | A kind of training method based on CNN models, human face posture estimating and measuring method and device |
CN108460342A (en) * | 2018-02-05 | 2018-08-28 | 西安电子科技大学 | Hyperspectral image classification method based on convolution net and Recognition with Recurrent Neural Network |
CN108460342B (en) * | 2018-02-05 | 2021-01-01 | 西安电子科技大学 | Hyperspectral image classification method based on convolutional neural network and cyclic neural network |
CN108764303A (en) * | 2018-05-10 | 2018-11-06 | 电子科技大学 | A kind of remote sensing images spatial term method based on attention mechanism |
US10643092B2 (en) | 2018-06-21 | 2020-05-05 | International Business Machines Corporation | Segmenting irregular shapes in images using deep region growing with an image pyramid |
US10776923B2 (en) | 2018-06-21 | 2020-09-15 | International Business Machines Corporation | Segmenting irregular shapes in images using deep region growing |
CN109002771A (en) * | 2018-06-26 | 2018-12-14 | 中国科学院遥感与数字地球研究所 | A kind of Classifying Method in Remote Sensing Image based on recurrent neural network |
CN109002771B (en) * | 2018-06-26 | 2022-04-08 | 中国科学院遥感与数字地球研究所 | Remote sensing image classification method based on recurrent neural network |
CN109460471B (en) * | 2018-11-01 | 2021-09-24 | 信融源大数据科技(北京)有限公司 | Method for establishing fiber category map library based on self-learning mode |
CN109460471A (en) * | 2018-11-01 | 2019-03-12 | 信融源大数据科技(北京)有限公司 | A method of the mode based on self study establishes kinds of fibers spectrum library |
CN109670042A (en) * | 2018-12-04 | 2019-04-23 | 广东宜教通教育有限公司 | A kind of examination question classification and grade of difficulty method based on recurrent neural network |
CN109615008A (en) * | 2018-12-11 | 2019-04-12 | 华中师范大学 | Hyperspectral image classification method and system based on stack width learning |
CN109615008B (en) * | 2018-12-11 | 2022-05-13 | 华中师范大学 | Hyperspectral image classification method and system based on stack width learning |
CN109711466A (en) * | 2018-12-26 | 2019-05-03 | 陕西师范大学 | A kind of CNN hyperspectral image classification method retaining filtering based on edge |
CN109816002A (en) * | 2019-01-11 | 2019-05-28 | 广东工业大学 | The single sparse self-encoding encoder detection method of small target migrated certainly based on feature |
CN109816002B (en) * | 2019-01-11 | 2022-09-06 | 广东工业大学 | Single sparse self-encoder weak and small target detection method based on feature self-migration |
CN109978041A (en) * | 2019-03-19 | 2019-07-05 | 上海理工大学 | A kind of hyperspectral image classification method based on alternately update convolutional neural networks |
CN109978041B (en) * | 2019-03-19 | 2022-11-29 | 上海理工大学 | Hyperspectral image classification method based on alternative updating convolutional neural network |
CN110188794B (en) * | 2019-04-23 | 2023-02-28 | 深圳大学 | Deep learning model training method, device, equipment and storage medium |
CN110188794A (en) * | 2019-04-23 | 2019-08-30 | 深圳大学 | A kind of training method, device, equipment and the storage medium of deep learning model |
CN110163293A (en) * | 2019-05-28 | 2019-08-23 | 武汉轻工大学 | Red meat classification method, device, equipment and storage medium based on deep learning |
CN110363078A (en) * | 2019-06-05 | 2019-10-22 | 汕头大学 | A kind of hyperspectral image classification method and device based on ADMM-Net |
CN110866439A (en) * | 2019-09-25 | 2020-03-06 | 南京航空航天大学 | Hyperspectral image joint classification method based on multi-feature learning and superpixel kernel sparse representation |
CN110852451A (en) * | 2019-11-27 | 2020-02-28 | 电子科技大学 | Recursive kernel self-adaptive filtering method based on kernel function |
CN110852451B (en) * | 2019-11-27 | 2022-03-01 | 电子科技大学 | Recursive kernel self-adaptive filtering method based on kernel function |
CN111582330A (en) * | 2020-04-22 | 2020-08-25 | 北方民族大学 | Integrated ResNet-NRC method for dividing sample space based on lung tumor image |
CN111860654A (en) * | 2020-07-22 | 2020-10-30 | 河南大学 | Hyperspectral image classification method based on recurrent neural network |
CN111860654B (en) * | 2020-07-22 | 2024-02-02 | 河南大学 | Hyperspectral image classification method based on cyclic neural network |
CN113128669A (en) * | 2021-04-08 | 2021-07-16 | 中国科学院计算技术研究所 | Neural network model for semi-supervised learning and semi-supervised learning method |
CN113139532B (en) * | 2021-06-22 | 2021-09-21 | 中国地质大学(武汉) | Classification method based on multi-output classification model, computer equipment and medium |
CN113139532A (en) * | 2021-06-22 | 2021-07-20 | 中国地质大学(武汉) | Classification method based on multi-output classification model, computer equipment and medium |
CN113887656A (en) * | 2021-10-21 | 2022-01-04 | 江南大学 | Hyperspectral image classification method combining deep learning and sparse representation |
CN113887656B (en) * | 2021-10-21 | 2024-04-05 | 江南大学 | Hyperspectral image classification method combining deep learning and sparse representation |
CN117649943A (en) * | 2024-01-30 | 2024-03-05 | 吉林大学 | Shaping data intelligent analysis system and method based on machine learning |
CN117649943B (en) * | 2024-01-30 | 2024-04-30 | 吉林大学 | Shaping data intelligent analysis system and method based on machine learning |
Also Published As
Publication number | Publication date |
---|---|
CN106815601B (en) | 2019-10-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106815601B (en) | Hyperspectral image classification method based on recurrent neural network | |
CN110348319B (en) | Face anti-counterfeiting method based on face depth information and edge image fusion | |
CN105809198B (en) | SAR image target recognition method based on depth confidence network | |
CN103971123B (en) | Hyperspectral image classification method based on linear regression Fisher discrimination dictionary learning (LRFDDL) | |
CN104392463B (en) | Image salient region detection method based on joint sparse multi-scale fusion | |
CN108564109A (en) | A kind of Remote Sensing Target detection method based on deep learning | |
CN109389080A (en) | Hyperspectral image classification method based on semi-supervised WGAN-GP | |
CN109766858A (en) | Three-dimensional convolution neural network hyperspectral image classification method combined with bilateral filtering | |
CN106682569A (en) | Fast traffic signboard recognition method based on convolution neural network | |
CN106295124A (en) | Utilize the method that multiple image detecting technique comprehensively analyzes gene polyadenylation signal figure likelihood probability amount | |
CN113627472B (en) | Intelligent garden leaf feeding pest identification method based on layered deep learning model | |
CN109978882A (en) | A kind of medical imaging object detection method based on multi-modal fusion | |
CN102208034A (en) | Semi-supervised dimension reduction-based hyper-spectral image classification method | |
CN107657271A (en) | Hyperspectral image classification method based on long memory network in short-term | |
CN103093478B (en) | Based on the allos image thick edges detection method of quick nuclear space fuzzy clustering | |
CN105405132A (en) | SAR image man-made target detection method based on visual contrast and information entropy | |
CN112950780B (en) | Intelligent network map generation method and system based on remote sensing image | |
CN109034213B (en) | Hyperspectral image classification method and system based on correlation entropy principle | |
Hussein et al. | A comprehensive evaluation framework and a comparative study for human detectors | |
CN108596195A (en) | A kind of scene recognition method based on sparse coding feature extraction | |
CN107704840A (en) | A kind of remote sensing images Approach for road detection based on deep learning | |
Tombe et al. | Adaptive deep co-occurrence feature learning based on classifier-fusion for remote sensing scene classification | |
CN106886754A (en) | Object identification method and system under a kind of three-dimensional scenic based on tri patch | |
CN107203779A (en) | The EO-1 hyperion dimension reduction method kept based on empty spectrum information | |
CN111881965B (en) | Hyperspectral pattern classification and identification method, device and equipment for medicinal material production place grade |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |