CN101777125A - Method for supervising and classifying complex class of high-resolution remote sensing image - Google Patents
Method for supervising and classifying complex class of high-resolution remote sensing image Download PDFInfo
- Publication number
- CN101777125A CN101777125A CN201010107379A CN201010107379A CN101777125A CN 101777125 A CN101777125 A CN 101777125A CN 201010107379 A CN201010107379 A CN 201010107379A CN 201010107379 A CN201010107379 A CN 201010107379A CN 101777125 A CN101777125 A CN 101777125A
- Authority
- CN
- China
- Prior art keywords
- remote sensing
- sample
- vector
- dimension
- pyramid
- 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
Images
Landscapes
- Image Analysis (AREA)
Abstract
The invention relates to the technical field of remote sensing image processing, in particular to a method for supervising and classifying the complex class of the high-resolution remote sensing image. The invention describes the image content in more details and more accuracy by extracting local feature of the high-resolution remote sensing image or using local feature vector set to represent an image in case of that all classes of feature information mix with each other when global feature is used; the supervising and classifying method of defaulting the input to be a vector is applied by the way that a local characteristic vector set is converted into one expression vector according to the multi-dimension pyramid expression, wherein the multi-dimension pyramid expression method can ensure a relative high computational efficiency and a classification accuracy when the local feature is the high demensional feature; AdaBoost is utilized to choose strong discriminative minority dimension to construct a strong classifier for preventing the negative information from effecting classification accuracy on the basis of obtaining a multi-dimension pyramid expression vector. The method provided by the invention has high computational efficiency, which is adaptable for the high-resolution remote sensing image and can effectively improve the classification accuracy of the complex class of the high-resolution remote sensing image.
Description
Technical field
The present invention relates to technical field of remote sensing image processing, relate in particular to a kind of supervised classification method of complex class of high-resolution remote sensing image.
Background technology
Supervised classification general flow at remote sensing images is: at first obtain remote sensing images sample of all categories, extract the feature of sample, determine discriminant function (being sorter) according to the priori of sample information and classification; The remote sensing images for the treatment of classification then extract-same feature, and the substitution sorter obtains classification output.Feature Extraction and selection, and choosing of sorting algorithm is the key that influences nicety of grading.
Can extract many low-level image features from remote sensing images, common has: based on the statistical nature of gradation of image value, and edge feature, textural characteristics and shape facility etc.Along with improving constantly of remote sensing images resolution, the complicacy of details and structure and randomness are obvious further, and the simple classification that had " feature consistance " originally may become the complex class of " feature is inconsistent ".This can make the sorting algorithm precision based on low-level image feature reduce.At computer vision and area of pattern recognition, at high-resolution optical imagery classification, popular trend is to use earlier to detect the critical area that operator finds image, on critical area, calculate local feature then, obtain local feature vectors set and image is described, carry out Classification and Identification at last by descriptive model.This method is suitable in high-resolution remote sensing image too.But a lot of traditional machine learning methods can not directly utilize the local feature vectors set as input, because the input of these machine learning method acquiescences all is an eigenvector.In order to address this problem, need close again local feature vector set and express, it is mapped as a vector, be beneficial to the conventional machines learning method and use.The method that the local feature vectors set is expressed as vector commonly used has: bag-of-features (BOF) and pyramid are expressed.
BOF sets up the feature prototype library by local feature vector set being closed cluster (as using k-means) in advance, how the vector expression image that constitutes of the probability that occurs in image with the feature prototype determines automatically that the cluster centre number of the best is that BOF has problem to be solved then.
With reference to disclosed pyramid expression in the U.S. Pat 2007/0217676, obtain the feature prototype library by direct division feature space, avoided clustering problem, and pyramid express to feature space from fine to coarse carried out the multilayer distribution of eigenvector of having divided (at every turn dividing) more careful portrayal at feature space corresponding to pyramidal one deck.But pyramid is expressed and had two shortcomings: (1) pyramid is expressed vector length (being dimension) and is exponential relationship with the local feature dimension, and is linear with the pyramidal number of plies.Using under the higher-dimension local feature situation, its expression efficiency is unacceptable; (2) because the higher-dimension local feature vectors often just accumulates in certain little corner of feature space, the pyramid of finite layer (restriction of counting yield) can not this is little " corner " fine division, causes pyramid to express vector and lack differentiation power in the classification application of using high dimensional feature.
Because use local feature description's high-resolution remote sensing image, some irrelevant informations must exert an influence to final expression vector, so need to select a kind of supervised classification method of feature of can selecting to remove the interference of these irrelevant informations.
In sum, at the complex class supervised classification of high-resolution remote sensing image, at first need to extract the local feature of high-resolution remote sensing image; Secondly, need a kind of expression that the local feature vectors set is expressed as the use of a vector with convenient traditional supervised classification method, this method is preferably in and imports when being the higher-dimension local feature simultaneously, can guarantee that also the expression vector of exporting has meticulous information and counting yield height; At last, need to select the supervised classification method of feature to remove the interference of irrelevant information.
Summary of the invention
The supervised classification method that the purpose of this invention is to provide a kind of complex class of high-resolution remote sensing image is to improve the nicety of grading and the efficient of complex class of high-resolution remote sensing image.
For achieving the above object, the present invention adopts following technical scheme:
Step 1 is opened high-resolution remote sensing image by the high-resolution remote sensing image handling procedure;
Step 2 distributes and the class categories number according to actual atural object, utilizes sample district instrument to select training sample on high-resolution remote sensing image, deposits training sample in training sample database;
Step 3 is utilized point of interest to detect operator and detect area-of-interest on training sample;
Step 4 is extracted local feature detecting on the zone, obtain local feature vectors, and the local feature vectors that obtains on the All Ranges constitutes the local feature vectors set, deposits the local feature vectors set in feature database;
Step 5 changes into one with multidimensional pyramid expression with the local feature vectors set and expresses vector, and deposits the feature representation storehouse in;
Step 6 is selected next training sample, and repeating step 3 up to the expression of finishing all training samples, obtains complete feature representation storehouse to step 5;
Step 7, the expression vector that will belong to the first kind is labeled as positive sample, and all expression vectors that do not belong to such are labeled as negative sample;
Step 8, input AdaBoost, training strong classifier;
Step 9, the sample labeling that will belong to next class is positive sample, and all sample labelings that do not belong to such are negative sample, and repeating step 8 is up to the training of the strong classifier of finishing all categories;
Step 15 is selected next image subblock, and repeating step 11 is to step 14, up to the classification of all pixels of finishing the remote sensing images of waiting to classify.
In the described step 5 expression of multidimensional pyramid is finished in each local feature vectors set and is comprised following substep:
Step 5.1 is utilized dimension reduction method, with the local feature vectors dimensionality reduction;
Step 5.2, first dimension of the feature behind the selection dimensionality reduction;
Step 5.3 is calculated pyramid and is expressed vector on this dimension data;
Step 5.4, the following one dimension of the feature behind the selection dimensionality reduction, repeating step 5.3 is up to finish the calculating that pyramid is expressed vector on all dimensions;
Step 5.5, the pyramid that each dimension is obtained are expressed vector and are coupled together and constitute a multidimensional pyramid and express vector.
Utilize AdaBoost training strong classifier to comprise following substep in the described step 8:
Step 8.1 is supposed a total N positive negative sample, and the weight of each sample is 1/N, makes iteration ordinal number t=1;
Step 8.2, get positive negative sample first the dimension data, with all data according to big minispread;
Step 8.3 is got the intermediate value of first value and second value, as thresholding;
Step 8.4, statistics utilizes this thresholding to distinguish the error rate of positive negative sample;
Step 8.5, the intermediate value of taking off a pair of consecutive value are as thresholding, and repeating step 8.4 is up to the error in classification rate that calculates under all thresholdings;
Step 8.6, get error in classification rate minimum thresholding and this moment two classification relative thresholds direction as the Weak Classifier of this dimension selection;
Step 8.7 is got the following one-dimensional data of positive negative sample, and according to big minispread, repeating step 8.3 is to step 8.6 with all data, calculates up to the Weak Classifier of all dimensions to finish;
Step 8.8, the Weak Classifier of selection sort error rate minimum is as best Weak Classifier h
t(x);
Step 8.9 is calculated h by following formula
t(x) weight,
Wherein, ε
tBe h
t(x) error in classification rate;
Step 8.10, upgrade the weight of all samples by following formula:
Wherein, y
nBe the label of sample, Z
tBe to make
The normalization constant;
Step 8.11, t=t+1, repeating step 8.2 arrives step 8.10, any one condition below satisfying:
①ε
t=0;②
(3) t is greater than setting value;
Step 8.12, calculate the strong classifier of output by following formula:
The present invention has the following advantages and good effect:
1) by high-resolution remote sensing image is extracted local feature, described picture material in more detail, exactly, when having avoided the use global characteristics, various types of other characteristic information is mixed in together;
2) on the basis that obtains the local feature vectors set, obtain expressing vector by the expression of multidimensional pyramid, the supervised classification method that feasible acquiescence input is a vector is used, and multidimensional pyramid expression still guarantees higher counting yield and nicety of grading when local feature is high dimensional feature;
3) on the basis that obtains multidimensional pyramid expression vector, select the strong minority Wesy of differentiation power in making up strong classifier by AdaBoost, avoided the influence of irrelevant information to nicety of grading;
4) the counting yield height is applicable to high-resolution remote sensing image, can effectively improve the nicety of grading of complex class of high-resolution remote sensing image.
Description of drawings
Fig. 1 is the FB(flow block) that the multidimensional pyramid that proposes among the present invention is expressed.
Fig. 2 is the FB(flow block) of the supervised classification method of the complex class of high-resolution remote sensing image that proposes of the present invention.
Embodiment
The invention will be further described in conjunction with the accompanying drawings with specific embodiment below:
For the ease of understanding the supervised classification method of the complex class of high-resolution remote sensing image that the present invention proposes, at first the multidimensional pyramid expression theory that the present invention is proposed is set forth, Fig. 1 is the FB(flow block) that the multidimensional pyramid that proposes among the present invention is expressed, and the flow process that the multidimensional pyramid is expressed comprises following three steps:
1, local feature vector set is closed 10 dimensionality reductions 11 to remove the correlativity between each dimension; To the local feature vectors dimensionality reduction, on the one hand the high dimensional feature vector is changed into the low-dimensional eigenvector, simplify and calculate, uncorrelated between each dimension of the feature behind the dimensionality reduction on the other hand, make things convenient for next step that each dimension is handled respectively.
In the specific implementation, the user will select dimension reduction method according to the internal relation between each dimension of local feature, for example, if each dimension of local feature was linear dependence originally, available PCA dimensionality reduction, if each dimension of local feature was nonlinear dependence originally, available flow pattern method dimensionality reduction etc.
2, on each dimension of the feature behind the dimensionality reduction, calculate pyramid respectively and express vector 12;
Suppose that I represents an image, use the point of interest detection algorithm from I, to detect m marking area { p
1, p
2..., p
m.At each regional p
iThe middle D dimension local feature vectors of calculating
Result images I gathers X={x with an eigenvector
1..., x
mDescribe.Through dimensionality reduction, obtain d dimensional feature vector
This moment, the eigenvector set became
Suppose to calculate pyramid and express vector on the j dimension of eigenvector, promptly this moment, feature space was F
j d, the eigenvector set becomes
The maximal value of supposing the j dimension data again is V.Use statistics with histogram X
d(j) distribution, the size of initial seasonal histogram bin are 1, and the gained histogram table is shown H
0(X
d(j)).Subsequently, histogram bin is enlarged gradually (make it be respectively 2,4 ..., 2
i... V) and calculate corresponding histogram, can obtain altogether
Individual histogram.All histograms are coupled together to constitute a vector be exactly that pyramid is expressed vector Ψ (X
d(j))=[H
0(X
d(j)) ..., H
L-1(X
d(j))], H wherein
i(X
d(j)) i histogram of expression, i=0,1 ..., L-1, L are also referred to as the pyramidal number of plies.
3, the pyramid that each dimension obtained is expressed vector and is coupled together and 13 constitute the multidimensional pyramids and express vectors 14.
The multidimensional pyramid is expressed vector representation: MP (X
d)=[Ψ (X
d(1)) ..., Ψ (X
d(d))].
The dimension that the multidimensional pyramid is expressed vector is:
As seen the dimension of multidimensional pyramid expression vector is proportional with the dimension of feature, also proportional with the pyramidal number of plies, even input is a high dimensional feature, multidimensional pyramid expression still keeps higher counting yield, and to high-dimensional feature space arbitrarily careful division can significantly not reduce counting yield yet, make the multidimensional pyramid of output express vector simultaneously again and have meticulous information.
The supervised classification method of complex class of high-resolution remote sensing image provided by the invention as shown in Figure 2, may further comprise the steps:
Step 1 is opened high-resolution remote sensing image by the high-resolution remote sensing image handling procedure;
Step 2 distributes and the class categories number according to actual atural object, utilizes sample district instrument to select training sample 21,22 on high-resolution remote sensing image, deposits training sample in training sample database;
Step 3 is utilized point of interest to detect operator and detect area-of-interest on training sample;
During concrete enforcement, detect operator and can use uniform grid, the DoG that stochastic sampling or Lowe propose detects son etc.Note using the set of a pixel of " zone " expression here, i.e. image subset arbitrarily.
Step 4 is extracted local feature detecting on the zone, obtain local feature vectors 23, and the local feature vectors that obtains on the All Ranges constitutes the local feature vectors set, deposits the local feature vectors set in feature database;
Step 5 changes into one with multidimensional pyramid expression with the local feature vectors set and expresses vector 24, and deposits the feature representation storehouse in, and the expression of multidimensional pyramid is finished in each local feature vectors set needs following step:
Step 5.1 is utilized dimension reduction method, with the local feature vectors dimensionality reduction;
Step 5.2, first dimension of the feature behind the selection dimensionality reduction;
Step 5.3 is calculated pyramid and is expressed vector on this dimension data;
Step 5.4, the following one dimension of the feature behind the selection dimensionality reduction, repeating step 5.3 is up to finish the calculating that pyramid is expressed vector on all dimensions;
Step 5.5, the pyramid that each dimension is obtained are expressed vector and are coupled together and constitute a multidimensional pyramid and express vector.
Step 6 is selected next training sample, and repeating step 3 up to the expression of finishing all training samples, obtains complete feature representation storehouse to step 5.
Step 7, the expression vector that will belong to the first kind is labeled as positive sample, and all expression vectors that do not belong to such are labeled as negative sample.
Step 8, training strong classifier 25 among the input AdaBoost, utilize AdaBoost training strong classifier 26 to need following several steps:
(Adaboost is a sorter learning method well known in the art, does not repeat them here.)
Step 8.1 is supposed a total N positive negative sample, and the weight of each sample is 1/N, iterations t=1.
Step 8.2, get positive negative sample first the dimension data, with all data according to big minispread;
Step 8.3 is got the intermediate value of first value and second value, as thresholding.
Step 8.4, statistics utilizes this thresholding to distinguish the error rate of positive negative sample;
Step 8.5, the intermediate value of taking off a pair of consecutive value are as thresholding, and repeating step 8.4 is up to the error in classification rate that calculates under all thresholdings;
Step 8.6, get error in classification rate minimum thresholding and this moment two classification relative thresholds direction as the Weak Classifier of this dimension selection;
Step 8.7 is got the following one-dimensional data of positive negative sample, and according to big minispread, repeating step 8.3 is to step 8.6 with all data, calculates up to the Weak Classifier of all dimensions to finish;
Step 8.8, the Weak Classifier of selection sort error rate minimum is as best Weak Classifier h
t(x);
Step 8.9 is calculated h by following formula
t(x) weight,
Wherein, ε
tBe h
t(x) error in classification rate.
Step 8.10, upgrade the weight of all samples by following formula:
Wherein, y
nBe the label of sample, Z
tBe to make
The normalization constant.
Step 8.11, t=t+1, repeating step 8.2 arrives step 8.10, any one condition below satisfying: (1) ε
t=0; (2)
(3) t is greater than setting value.
Step 8.12, calculate the strong classifier of output by following formula:
Step 9, the sample labeling that will belong to next class is positive sample, all sample labelings that do not belong to such are negative sample.Repeating step 8 is up to the training of the strong classifier of finishing all categories;
Step 15 is selected next image subblock, and repeating step 11 is to step 14, up to the classification 33 of all pixels of finishing the remote sensing images of waiting to classify;
Present embodiment is applied in technical scheme provided by the present invention in the High Resolution SAR Image Classification.
(1) utilizes the Flame Image Process instrument, open the high resolution SAR image that a width of cloth size is 20000 * 20000 pixels.This image is the somewhere, Toronto city that TerraSAR-X takes.
(2) roughly comprise unartificial building, culture public place and residential area 3 class atural objects on this image.Distribute according to actual atural object, utilize sample district instrument to select training sample on this image, deposit training sample in training sample database, every class is selected 100 width of cloth training samples, totally 300 width of cloth training sample composing training sample storehouses, and the size of every image is 128 * 128 pixels.
(3) divide uniform grid on every training sample, sizing grid is 16 * 16 pixels, and is overlapping 50% between grid, obtains 225 area-of-interests altogether.
(4) extract 32 dimension histograms as local feature on each grid, this histogram can be by obtaining after the histogram uniform quantization with 16 bit data.Every local feature vectors set description that image is made of 225 32 n dimensional vector ns.
(5) use PCA that 4 dimensional feature set of vectors are reduced in 32 dimension local feature vectors set, calculate 5 layers of pyramid in each dimension and express vector, the pyramid expression vector that again each dimension is obtained couples together and constitutes multidimensional pyramid expression vector, and the dimension that the multidimensional pyramid that obtains is expressed vector is 85.
(6) on all training samples of training sample database, finish the operation that (3) arrive (5).
(7) sorter of the training first kind, all training samples that belong to the first kind are labeled as positive sample, and all training samples that do not belong to the first kind are labeled as negative sample.
(8) AdaBoost iteration 100 times and use decision tree to train strong classifier as Weak Classifier.
(9) obtain 3 strong classifiers altogether.
(10) to be classified is another significantly high resolution SAR image, and the image size is 1680 * 1984 pixels, and resolution is 1.25 meters.The SAR image of will waiting to classify is divided into the image subblock of 128 * 128 pixels and (compares with nonoverlapping sub-piece, the easier elimination of the classification results serrated boundary that overlapping sub-piece obtains, and also can to utilize dividing method to obtain irregular image subblock, to use the sub-piece of nonoverlapping rule in order simplifying to calculate here).
(11) dividing size on each sub-piece is the grid of 16 * 16 pixels, between grid overlapping 50%.Histogram after extracting 32 dimension uniform quantizations on each grid.The set description that each image subblock is made of 225 32 eigenvectors of tieing up.Because AdaBoost has selected the feature dimensions that is used to classify in (8), do not need this moment the multidimensional pyramid of each image subblock to express all dimensions of vector, and the multidimensional pyramid that only calculates this image subblock is expressed vector corresponding value on the selected feature dimensions of AdaBoost and is got final product.
(12) value on the selected feature dimensions that will obtain is imported respectively in 3 strong classifiers, obtain the probability that this image subblock belongs to three classifications respectively, get the classification of that strong classifier corresponding class of probability maximum wherein as this image subblock, promptly the classification of image subblock is G (MP (X))=argmax
Y ∈ Yf
y(MP (X)).
(13) classification of sub-piece is given all pixels of this sub-piece.
(14) give the pixel that it covers after the sub-piece of all images on the image is all calculated classification by aforesaid operations, obtain final classification results figure, finish classification.
Classification results in the example shows, the sorting technique at complex class of high-resolution remote sensing image that proposes among the present invention can obtain higher nicety of grading, especially when local feature is high dimensional feature, not only counting yield is high but also guarantee nicety of grading preferably.
Claims (2)
1. the supervised classification method of a complex class of high-resolution remote sensing image is characterized in that, may further comprise the steps:
Step 1 is opened high-resolution remote sensing image by the high-resolution remote sensing image handling procedure;
Step 2 distributes and the class categories number according to actual atural object, utilizes sample district instrument to select training sample on high-resolution remote sensing image, deposits training sample in training sample database;
Step 3 utilizes the detection operator to detect area-of-interest on training sample;
Step 4 is extracted local feature detecting on the zone, obtain local feature vectors, and the local feature vectors that obtains on the All Ranges constitutes the local feature vectors set, deposits the local feature vectors set in feature database;
Step 5 changes into one with multidimensional pyramid expression with the local feature vectors set and expresses vector, and deposits the feature representation storehouse in;
Step 6 is selected next training sample, and repeating step 3 up to the expression of finishing all training samples, obtains complete feature representation storehouse to step 5;
Step 7, the expression vector that will belong to the first kind is labeled as positive sample, and all expression vectors that do not belong to such are labeled as negative sample;
Step 8, input AdaBoost, training strong classifier;
Step 9, the sample labeling that will belong to next class is positive sample, and all sample labelings that do not belong to such are negative sample, and repeating step 8 is up to the training of the strong classifier of finishing all categories;
Step 10 is decomposed into image subblock with high-resolution remote sensing image to be classified;
Step 11 to step 5, but is only calculated the value that multidimensional pyramid that strong classifier need use is expressed the feature dimensions of vector to an image subblock repeating step 3, obtains the expression vector of this image subblock;
Step 12 in the strong classifier that the expression vector input of image subblock is all kinds of, is calculated this expression vector and is belonged to all kinds of probability;
Step 13 is selected the classification of that classification of probability maximum as image subblock;
Step 14 is given this all pixels that sub-piece covers with the classification of sub-piece;
Step 15 is selected next image subblock, and repeating step 11 is to step 14, up to the classification of all pixels of finishing the remote sensing images of waiting to classify.
2. the supervised classification method of complex class of high-resolution remote sensing image according to claim 1 is characterized in that:
In the described step 5 expression of multidimensional pyramid is finished in each local feature vectors set and is comprised following substep:
Step 5.1 is utilized dimension reduction method, with the local feature vectors dimensionality reduction;
Step 5.2, first dimension of the feature behind the selection dimensionality reduction;
Step 5.3 is calculated pyramid and is expressed vector on this dimension data;
Step 5.4, the following one dimension of the feature behind the selection dimensionality reduction, repeating step 5.3 is up to finish the calculating that pyramid is expressed vector on all dimensions;
Step 5.5, the pyramid that each dimension is obtained are expressed vector and are coupled together and constitute a multidimensional pyramid and express vector.
3. the supervised classification method of complex class of high-resolution remote sensing image according to claim 1 and 2 is characterized in that:
Utilize AdaBoost training strong classifier to comprise following substep in the described step 8:
Step 8.1 is supposed a total N positive negative sample, and the weight of each sample is 1/N, makes iteration ordinal number t=1;
Step 8.2, get positive negative sample first the dimension data, with all data according to big minispread;
Step 8.3 is got the intermediate value of first value and second value, as thresholding;
Step 8.4, statistics utilizes this thresholding to distinguish the error rate of positive negative sample;
Step 8.5, the intermediate value of taking off a pair of consecutive value are as thresholding, and repeating step 8.4 is up to the error in classification rate that calculates under all thresholdings;
Step 8.6, get error in classification rate minimum thresholding and this moment two classification relative thresholds direction as the Weak Classifier of this dimension selection;
Step 8.7 is got the following one-dimensional data of positive negative sample, and according to big minispread, repeating step 8.3 is to step 8.6 with all data, calculates up to the Weak Classifier of all dimensions to finish;
Step 8.8, the Weak Classifier of selection sort error rate minimum is as best Weak Classifier h
t(x);
Step 8.9 is calculated h by following formula
t(x) weight,
Wherein, ε
tBe h
t(x) error in classification rate;
Step 8.10, upgrade the weight of all samples by following formula:
Step 8.11, t=t+1, repeating step 8.2 arrives step 8.10, any one condition below satisfying:
1. ε
t=0; 2.
3. t is greater than setting value; Step 8.12, calculate the strong classifier of output by following formula:
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101073798A CN101777125B (en) | 2010-02-03 | 2010-02-03 | Method for supervising and classifying complex category of high-resolution remote sensing image |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN2010101073798A CN101777125B (en) | 2010-02-03 | 2010-02-03 | Method for supervising and classifying complex category of high-resolution remote sensing image |
Publications (2)
Publication Number | Publication Date |
---|---|
CN101777125A true CN101777125A (en) | 2010-07-14 |
CN101777125B CN101777125B (en) | 2012-01-11 |
Family
ID=42513583
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN2010101073798A Expired - Fee Related CN101777125B (en) | 2010-02-03 | 2010-02-03 | Method for supervising and classifying complex category of high-resolution remote sensing image |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN101777125B (en) |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102096826A (en) * | 2010-11-24 | 2011-06-15 | 清华大学 | Compound classification method for multi-resolution remote sensing image on basis of real likelihood characteristic |
CN102110227A (en) * | 2010-11-24 | 2011-06-29 | 清华大学 | Compound method for classifying multiresolution remote sensing images based on context |
CN102663410A (en) * | 2012-02-27 | 2012-09-12 | 北京交通大学 | Method and system for detecting microcalcifications in mammogram |
CN102831411A (en) * | 2012-09-07 | 2012-12-19 | 云南晟邺科技有限公司 | Quick face detection method |
CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
CN105005789A (en) * | 2015-07-01 | 2015-10-28 | 北京理工大学 | Vision lexicon based remote sensing image terrain classification method |
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 |
CN107249434A (en) * | 2015-02-12 | 2017-10-13 | 皇家飞利浦有限公司 | Robust classification device |
CN109284706A (en) * | 2018-09-12 | 2019-01-29 | 北京英视睿达科技有限公司 | Hot spot grid agglomeration of industries area recognizing method based on Multi-sensor satellite remote sensing |
CN113537290A (en) * | 2021-06-16 | 2021-10-22 | 广东工业大学 | Image matching method based on ultra-high dimensional data element clustering |
-
2010
- 2010-02-03 CN CN2010101073798A patent/CN101777125B/en not_active Expired - Fee Related
Cited By (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102110227A (en) * | 2010-11-24 | 2011-06-29 | 清华大学 | Compound method for classifying multiresolution remote sensing images based on context |
CN102096826A (en) * | 2010-11-24 | 2011-06-15 | 清华大学 | Compound classification method for multi-resolution remote sensing image on basis of real likelihood characteristic |
CN102663410A (en) * | 2012-02-27 | 2012-09-12 | 北京交通大学 | Method and system for detecting microcalcifications in mammogram |
CN102663410B (en) * | 2012-02-27 | 2014-04-02 | 北京交通大学 | Method and system for detecting microcalcifications in mammogram |
CN102831411B (en) * | 2012-09-07 | 2016-05-04 | 云南晟邺科技有限公司 | A kind of fast face detecting method |
CN102831411A (en) * | 2012-09-07 | 2012-12-19 | 云南晟邺科技有限公司 | Quick face detection method |
CN102945374A (en) * | 2012-10-24 | 2013-02-27 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
CN102945374B (en) * | 2012-10-24 | 2015-07-08 | 北京航空航天大学 | Method for automatically detecting civil aircraft in high-resolution remote sensing image |
CN107249434A (en) * | 2015-02-12 | 2017-10-13 | 皇家飞利浦有限公司 | Robust classification device |
US10929774B2 (en) | 2015-02-12 | 2021-02-23 | Koninklijke Philips N.V. | Robust classifier |
CN105005789A (en) * | 2015-07-01 | 2015-10-28 | 北京理工大学 | Vision lexicon based remote sensing image terrain classification method |
CN105005789B (en) * | 2015-07-01 | 2018-04-03 | 北京理工大学 | A kind of remote sensing images terrain classification method of view-based access control model vocabulary |
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 |
CN109284706A (en) * | 2018-09-12 | 2019-01-29 | 北京英视睿达科技有限公司 | Hot spot grid agglomeration of industries area recognizing method based on Multi-sensor satellite remote sensing |
CN109284706B (en) * | 2018-09-12 | 2023-12-01 | 国际商业机器(中国)投资有限公司 | Hot spot grid industrial aggregation area identification method based on multi-source satellite remote sensing data |
CN113537290A (en) * | 2021-06-16 | 2021-10-22 | 广东工业大学 | Image matching method based on ultra-high dimensional data element clustering |
CN113537290B (en) * | 2021-06-16 | 2022-08-12 | 广东工业大学 | Image matching method based on ultra-high dimensional data element clustering |
Also Published As
Publication number | Publication date |
---|---|
CN101777125B (en) | 2012-01-11 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN101777125B (en) | Method for supervising and classifying complex category of high-resolution remote sensing image | |
Agarwal et al. | Hyperfeatures–multilevel local coding for visual recognition | |
CN108596154B (en) | Remote sensing image classification method based on high-dimensional feature selection and multilevel fusion | |
CN100461204C (en) | Method for recognizing facial expression based on 2D partial least square method | |
CN102208038B (en) | Image classification method based on visual dictionary | |
CN108171136A (en) | A kind of multitask bayonet vehicle is to scheme to search the system and method for figure | |
CN104281849B (en) | A kind of cloth color of image feature extracting method | |
CN103761531A (en) | Sparse-coding license plate character recognition method based on shape and contour features | |
CN102254303A (en) | Methods for segmenting and searching remote sensing image | |
CN103489005A (en) | High-resolution remote sensing image classifying method based on fusion of multiple classifiers | |
CN102393914A (en) | Land-based cloud chart recognition method based on classification trees of support vector machine | |
CN102324038A (en) | A kind of floristics recognition methods based on digital picture | |
CN106780503A (en) | Remote sensing images optimum segmentation yardstick based on posterior probability information entropy determines method | |
CN112733936A (en) | Recyclable garbage classification method based on image recognition | |
Agarwal et al. | Multilevel image coding with hyperfeatures | |
CN104850859A (en) | Multi-scale analysis based image feature bag constructing method | |
CN104268507A (en) | Manual alphabet identification method based on RGB-D image | |
CN104156690A (en) | Gesture recognition method based on image space pyramid bag of features | |
Zhu et al. | Learning reconfigurable scene representation by tangram model | |
CN107563393A (en) | A kind of extraction of inscriptions on bones or tortoise shells picture Local textural feature and matching process and system | |
CN113743523B (en) | Building rubbish fine classification method guided by visual multi-feature | |
CN105894035B (en) | SAR image classification method based on SAR-SIFT and DBN | |
CN104268557A (en) | Polarization SAR classification method based on cooperative training and depth SVM | |
CN105844299A (en) | Image classification method based on bag of words | |
Selvarajah et al. | Combined feature descriptor for content based image retrieval |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
C14 | Grant of patent or utility model | ||
GR01 | Patent grant | ||
CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20120111 Termination date: 20150203 |
|
EXPY | Termination of patent right or utility model |