CN105069468A - Hyper-spectral image classification method based on ridgelet and depth convolution network - Google Patents

Hyper-spectral image classification method based on ridgelet and depth convolution network Download PDF

Info

Publication number
CN105069468A
CN105069468A CN201510451326.0A CN201510451326A CN105069468A CN 105069468 A CN105069468 A CN 105069468A CN 201510451326 A CN201510451326 A CN 201510451326A CN 105069468 A CN105069468 A CN 105069468A
Authority
CN
China
Prior art keywords
image
training sample
network
layer
spectral
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
Application number
CN201510451326.0A
Other languages
Chinese (zh)
Other versions
CN105069468B (en
Inventor
刘芳
石程
郝红侠
焦李成
李玲玲
尚荣华
马文萍
杨淑媛
马晶晶
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Xidian University
Original Assignee
Xidian University
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Xidian University filed Critical Xidian University
Priority to CN201510451326.0A priority Critical patent/CN105069468B/en
Publication of CN105069468A publication Critical patent/CN105069468A/en
Application granted granted Critical
Publication of CN105069468B publication Critical patent/CN105069468B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Processing (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a hyper-spectral image classification method based on ridgelet and a depth convolution network, and mainly aims to solve the problem that the accuracy is low and the computational complexity is high for hyper-spectral image classification through the existing technology. The method comprises the following steps: (1) selecting training samples in a hyper-spectral image; (2) extracting spectral information and spatial information of the training samples; (3) forming training sample sets based on the spectral information and spatial information; (4) constructing a five-layer depth convolution network, and designing a ridgelet filter to initialize the network; (5) using the training sample sets to train a constructed neural network; and (6) using the trained neural network to classify the other training samples, thus completing image classification. The method has the advantages of high classification accuracy and high classification speed, and can be used in weather monitoring, environmental monitoring, urban planning, and disaster prevention and mitigation.

Description

Based on the hyperspectral image classification method of ridge ripple and degree of depth convolutional network
Technical field
The invention belongs to technical field of image processing, particularly a kind of hyperspectral image classification method, can be used for weather monitoring, environmental monitoring, Land_use change, city planning and prevent and reduce natural disasters.
Background technology
High spectral resolution remote sensing refers to and utilizes a lot of very narrow electromagnetic wave bands to obtain relevant data from interested object.Its maximum feature is exactly while acquisition target image two-dimensional space scene information, the spectral information that high-resolution one dimension characterizes its physical attribute can also be obtained, namely having the characteristic of " collection of illustrative plates unification ", is one of new technique representing remote sensing last word.The key distinction of high-spectrum remote-sensing and conventional remotely-sensed data is that high-spectrum remote-sensing is narrow wave band imaging, can obtain continuous print spectral information, detects not detectable material in conventional remote sensing.Therefore, when a broadband system roughly can only distinguish different material kind, EO-1 hyperion sensor but for the detailed qualification of material and can more accurately estimate that its degree of enriching provides potential possibility.
Main contents of hyperspectral data processing are exactly ground object target classification.Classification is a kind of analytical technology describing ground object target or kind, its main task gives a category label to produce a kind of process of thematic maps to each pixel of data volume, and it is that people are from one of remote sensing image important channel of extracting useful information.The thematic maps produced after classification clearly can reflect the space distribution of atural object, is convenient to people and is therefrom familiar with and finds its rule, makes high-spectrum remote sensing have real use value and effectively put in practical application.
Traditional image classification method is visual interpretation, and visual technology make use of the spatial model that the outstanding thinking ability of the mankind is come in qualitative evaluation image.There is certain shortcoming in this method, it needs image visual interpretation person to have abundant Geo knowledge and visual interpretation experience, and labour intensity is large, requires a great deal of time.Spectral characteristic is not can evaluate by the method for visual interpretation comprehensively in addition.In order to improve quality and the efficiency of classification, from the seventies in last century, people start to pay attention to the technique study by thematic information in computing machine automatic acquisition remote sensing images.Traditional statistical pattern recognition method was mainly utilized to carry out remote sensing computer interpretation at that time, nicety of grading can not be satisfactory, along with the continuous development and change of remote sensing image, also constantly new requirement is being proposed to sorting algorithm, therefore improve existing sorting algorithm, find new method is one of focus in remote sensing application research always.
The feature extraction of high spectrum image is a pith of classification hyperspectral imagery, has a great impact nicety of grading.At present, the high spectrum image feature extracting method that market uses mainly contains the feature extracting method based on spectral information, based on the feature extracting method of spatial information, and the feature extracting method of combining space information and spectral information.
Based in the feature extracting method of spectral information, each pixel in high-spectral data shows as a spectrum response curve in spectral space.Different atural object has different wave spectrum reflectivity and absorption characteristic; Identical atural object also has different reflectance spectrum rates at different wave bands, shows as different radiation intensity in remotely-sensed data.The spectral profile form of therefore different atural object is different; The spectral profile of same atural object is not smooth but fluctuations yet, usually has multiple peak dot and valley point.Sorting technique based on spectral information is exactly utilize the spectral profile of different atural object to classify to atural object, conventional feature extraction algorithm has sparse PCA (PrincipalComponentsAnalysis), ICA (IndependentComponentAnalysis) and LDA (LatentDirichletAllocation) etc.But object spectrum response can be subject to the impact of several factors, as solar illumination, atmospheric transparency and wind speed etc., and these factors are all very difficult Measurement accuracy usually, so the object spectrum response curve that actual measurement obtains may have very large difference with actual curve.In this case, spectral space describing mode well can not adapt to the analysis of high-spectral data, so the nicety of grading usually obtained based on the feature extracting method of spectral information is not ideal enough.
Feature extracting method based on spatial information only utilizes the spatial information of high spectrum image to classify, and typical method is as the feature extracting method of the feature extracting method based on variance, the feature extracting method based on gray level co-occurrence matrixes and wavelet analysis.These class methods are feature extracting methods of a kind of artificial experience, need the feature knowing image in advance, more corresponding selection appropriate method, so these class methods need good priori just can reach good classifying quality.
For this reason, Many researchers proposes the feature extracting method in conjunction with space and spectral information, nicety of grading is improved, as IFRF (ImageFusionandRecursiveFiltering), EPF (Edge-PreservingFilters) and NMFL (NonlinearandLinearMultipleFeatureLearning) method by means of the spectrum of high spectrum image and spatial information.Although these class methods overcome the wrong point problem problem of the atural object only using spectral information or spatial information to cause to a certain extent, still need more priori can obtain good nicety of grading.
Neural network is the method that a class effectively extracts empty spectrum signature, is also a kind of method of feature learning of active, does not need there is priori to image, and typical neural network is as BP neural network, wavelet neural network and ridge ripple neural network.But these are all the neural networks of shallow-layer, all only comprise 3 layers, in order to better excavate the feature of image more deep layer, the model of deep neural network is suggested, and typical deep neural network has own coding degree of depth network, limited Boltzmann machine degree of depth network, degree of depth convolutional network etc.Because degree of depth convolutional network is a real two-dimentional neural network, for the image of two dimension, degree of depth convolutional network better can represent the feature of image.But the initialization of traditional degree of depth convolutional network wave filter is all adopt random initializtion, or Gaussian function initialization, good initialization for network performance and approach speed and have a great impact, and these traditional initial methods are difficult to reach a desirable effect.
Summary of the invention
The object of the invention is to the deficiency for above-mentioned prior art, a kind of hyperspectral image classification method based on ridge ripple and degree of depth convolutional network is proposed, in classification hyperspectral imagery problem, study is difficult to effective characteristic of division to solve prior art, and traditional degree of depth convolutional network is difficult to reach the problem of less computation complexity, put forward precision and the speed of classification hyperspectral imagery.
For achieving the above object, performing step of the present invention is as follows:
1) input the class mark of terrestrial object information in a panel height spectrum picture and this image, from this high spectrum image, select the pixel of 10% as training sample;
2) spectral information of training sample is extracted: along the dimension of high spectrum image spectrum, extract the spectral information of each training sample, composition spectral vector f j, j=1 ..., J, J are the numbers of training sample;
3) dimension-reduction treatment is carried out to high spectrum image, retain front 4 principal components, the image after composition dimensionality reduction;
4) spatial information of training sample is extracted: centered by each training sample, in the image after dimensionality reduction on every one dimension, choose the window of 7 × 7 sizes, obtain the spatial information of this sample in this dimension
5) by the spatial information of each training sample with spectral vector f jform a foursquare training sample image block, and this image is normalized, obtain the training sample square image blocks F after normalization j;
6) degree of depth convolutional network of 5 layers is constructed, and with the training sample square image blocks F after normalization jas the input of this convolutional network, this network is trained, obtain the network trained;
7) using remaining the pixel of 90% in high spectrum image as test sample book, the spectral vector f of each sample is extracted q, and space vector n i, i=1 ..., 4, form a foursquare test sample image block, and this image block is normalized, obtain the test sample book square image blocks F after normalization q, q=1 ..., Q, Q are the numbers of test sample book;
8) by the test sample book square image blocks F after normalization qbe input to step 6) in the network that trains, according to the parameter value trained in network, obtain the class scale value of this sample, complete classification.
The present invention compared with prior art, has following effect:
A () present invention uses the hyperspectral image classification method of spatial information and spectral information combination, overcome in traditional hyperspectral image classification method and only adopt spectral information, have ignored the problem effectively utilizing spatial information, improve nicety of grading.
B () the present invention is under the framework of conventional depth convolutional neural networks, ridge wave function is adopted to the initialization of convolutional layer wave filter in network, overcome conventional filter initial method to be difficult to approach the space of high spectrum image and the problem of spectral information effectively and rapidly, improve the speed of classification.
Accompanying drawing explanation
Fig. 1 is the hyperspectral image classification method FB(flow block) based on ridge ripple and degree of depth convolutional network of the present invention;
Fig. 2 is that the present invention tests use image and true atural object classification chart thereof;
Fig. 3 uses the present invention and prior art to the classification results comparison diagram of Fig. 2;
Fig. 4 uses the present invention and existing method to the error decline comparison diagram of Fig. 2.
Embodiment
Referring to accompanying drawing, technical scheme of the present invention and effect are described in further detail.
With reference to Fig. 1, performing step of the present invention is as follows:
Step 1, input picture.
Input a panel height spectrum picture, as shown in the figure, wherein 2 (a) is the high spectrum image of input, and Fig. 2 (b) is the class logo image that 2 (a) is corresponding, selects the pixel of 10% as training sample from 2 (a).
Step 2, extracts the spectral information of training sample.
If the Spectral dimension of the high spectrum image of input is V in step 1, to each training sample, extract the spectral value of the every one dimension of this sample, composition spectral vector f j, j=1 ..., J, J are the numbers of training sample, spectral vector f jdimension be V.
Step 3, to high-spectrum image dimensionality reduction.
The method of image being carried out to dimensionality reduction has the methods such as sparse PCA, LDA, PCA, ICA, and the present invention uses the high spectrum image of PCA method to input in step 1 to carry out dimension-reduction treatment, and concrete steps are as follows:
3a) obtain the covariance matrix of the high spectrum image of input in step 1;
The component map of the high spectrum image of input in step 1 3b) is obtained according to covariance matrix;
3c) finally retain the maximum component map of front 4 energy values, by the image after these 4 component map composition dimensionality reductions.
Step 4, extracts the spatial information of training sample.
In each component map in image in step 2 after dimensionality reduction, centered by each training sample, choose the window of 7 × 7 sizes, obtain the spatial information of this sample in component map
Step 5, combines spatial information with spectral information.
5a) by the image block of the representative image spatial information of 47 × 7 sizes be combined into the spatial information square image blocks that a size is 14*14 G = g 1 g 2 g 3 g 4 ;
5b) by spectral vector f jbeing rearranged into a size is H*l 2spectral information rectangle image block F, H is the length of rectangle image block, l 2wide for rectangle image block, H*l 2=V, wherein V is the dimension of spectral vector;
5c) in spatial information square image blocks G, random selecting size is (H-14) * (H-l 2) image block B, according to this image block B and spatial information square image blocks G and spectral information rectangle image block F, building size is the training sample square image blocks of H*H: A = G F B ;
5d) the training sample square image blocks A obtained is normalized, obtains normalized training sample square image blocks F j, j=1 ..., J, J are the numbers of training sample.
Step 6, constructs 5 layer depth convolutional network and trains it.
6a) construct 5 layer depth convolutional network: wherein the 1st layer is input layer, and layers 2 and 3 is convolutional layer, and the 4th layer is full linking layer, and the 5th layer is softmax sorter; 1st layer be input as normalized training sample square image blocks F j; 2nd layer comprises L 1individual wave filter; 3rd layer comprises L 2individual wave filter; 4th layer comprises 100 node unit; This layer 5 exports the class mark of training sample.
6b) train 5 layer depth convolutional network, its step is as follows:
6b1) use the wave filter of ridge wave function initialization two convolutional layers, that is:
First, discretize is carried out to scale parameter a, the displacement parameter b of continuous ridge wave function and direction parameter θ: set the span of scale parameter a as a ∈ (0,3], discretize is spaced apart 1, the span of direction parameter θ is θ ∈ [0, π), discretize is spaced apart π/18, and the span of displacement parameter b is shown below:
b ∈ [ 0 , n × ( s i n θ + c o s θ ) ] θ ∈ [ 0 , π / 2 ) b ∈ [ n × cos θ , n × s i n θ ] θ ∈ [ π / 2 , π ) , Discretize is spaced apart 1;
Then, according to the wave filter of layers 2 and 3 setting in degree of depth convolutional network, and the parameter of above-mentioned discretize, obtain the bank of filters that includes K wave filter, K is the number of this bank of filters median filter;
Finally, difference Stochastic choice L from bank of filters 1and L 2individual wave filter, as the initial value of above-mentioned degree of depth convolutional network layers 2 and 3 wave filter, wherein L 1< K, L 2< K;
6b2) by each normalized training sample square image blocks F jas the input of input layer, through propagated forward, obtain the output class mark of network;
6b3) using network output class mark and training sample true class target least mean-square error as cost function;
6b4) adopt back-propagation algorithm to minimize cost function, obtain the network parameter trained.
Step 7, extracts spectral information and the spatial information of test sample book.
The pixel of 90% is remained as test sample book in high spectrum image 7a) step 1 inputted;
7b) extract the spectral information of test sample book: set the Spectral dimension of the high spectrum image of input in step 1 as V, to each test sample book, extract the spectral value of the every one dimension of this test sample book, composition spectral vector f q, q=1 ..., Q, Q are the numbers of test sample book, spectral vector f qdimension be V;
7c) extract the spatial information of test sample book: in each component map in the image in step 2 after dimensionality reduction, centered by each test sample book, choose the window of 7 × 7 sizes, obtain the spatial information n of this test sample book in component map i, i=1 ..., 4;
7d) according to method in step 5, by the spatial information of each test sample book and n i, i=1 ..., 4 spectral vector f qform a test sample book square image blocks, and this test sample book square image blocks is normalized, obtain normalized test sample book square image blocks F q, q=1 ..., Q, Q are the numbers of test sample book.
Step 8, by normalized test sample book square image blocks F qbe input in the network trained in step 6, carry out propagated forward.
Described propagated forward process is:
First, by normalized test sample book square image blocks F qcarry out convolution with the wave filter of the 2nd layer, obtain the 2nd layer of characteristic pattern exported;
Then, the characteristic pattern exported with the 2nd layer and the wave filter of the 3rd layer carry out convolution, obtain the 3rd layer of characteristic pattern exported;
Then, the characteristic pattern that the 3rd layer exports is input to the 4th layer, through the output calculating the 4th layer of the 4th layer;
Finally, the output of the 4th layer is input to the class scale value obtaining test sample book in the softmax sorter of the 5th layer, completes classification.
Effect of the present invention can further illustrate by following emulation experiment:
(1) simulated conditions
The hardware condition of emulation of the present invention is: windowsXP, SPI, CPUPentium (R) 4, and basic frequency is 2.4GHZ; Software platform is: MatlabR2012a;
Emulate the high spectrum image that the image credit selected is PaviaUniversity, have 9 class atural objects in this image, as shown in Fig. 2 (a), Fig. 2 (b) is the class logo image of Fig. 2 (a) correspondence.
Data in table 1 are to the training sample of each class atural object selection in this image and the number of test sample book.
Table 1
Emulation mode uses the inventive method and existing PCA, sparse PCA, IFRF, EPF and NMFL method respectively.
(2) content and result is emulated
Emulation 1, carries out classification with the present invention and described existing Lung biopsy to Fig. 2 (a) and emulates, result as Fig. 3, wherein:
Fig. 3 (a) is the classification results figure by PCA method,
Fig. 3 (b) is the classification results figure by sparse PCA method,
Fig. 3 (c) is the classification results figure by IFRF method,
Fig. 3 (d) is the classification result figure with EPF,
Fig. 3 (e) is the classification results figure by NMFL method,
Fig. 3 (f) is the classification results figure by the inventive method.
From the classification results figure of Fig. 3 (a)-3 (f), sorting technique precision of the present invention and classifying quality better.
Emulation 2, carries out classification with the method for ridge ripple initialization convolutional layer wave filter of the present invention and existing random initializtion method and these two kinds of initial methods of Gauss's initial method to Fig. 2 (a) and emulates, obtain error decline figure as shown in Figure 4.The horizontal ordinate of Fig. 4 is iterations, and ordinate is the output class mark of training sample and true class target least mean-square error, and along with the increase of iterations, least mean-square error value reduces gradually.
As can be seen from Figure 4, graph of errors of the present invention has fall off rate faster, can reach good nicety of grading with minimum computing time.
Above experimental result shows: compared with the technology of prior art, and the present invention, solving on the adaptive learning Characteristic Problem in classification hyperspectral imagery problem, has obvious advantage, and shortens computing time.

Claims (5)

1., based on the hyperspectral image classification method of ridge ripple and degree of depth convolutional network, comprise the steps:
1) input the class mark of terrestrial object information in a panel height spectrum picture and this image, from this high spectrum image, select the pixel of 10% as training sample;
2) spectral information of training sample is extracted: along the dimension of high spectrum image spectrum, extract the spectral information of each training sample, composition spectral vector f j, j=1 ..., J, J are the numbers of training sample;
3) dimension-reduction treatment is carried out to high spectrum image, retain front 4 principal components, the image after composition dimensionality reduction;
4) spatial information of training sample is extracted: centered by each training sample, in the image after dimensionality reduction on every one dimension, choose the window of 7 × 7 sizes, obtain the spatial information of this sample in this dimension i=1 ..., 4;
5) by the spatial information of each training sample with spectral vector f jform a foursquare training sample image block, and this image is normalized, obtain the training sample square image blocks F after normalization j;
6) degree of depth convolutional network of 5 layers is constructed, and with the training sample square image blocks F after normalization jas the input of this convolutional network, this network is trained, obtain the network trained;
7) using remaining the pixel of 90% in high spectrum image as test sample book, the spectral vector f of each sample is extracted q, and space vector n i, i=1 ..., 4, form a foursquare test sample image block, and this image block is normalized, obtain the test sample book square image blocks F after normalization q, q=1 ..., Q, Q are the numbers of test sample book;
8) by the test sample book square image blocks F after normalization qbe input to step 6) in the network that trains, according to the parameter value trained in network, obtain the class scale value of this sample, complete classification.
2. the hyperspectral image classification method based on ridge ripple and degree of depth convolutional network according to claim 1, wherein said step 5) in by the spatial information of each training sample i=1 ..., 4 and spectral vector f jform a foursquare image block, carry out as follows:
5a) by the image block of the representative image spatial information of 47 × 7 sizes i=1 ..., 4 are combined into the spatial information square image blocks that a size is 14*14
5b) by spectral vector f jbeing rearranged into a size is H*l 2rectangle image block F, H is the length of rectangle image block, l 2wide for rectangle image block, H*l 2=V, wherein V is the dimension of spectral vector;
5c) in spatial information square image blocks G, random selecting size is (H-14) * (H-l 2) image block B, according to this image block B and spatial information square image blocks G and spectral information rectangle image block F, building size is the training sample square image blocks of H*H: .
3. the hyperspectral image classification method based on ridge ripple and degree of depth convolutional network according to claim 1, wherein said step 6) in 5 layer depth convolutional network, its 1st layer is input layer, and layers 2 and 3 is convolutional layer, 4th layer is full linking layer, and the 5th layer is softmax sorter;
Described 1st layer be input as training sample square image blocks F j;
Described 2nd layer comprises L 1individual wave filter;
Described 3rd layer comprises L 2individual wave filter;
Described 4th layer comprises 100 node unit;
Described layer 5 exports the class mark of training sample.
4. the hyperspectral image classification method based on ridge ripple and degree of depth convolutional network according to claim 1, wherein said step 6) in 5 layer depth convolutional network are trained, its step is as follows:
6a) use the wave filter of ridge wave function initialization two convolutional layers;
6b) by the training sample square image blocks F after each normalization jas the input of input layer, through propagated forward, obtain the output class mark of network;
6c) using network output class mark and training sample true class target least mean-square error as cost function;
6d) adopt back-propagation algorithm to minimize cost function, obtain the network parameter trained.
5. the hyperspectral image classification method based on ridge ripple and degree of depth convolutional network according to claim 4, wherein said step 6a) in use the wave filter of ridge wave function initialization two convolutional layers, carry out in accordance with the following steps:
6a1) discretize is carried out to the parameter of continuous ridge wave function:
Described continuous ridge wave function formula is: wherein (x 1, x 2) be the coordinate figure of wave filter, a is scale parameter, and b is displacement parameter, and θ is direction parameter, and ψ (x) is wavelet function,
If the span of scale parameter a be a ∈ (0,3], discretize is spaced apart 1, the span of direction parameter θ be θ ∈ [0, π), discretize is spaced apart π/18, and the span of displacement parameter b is shown below: discretize is spaced apart 1;
6a2) according to the wave filter of layers 2 and 3 setting in degree of depth convolutional network, and step 6a1) the middle parameter arranged, obtain the bank of filters that includes K wave filter, K is the number of this bank of filters median filter;
6a3) difference Stochastic choice L from bank of filters 1and L 2individual wave filter, as the initial value of above-mentioned degree of depth convolutional network layers 2 and 3 wave filter, wherein L 1< K, L 2< K.
CN201510451326.0A 2015-07-28 2015-07-28 Hyperspectral image classification method based on ridge ripple and depth convolutional network Active CN105069468B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201510451326.0A CN105069468B (en) 2015-07-28 2015-07-28 Hyperspectral image classification method based on ridge ripple and depth convolutional network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201510451326.0A CN105069468B (en) 2015-07-28 2015-07-28 Hyperspectral image classification method based on ridge ripple and depth convolutional network

Publications (2)

Publication Number Publication Date
CN105069468A true CN105069468A (en) 2015-11-18
CN105069468B CN105069468B (en) 2018-04-17

Family

ID=54498829

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201510451326.0A Active CN105069468B (en) 2015-07-28 2015-07-28 Hyperspectral image classification method based on ridge ripple and depth convolutional network

Country Status (1)

Country Link
CN (1) CN105069468B (en)

Cited By (20)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105631480A (en) * 2015-12-30 2016-06-01 哈尔滨工业大学 Hyperspectral data classification method based on multi-layer convolution network and data organization and folding
CN105718957A (en) * 2016-01-26 2016-06-29 西安电子科技大学 Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network
CN106023065A (en) * 2016-05-13 2016-10-12 中国矿业大学 Tensor hyperspectral image spectrum-space dimensionality reduction method based on deep convolutional neural network
CN106203482A (en) * 2016-06-30 2016-12-07 东南大学 Characteristics of The Remote Sensing Images dimension reduction method based on mRMR and KPCA
CN106326899A (en) * 2016-08-18 2017-01-11 郑州大学 Tobacco leaf grading method based on hyperspectral image and deep learning algorithm
CN106529570A (en) * 2016-10-14 2017-03-22 西安电子科技大学 Image classification method based on deep ridgelet neural network
CN106997380A (en) * 2017-03-21 2017-08-01 北京工业大学 Imaging spectrum safe retrieving method based on DCGAN depth networks
CN107122708A (en) * 2017-03-17 2017-09-01 广东工业大学 Classification hyperspectral imagery algorithm based on convolutional neural networks and the learning machine that transfinites
CN107194373A (en) * 2017-06-16 2017-09-22 河海大学 A kind of target in hyperspectral remotely sensed image feature extraction and classifying method
CN107292343A (en) * 2017-06-23 2017-10-24 中南大学 A kind of Classification of hyperspectral remote sensing image method based on six layers of convolutional neural networks and spectral space information consolidation
CN107862324A (en) * 2017-10-19 2018-03-30 北京化工大学 A kind of CBR forecast model intellectuality method for early warning based on MWSPCA
CN108171130A (en) * 2017-12-15 2018-06-15 安徽四创电子股份有限公司 A kind of EO-1 hyperion terrain classification recognition methods
CN108229515A (en) * 2016-12-29 2018-06-29 北京市商汤科技开发有限公司 Object classification method and device, the electronic equipment of high spectrum image
CN108428021A (en) * 2018-05-21 2018-08-21 国网山东省电力公司青岛供电公司 Micro-capacitance sensor Short-term Load Forecasting Model based on HSA-RRNN
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
CN108830243A (en) * 2018-06-22 2018-11-16 西安电子科技大学 Hyperspectral image classification method based on capsule network
CN109389080A (en) * 2018-09-30 2019-02-26 西安电子科技大学 Hyperspectral image classification method based on semi-supervised WGAN-GP
CN110458208A (en) * 2019-07-24 2019-11-15 哈尔滨工业大学 Hyperspectral image classification method based on information measure
CN111222576A (en) * 2020-01-08 2020-06-02 西安理工大学 High-resolution remote sensing image classification method
CN112183669A (en) * 2020-11-04 2021-01-05 北京航天泰坦科技股份有限公司 Image classification method and device, equipment and storage medium

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130082858A1 (en) * 2010-08-26 2013-04-04 David H. Chambers Classification of subsurface objects using singular values derived from signal frames
CN104050507A (en) * 2014-06-30 2014-09-17 南京理工大学 Hyper spectrum image classification method based on multilayer neural network
CN104102929A (en) * 2014-07-25 2014-10-15 哈尔滨工业大学 Hyperspectral remote sensing data classification method based on deep learning
CN104700116A (en) * 2015-03-13 2015-06-10 西安电子科技大学 Polarized SAR (synthetic aperture radar) image object classifying method based on multi-quantum ridgelet representation

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130082858A1 (en) * 2010-08-26 2013-04-04 David H. Chambers Classification of subsurface objects using singular values derived from signal frames
CN104050507A (en) * 2014-06-30 2014-09-17 南京理工大学 Hyper spectrum image classification method based on multilayer neural network
CN104102929A (en) * 2014-07-25 2014-10-15 哈尔滨工业大学 Hyperspectral remote sensing data classification method based on deep learning
CN104700116A (en) * 2015-03-13 2015-06-10 西安电子科技大学 Polarized SAR (synthetic aperture radar) image object classifying method based on multi-quantum ridgelet representation

Cited By (29)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105631480B (en) * 2015-12-30 2018-10-26 哈尔滨工业大学 The Hyperspectral data classification method folded based on multilayer convolutional network and data recombination
CN105631480A (en) * 2015-12-30 2016-06-01 哈尔滨工业大学 Hyperspectral data classification method based on multi-layer convolution network and data organization and folding
CN105718957A (en) * 2016-01-26 2016-06-29 西安电子科技大学 Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network
CN106023065A (en) * 2016-05-13 2016-10-12 中国矿业大学 Tensor hyperspectral image spectrum-space dimensionality reduction method based on deep convolutional neural network
CN106023065B (en) * 2016-05-13 2019-02-19 中国矿业大学 A kind of tensor type high spectrum image spectral-spatial dimension reduction method based on depth convolutional neural networks
CN106203482A (en) * 2016-06-30 2016-12-07 东南大学 Characteristics of The Remote Sensing Images dimension reduction method based on mRMR and KPCA
CN106326899A (en) * 2016-08-18 2017-01-11 郑州大学 Tobacco leaf grading method based on hyperspectral image and deep learning algorithm
CN106529570B (en) * 2016-10-14 2019-06-18 西安电子科技大学 Image classification method based on depth ridge ripple neural network
CN106529570A (en) * 2016-10-14 2017-03-22 西安电子科技大学 Image classification method based on deep ridgelet neural network
CN108229515A (en) * 2016-12-29 2018-06-29 北京市商汤科技开发有限公司 Object classification method and device, the electronic equipment of high spectrum image
CN107122708A (en) * 2017-03-17 2017-09-01 广东工业大学 Classification hyperspectral imagery algorithm based on convolutional neural networks and the learning machine that transfinites
CN106997380B (en) * 2017-03-21 2019-07-12 北京工业大学 Imaging spectrum safe retrieving method based on DCGAN depth network
CN106997380A (en) * 2017-03-21 2017-08-01 北京工业大学 Imaging spectrum safe retrieving method based on DCGAN depth networks
CN107194373A (en) * 2017-06-16 2017-09-22 河海大学 A kind of target in hyperspectral remotely sensed image feature extraction and classifying method
CN107292343A (en) * 2017-06-23 2017-10-24 中南大学 A kind of Classification of hyperspectral remote sensing image method based on six layers of convolutional neural networks and spectral space information consolidation
CN107862324A (en) * 2017-10-19 2018-03-30 北京化工大学 A kind of CBR forecast model intellectuality method for early warning based on MWSPCA
CN107862324B (en) * 2017-10-19 2021-11-02 北京化工大学 MWSPCA-based CBR prediction model intelligent early warning method
CN108171130A (en) * 2017-12-15 2018-06-15 安徽四创电子股份有限公司 A kind of EO-1 hyperion terrain classification recognition methods
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
CN108428021B (en) * 2018-05-21 2021-10-12 国网山东省电力公司青岛供电公司 Micro-grid short-term load prediction model based on HSA-RRNN
CN108428021A (en) * 2018-05-21 2018-08-21 国网山东省电力公司青岛供电公司 Micro-capacitance sensor Short-term Load Forecasting Model based on HSA-RRNN
CN108830243A (en) * 2018-06-22 2018-11-16 西安电子科技大学 Hyperspectral image classification method based on capsule network
CN109389080A (en) * 2018-09-30 2019-02-26 西安电子科技大学 Hyperspectral image classification method based on semi-supervised WGAN-GP
CN110458208A (en) * 2019-07-24 2019-11-15 哈尔滨工业大学 Hyperspectral image classification method based on information measure
CN111222576A (en) * 2020-01-08 2020-06-02 西安理工大学 High-resolution remote sensing image classification method
CN111222576B (en) * 2020-01-08 2023-03-24 西安理工大学 High-resolution remote sensing image classification method
CN112183669A (en) * 2020-11-04 2021-01-05 北京航天泰坦科技股份有限公司 Image classification method and device, equipment and storage medium
CN112183669B (en) * 2020-11-04 2024-02-13 航天科工(北京)空间信息应用股份有限公司 Image classification method, device, equipment and storage medium

Also Published As

Publication number Publication date
CN105069468B (en) 2018-04-17

Similar Documents

Publication Publication Date Title
CN105069468A (en) Hyper-spectral image classification method based on ridgelet and depth convolution network
CN110136170B (en) Remote sensing image building change detection method based on convolutional neural network
CN102073879B (en) Method for identifying characteristic land categories of ocean remote sensing images of coast on basis of semi-supervised learning
CN103413151A (en) Hyperspectral image classification method based on image regular low-rank expression dimensionality reduction
CN106529508A (en) Local and non-local multi-feature semantics-based hyperspectral image classification method
Ahmed Urban land cover change detection analysis and modeling spatio-temporal Growth dynamics using Remote Sensing and GIS Techniques: A case study of Dhaka, Bangladesh
CN112232371B (en) American license plate recognition method based on YOLOv3 and text recognition
CN106991374A (en) Handwritten Digit Recognition method based on convolutional neural networks and random forest
CN107463948A (en) Classification of Multispectral Images method based on binary channels multiple features fusion network
CN113343563B (en) Landslide susceptibility evaluation method based on automatic sample selection and surface deformation rate
Cao et al. Facade geometry generation from low-resolution aerial photographs for building energy modeling
CN104298999B (en) EO-1 hyperion feature learning method based on recurrence autocoding
CN104331698A (en) Remote sensing type urban image extracting method
CN114821342B (en) Remote sensing image road extraction method and system
CN112287983B (en) Remote sensing image target extraction system and method based on deep learning
CN112232328A (en) Remote sensing image building area extraction method and device based on convolutional neural network
CN104778482A (en) Hyperspectral image classifying method based on tensor semi-supervised scale cutting dimension reduction
CN112633140A (en) Multi-spectral remote sensing image urban village multi-category building semantic segmentation method and system
CN111639587A (en) Hyperspectral image classification method based on multi-scale spectrum space convolution neural network
CN106846322A (en) Based on the SAR image segmentation method that curve wave filter and convolutional coding structure learn
CN104463223A (en) Hyperspectral image group sparse demixing method based on empty spectral information abundance restraint
CN114202539A (en) Hyperspectral image anomaly detection method based on end-to-end RX
CN110163294A (en) Remote Sensing Imagery Change method for detecting area based on dimensionality reduction operation and convolutional network
CN103425995A (en) Hyperspectral image classification method based on area similarity low rank expression dimension reduction
Li et al. An aerial image segmentation approach based on enhanced multi-scale convolutional neural network

Legal Events

Date Code Title Description
C06 Publication
PB01 Publication
C10 Entry into substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant