CN103955712A - Method for automatically classifying satellite image scene based on morphological component analysis - Google Patents
Method for automatically classifying satellite image scene based on morphological component analysis Download PDFInfo
- Publication number
- CN103955712A CN103955712A CN201410218429.8A CN201410218429A CN103955712A CN 103955712 A CN103955712 A CN 103955712A CN 201410218429 A CN201410218429 A CN 201410218429A CN 103955712 A CN103955712 A CN 103955712A
- Authority
- CN
- China
- Prior art keywords
- satellite image
- scene
- layer
- classification
- base map
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Processing (AREA)
- Image Analysis (AREA)
Abstract
The invention belongs to the technical field of remote sensing data application and satellite image processing and particularly relates to a method for automatically classifying a satellite image scene based on morphological component analysis. The method comprises the following steps of (1) constructing a dictionary matrix through an independent meta-analysis method; (2) decomposing morphological components of a satellite image based on the morphological component analysis theory; (3) decomposing the satellite image into a texture layer and a bottom layer; (4) performing quantitative calculation on the total probability of classification of an objective image scene according to a mechanism of maximum likelihood estimation with combination of features of the texture layer and the bottom layer, and automatically classifying the satellite image scene with the quantitative calculation on the total probability of classification of objective image scenes as the basis. The method combines with a machine vision method and can be applied to a computer easily. System construction is performed more quickly and accurately through the computer than through a manual method, artificial subjective factors are eliminated, and cost is reduced. Furthermore, the method has high automatic identification accuracy and identification stability.
Description
Technical field
The invention belongs to remote sensing data application and satellite image processing technology field, particularly a kind of satellite image scene automatic classification method of analyzing based on anatomic element.
Background technology
In nearly decades, remote sensing data application and satellite image analysis have caused that in many applications people pay close attention to widely, comprise mining industry, forestry, agricultural, military affairs, mapping, city planning, marine monitoring and prevent and reduce natural disasters etc.These application great majority can be summed up as a kind of classification problem, be equivalent to mark is carried out in the different terrain area of remotely sensed image.High-resolution satellite image ratio was more prone in the past and obtained easily in recent years, and the extremely abundant and valuable information simultaneously comprising in satellite image makes the people can be than carried out more exactly topographical surveying and earth observation in the past.Along with satellite image spatial resolution improves day by day, the various objects on earth surface and structural details thereof can be by acquiring satellites, more via satellite picture showing before people.The more important thing is, high-resolution satellite image makes people to differentiate in further detail differentiation to object and scene classification, and finally improves the precision of satellite image classification.
At present, people have proposed some technological means and have solved the problem of remote sensing data application field Satellite image scene automatic classification.The most representative certain methods comprises: Support Vector Machine, nearest neighbour classification device, neural network, border vector detection, linear discriminant analysis etc.Although people have proposed some sorting techniques, and some practical application achievements are obtained, but how to analyze more efficiently high-resolution satellite image, and accuracy how to utilize computer further to improve satellite image scene classification is all far from obtaining perfect solution.There is following defect in existing high-resolution satellite image scene sorting technique.
For high-resolution satellite image, scenario objects and structure thereof are often being dominated the process of Images Classification.The scenario objects belonging to a different category may have similar overall situation response for sorting algorithm, may have distinct overall situation response and originally belong to other scenario objects of same class for same sorting algorithm.Due to the physical presence of this phenomenon, the scene classification method based on view picture satellite image sometimes may cause the mistake of Images Classification.
In aforesaid satellite image scene classification method, Support Vector Machine has been proved to be a kind of high dimensional data that solves theoretically the strong instrument of supervised classification problem, and has obtained in actual applications good effect.But sorting technique based on Support Vector Machine is extremely responsive for the selection of model.More kill, because Support Vector Machine all needs to build a complicated mathematical model for each scene classification, this just causes this kind of method cannot expand in the application that thousands of scene kinds are classified going.
Recently, the nearest neighbour classification method of imparametrization has caused people's concern in Images Classification field.Although but these class methods are very popular, people are still very unclear for how appropriately to define suitably in mathematical meaning " real " neighborhood in high-resolution satellite image.Therefore, can these class methods be used widely, and prospect is still uncertain.
Generally speaking, collect image training sample that mark has been crossed and be suitable difficulty, expend time in and both expensive, and it is just simply too much to collect or directly generate the image training sample of unmarked mistake.In the application in remote sensing field, in order to obtain good classification performance and effect, the image training sample that often needs a large amount of marks to cross.Unfortunately, cannot obtain the abundant training sample of mark is a ubiquitous difficult problem in pattern classification problem.Worse, lack the abundant training sample of mark is a more serious problem in the related application of remotely-sensed data, because in the related application of remotely-sensed data, identification training sample is carried out to mark is extremely difficult and both expensive, sometimes or even cannot realize.Summary is got up, the high-resolution satellite image that has comprised surprising quantity of information obtaining for people, and along with the raising day by day of scene automatic classification demand, most of existing sorting algorithms are not competent.
Mark to satellite image training sample, adopts the method for artificial visual in conjunction with expertise knowledge more at present.Many human factors have caused wrong identification and the mark to satellite image training sample, and artificial cognition and labeling method also exist obvious uncertainty.
Summary of the invention
Existingly manually satellite image training sample identify and the deficiency of labeling method in order to overcome, the invention provides a kind of satellite image scene automatic classification method of analyzing based on anatomic element, the method combines the method for machine vision, be easy to realize on computers, there is higher automatic accuracy of identification and the stability of identification, can successfully manage the challenge running in existing high-resolution satellite Images Classification application.
The invention provides a kind of satellite image scene automatic classification method of analyzing based on anatomic element, the dictionary matrix that its method construct by independent component analysis uses in anatomic element analysis, on this basis, based on anatomic element analysis theories, satellite image is resolved into texture layer and base map layer.The satellite image decomposition method is here designed to the process of an iteration, makes it facilitate computer programming to realize.Last according to the mechanism of maximal possibility estimation, in conjunction with from texture layer and this two-layer feature separately of base map layer, calculate the overall probability size of target image being carried out to scene classification.Concrete steps are as follows:
Step 1: adopt independent component analysis structure dictionary matrix
Carry out independent component analysis by the satellite image that need to carry out scene classification to a width, obtain one group of independently substrate, construct the dictionary matrix of analyzing for anatomic element
;
Step 2: satellite image is carried out to anatomic element decomposition
Satellite image is decomposed and obtains P anatomic element by process of iteration according to real satellite view data:
(1) supplemental characteristic is prepared
According to real satellite view data, set correlation parameter, comprising: satellite image
, dictionary matrix
, iterations
, stop threshold value
and threshold value update scheme;
(2) initialization procedure
Initially dissolve: for arbitrarily
,
;
Initialization residual error:
;y represents the linear hybrid image observing;
Initialization threshold value: order
, and set
;
(3) major cycle process
From
arrive
;
From
arrive
;
1.1. calculate marginal residual error:
;
1.2. upgrade the by limiting threshold value
kthe coefficient of individual anatomic element:
;
Wherein, symbol
represent anatomic element to limit threshold value
process, limiting in the process of rigidity threshold value, if
, so
, otherwise
, in the process of the flexible threshold value of restriction,
;
1.3. upgrade
kindividual anatomic element:
;
2. upgrade residual error:
3. upgrade threshold value according to given threshold value update scheme:
If satisfied condition
, just stop iterative process;
(4) Output rusults
Output decomposes to satellite image the anatomic element obtaining:
.
Step 3: to the decomposition of texture layer and base map layer
With
represent the dictionary matrix of real satellite image texture layer, with
represent the dictionary matrix of real satellite image base map layer;
represent texture layer coefficient vector,
represent base map layer coefficients vector;
expression can be carried out the coefficient vector that optimization represents to texture layer,
expression can be carried out the coefficient vector that optimization represents to base map layer;
If actual satellite image
during not by noise, by solving the protruding optimization problem of the following belt restraining of concrete form, realize the decomposition to satellite image texture layer and base map layer:
And meet constraint condition:
;
If actual satellite image
during by noise, by solving the protruding optimization problem of the following belt restraining of concrete form, realize the decomposition to satellite image texture layer and base map layer:
And meet constraint condition:
, wherein
represent real satellite image
the rank of middle noise;
Step 4: the scene in satellite image is classified based on maximal possibility estimation
Suppose the actual satellite image given at
in have
the target object of kind, every kind has
individual class label,
, the wherein value of each class label
,
; The
in kind
individual class label
weight use
represent; The weighted value of every kind of scene classification being selected is used
weigh, wherein:
,
,
in texture layer
in kind of scene classification the
the coefficient of individual scene class label,
in base map layer
in kind of scene classification the
the coefficient of individual scene class label;
So in conjunction with the characteristic information from texture layer and base map layer, and according to the mechanism of maximal possibility estimation, by coefficient vector
with
after normalization, according to following formula to target satellite image
divide into
the overall probability of planting scene classification calculates:
And taking the above results as the automatic classification according to carrying out satellite image scene.
The present invention's advantage is compared with prior art:
The satellite image scene automatic classification method of analyzing based on anatomic element provided by the invention, utilize the satellite image texture layer morphological characteristic different with base map layer, each anatomic element is decomposed, and according to the mechanism of maximal possibility estimation, overall probability size to target image scene classification is quantitatively calculated, and carries out on this basis the automatic classification of satellite image scene.Adopt computing machine to carry out the realization of method and the structure of system, than manual method more quick and precisely, eliminated artificial subjective factor, reduced cost, also improved the stability of identification.
Brief description of the drawings
Fig. 1 is the process flow diagram of the satellite image scene automatic classification method based on anatomic element analysis.
Fig. 2 is the dictionary for satellite scene automatic classification method and system based on independent component analysis structure.Dictionary comprises 256 atoms, and the size of each atom is 16 × 16 pixels.
Fig. 3 is the original satellite image obtaining from Digital Global company.Campus scene satellite image size is 1300 × 800 pixels.
Fig. 4 is the satellite scene classification result that adopts the satellite image scene classification method and system based on anatomic element analysis to obtain.
Embodiment
Below in conjunction with accompanying drawing and embodiment, the present invention is described in further detail:
the first step: adopt independent component analysis structure dictionary matrix
Classical independent component analysis is a kind of signal processing method that blind source signal separates that is applied to.Independent Component Analysis can be separated into the mixed signal of source signal each independently composition.Carry out independent component analysis by the satellite image that need to carry out scene classification to a width, just can obtain one group of independently substrate.These substrates are proper vectors of higher order statistical, and this width satellite image just can be expressed as the linear combination of these separate substrate simultaneously.In the present invention, adopt Independent Component Analysis to construct the dictionary matrix of analyzing for anatomic element
.
second step: satellite image is carried out to anatomic element decomposition
In order to facilitate the description of specific embodiments, the term that the anatomic element that paper is once used decomposes.
In anatomic element analysis theories, suppose that a width comprises
the image of individual pixel
be
the linear combination of individual anatomic element.Image
likely by additive noise
institute pollutes, and has following expression-form:
The proposition of anatomic element analysis theories is intended to solve the Reverse Problem of above-mentioned hypothesis correspondence.That is to say, anatomic element analysis theories is for the linear hybrid image from observing
in recover each anatomic element
, wherein
.Each anatomic element
can represent by following expression:
Wherein,
for presentation video
?
individual base,
?
the coefficient that individual base is corresponding.
Above-mentioned these linear equations are combined, this linear system can be write as to following vector sum matrix form:
Wherein,
be input picture, it is expressed as one and by lexicographic order sequence, length is
one-dimensional vector.Matrix
(typical situation is:
) be one by
" dictionary " that individual base forms.
it is coefficient vector.By coefficient vector
with dictionary matrix
multiply each other, the linear combination of generation is exactly the image that recovery is constructed
.
Anatomic element analysis theories is by solving Constrained Optimization as follows, carrys out solving equation system:
, and estimate anatomic element:
,
.
, and make
That is:
, and make
Wherein,
,
typical value be a constant:
.Due to the imperfection of linear superposition model itself and the existence of noise, so need to introduce constraint condition in this optimization problem.If there is no noise, and linear superposition model is strictly accurately, meets
, can replace the inequality constrain condition here with equality constraint so.
In the present invention, the concrete steps that adopt anatomic element analysis to decompose satellite image are:
(1) supplemental characteristic is prepared
According to real satellite view data, set correlation parameter, comprising: satellite image:
, dictionary matrix:
, iterations:
, stop threshold value:
and threshold value update scheme.
(2) initialization procedure
Initially dissolve: for arbitrarily
,
Initialization residual error:
Initialization threshold value: order
, and set
(3) major cycle process
From
arrive
From
arrive
1.1 calculate marginal residual error:
1.2 upgrade the by limiting threshold value
kthe coefficient of individual anatomic element:
Wherein, symbol
represent anatomic element to limit threshold value
process.In the process of restriction rigidity threshold value, if
, so
, otherwise
.In the process of the flexible threshold value of restriction,
.
1.3. upgrade
kindividual anatomic element:
2. upgrade residual error:
3. upgrade threshold value according to given threshold value update scheme:
If satisfied condition
, just stop iterative process
(4) Output rusults
Output decomposes to satellite image the anatomic element obtaining:
the 3rd step: to the decomposition of texture layer and base map layer
In general, in a width real image, include geometric figure layer and texture layer.In real image, piecewise smooth content is called as base map layer, and this one deck has only carried the geometry information in real image.In real image, remaining texture content is called as texture layer.And anatomic element analytical approach has good decomposition effect for a width Given Graph as texture layer and the base map layer of neutral line combination.
Hypothesis matrix
the dictionary matrix for representing real satellite image texture layer, matrix
it is the dictionary matrix for representing real satellite image base map layer.And matrix
with
in can representing one deck content, can not represent other one deck content.That is to say, the texture layer of real satellite image can be based on matrix
represent, and the base map layer of real satellite image can be based on matrix
represent.
To the satellite image that a width is formed by texture layer and end map overlay arbitrarily
, can find a kind of based on associating dictionary (both Cover matrixes
, Cover matrix again
) satellite image is carried out to the appropriate method representing, need the protruding optimization problem of the belt restraining solving, concrete form is as follows:
And meet constraint condition:
.
Utilize this to separate, we just can separate the content of real satellite image, wherein
what comprise is the content of texture layer, and
what comprise is the content of base map layer.
If real satellite image is by noise, it just cannot be broken down into texture layer and base map layer completely.In this case, need the protruding optimization problem of the belt restraining solving, concrete form is as follows:
And meet constraint condition:
.
In this case, be only just a kind of approximate method for expressing for the decomposition of real satellite image.The error carrying over is can not, by texture layer dictionary matrix, can not be introduced by base map layer dictionary matrix representation content out again by those.Parameter
represent real satellite image
the rank of middle noise.
the 4th step: the scene in satellite image is classified based on maximal possibility estimation
To satellite image scene, classification is automatically the direct texture layer coefficient vector based on obtaining from anatomic element analysis is decomposed:
with base map layer coefficients vector:
.
We have hypothesis in a given real satellite image
the target object of kind.Every kind has
individual class label,
, the wherein value of each class label
,
.In order further to improve the effect of classification, our created symbol
,
represent
in kind
individual class label
weight.For example, in a given real satellite image, have 3 in the target object of classification, so
.Every kind has two class labels, so
.For first, second and third kind, their class label is respectively
,
with
.Because making expression formula, too much lower rotating savings becomes clear not, so we use
represent
,
with
three's set.Class label
with
value can be from
in choose, so
.So, for example symbol
just representing the 2nd class label in the 3rd kind
weight.
In order to quantize each scene classification in satellite image, we introduce the set of a coefficient:
the weighted value of weighing every kind of scene classification being selected, definition form is as follows:
,?
Wherein
in texture layer
in kind
the sparse coefficient of individual class label,
in base map layer
in kind
the sparse coefficient of individual class label.
By being defined as follows a weight matrix of form
:
We just can be write definition as following form again:
In conjunction with the characteristic information from texture layer and base map layer, and according to the mechanism of maximal possibility estimation, by target satellite image
divide into
the overall probability of planting scene classification can be defined as following form:
On calculate, before this decision rule, all coefficient vectors all should be normalized, the impact with the absolute value that reduces coefficient vector on the result of decision.Result of calculation based on above like this, just can carry out the automatic classification of satellite image scene.
One of detail display is adopted to the example of the inventive method and System Implementation below, this example is to carry out based on MATLAB R2011b software platform, it is upper that this software platform operates in notebook personal computer, adopts the Core i7-2620M central processing unit of Intel Company, and dominant frequency is 2.70GHz.In experiment, by choosing at random sample, the training sample that is used as being labeled, doing is like this in order to ensure that sample result does not rely on any special selection of the training data to being labeled.
In example, adopt the satellite image obtaining from Digital Global company.High-resolution satellite image is to acquire by the satellite of 3 WorldView-1 by name, WorldView-2 and QuickBird.The maximum space resolution of satellite image can reach 0.5 meter of every pixel.
In this example, by selecting the satellite image of school district, Fudan University Handan, performance of the present invention is described.The size of university's scene part used is
pixel.For from statistics viewpoint, make the scene classification of example more know significantly, the sample that has sufficient amount in satellite image has been carried out to mark for seven kinds of physics scenes of mark and training.Selected scene classification comprises: meadow, runway, road, roof, shadow region, trees and waters.In addition remaining pixel is put under " all the other " this classification.
In example, by satellite image itself is carried out to independent component analysis, build the dictionary matrix for satellite scene classification algorithm
.The dictionary that the study of training sample by being labeled for one group (account for all sample sizes 15%) obtains is illustrated in fig. 2 shown below.This dictionary includes 256 atoms, and each atom is aligned to a bulk, and size is
pixel.
Original satellite image for example is illustrated in fig. 3 shown below.On satellite image, adopt method proposed by the invention, by different color, different scenes are marked, the satellite scene classification result of obtaining is illustrated in fig. 4 shown below, result shows that the satellite image scene classification method and system based on anatomic element analysis proposed by the invention have extraordinary classify accuracy for seven kinds of chosen physics scene classifications.
In this external this example, 15% the sample that I have chosen all sample sizes that can obtain in every kind of scene classification at random carries out mark, and as training sample.Remaining sample is used for testing and assessing the actual classification accuracy of the satellite image scene automatic classification method of analyzing based on anatomic element proposing.Table 1 has been enumerated chosen physics scene classification, and the corresponding sample size for training and testing in every kind.Table 2 has been enumerated the classifying quality quantized result that different classes of physics scene is classified.
In the chosen physics scene classification of table 1 for the sample size of training and testing
(ratio: the ratio that refers to sample size for training and account for all sample sizes)
In the chosen physics scene classification of table 2, classify correct sample size and corresponding classify accuracy
。
Claims (1)
1. a satellite image scene automatic classification method of analyzing based on anatomic element, is characterized in that the dictionary matrix that first method construct by independent component analysis uses in anatomic element analysis; On this basis, based on anatomic element analysis theories, satellite image is carried out to anatomic element decomposition, and satellite image is resolved into texture layer and base map layer; According to the mechanism of maximal possibility estimation, in conjunction with from texture layer and this two-layer feature separately of base map layer, the overall probability size of target image scene classification is quantitatively calculated, and carried out on this basis the automatic classification of satellite image scene again; Specifically comprise the steps:
Step 1: adopt independent component analysis structure dictionary matrix
Carry out independent component analysis by the satellite image that need to carry out scene classification to a width, obtain one group of independently substrate, construct the dictionary matrix of analyzing for anatomic element
;
Step 2: satellite image is carried out to anatomic element decomposition
Pass through by process of iteration according to real satellite view data
n iteration iterations satellite image is decomposed and is obtained
pindividual anatomic element
;
(1) supplemental characteristic is prepared
According to real satellite view data, set correlation parameter, comprising: satellite image:
, dictionary matrix:
, iterations:
, stop threshold value:
and threshold value update scheme;
(2) initialization procedure
Initially dissolve: for arbitrarily
,
;
Initialization residual error:
;
y represent the linear hybrid image observing;
Initialization threshold value: order
, and set
;
(3) major cycle process
From
arrive
;
From
arrive
;
1.1. calculate marginal residual error:
;
1.2. upgrade the by limiting threshold value
kthe coefficient of individual anatomic element:
;
Wherein, symbol
represent anatomic element to limit threshold value
process, limiting in the process of rigidity threshold value, if
, so
, otherwise
, in the process of the flexible threshold value of restriction,
;
1.3. upgrade
kindividual anatomic element:
;
2. upgrade residual error:
3. upgrade threshold value according to given threshold value update scheme:
If satisfied condition
, just stop iterative process;
(4) Output rusults
Output decomposes to satellite image the anatomic element obtaining:
;
Step 3: to the decomposition of texture layer and base map layer
With
represent the dictionary matrix of real satellite image texture layer,
represent the dictionary matrix of real satellite image base map layer,
represent texture layer coefficient vector,
represent base map layer coefficients vector,
expression can be carried out the coefficient vector that optimization represents to texture layer,
expression can be carried out the coefficient vector that optimization represents to base map layer;
If actual satellite image
not by noise, by solving the protruding optimization problem of the following belt restraining of concrete form, realize satellite image
the decomposition of texture layer and base map layer:
And meet constraint condition:
;
If actual satellite image
by noise, by solving the protruding optimization problem of the following belt restraining of concrete form, realize satellite image
the decomposition of texture layer and base map layer:
And meet constraint condition:
, wherein
represent actual satellite image
the rank of middle noise;
Step 4: the scene in satellite image is classified based on maximal possibility estimation
Suppose the actual satellite image given at
in have
the target object of kind, every kind has
individual class label,
, the wherein value of each class label
,
; The
in kind
individual class label
weight use
represent; The weighted value of every kind of scene classification being selected is used
weigh, wherein:
,
,
in texture layer
in kind of scene classification the
the coefficient of individual scene class label,
in base map layer
in kind of scene classification the
the coefficient of individual scene class label;
So in conjunction with the characteristic information from texture layer and base map layer, and according to the mechanism of maximal possibility estimation, by coefficient vector
with
after normalization, according to following formula to target satellite image
divide into
the overall probability of planting scene classification calculates:
Again taking above-mentioned result of calculation as the automatic classification according to carrying out satellite image scene.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410218429.8A CN103955712A (en) | 2014-05-22 | 2014-05-22 | Method for automatically classifying satellite image scene based on morphological component analysis |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201410218429.8A CN103955712A (en) | 2014-05-22 | 2014-05-22 | Method for automatically classifying satellite image scene based on morphological component analysis |
Publications (1)
Publication Number | Publication Date |
---|---|
CN103955712A true CN103955712A (en) | 2014-07-30 |
Family
ID=51332986
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201410218429.8A Pending CN103955712A (en) | 2014-05-22 | 2014-05-22 | Method for automatically classifying satellite image scene based on morphological component analysis |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN103955712A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091305A (en) * | 2014-08-11 | 2014-10-08 | 詹曙 | Quick image segmentation method used for computer graph and image processing and based on GPU platform and morphological component analysis |
CN111582237A (en) * | 2020-05-28 | 2020-08-25 | 国家海洋信息中心 | High-resolution image airplane type identification method based on ATSM model |
-
2014
- 2014-05-22 CN CN201410218429.8A patent/CN103955712A/en active Pending
Non-Patent Citations (1)
Title |
---|
CHONG YU 等: ""Satellite Image Classification Using Morphological Component Analysis of Texture and Cartoon Layers"", 《IEEE GEOSCIENCE AND REMOTE SENSING LETTERS》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104091305A (en) * | 2014-08-11 | 2014-10-08 | 詹曙 | Quick image segmentation method used for computer graph and image processing and based on GPU platform and morphological component analysis |
CN104091305B (en) * | 2014-08-11 | 2016-05-18 | 詹曙 | A kind of for Fast image segmentation method computer graphic image processing, based on GPU platform and morphology PCA |
CN111582237A (en) * | 2020-05-28 | 2020-08-25 | 国家海洋信息中心 | High-resolution image airplane type identification method based on ATSM model |
CN111582237B (en) * | 2020-05-28 | 2022-08-12 | 国家海洋信息中心 | ATSM model-based high-resolution image airplane type identification method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
Duffy et al. | Spatial assessment of intertidal seagrass meadows using optical imaging systems and a lightweight drone | |
Pradhan et al. | An easy to use ArcMap based texture analysis program for extraction of flooded areas from TerraSAR-X satellite image | |
CN110287960A (en) | The detection recognition method of curve text in natural scene image | |
CN108171233A (en) | Use the method and apparatus of the object detection of the deep learning model based on region | |
Leon et al. | Improving the synoptic mapping of coral reef geomorphology using object-based image analysis | |
Lu et al. | Optimal spatial resolution of Unmanned Aerial Vehicle (UAV)-acquired imagery for species classification in a heterogeneous grassland ecosystem | |
BR112021015232A2 (en) | SHADOW AND CLOUD MASKING FOR REMOTE SENSING IMAGES IN AGRICULTURE APPLICATIONS USING MULTI-LAYER PERCEPTRON | |
Carbonneau et al. | UAV‐based training for fully fuzzy classification of Sentinel‐2 fluvial scenes | |
Hamada et al. | Estimating life-form cover fractions in California sage scrub communities using multispectral remote sensing | |
CN109948593A (en) | Based on the MCNN people counting method for combining global density feature | |
CN107506792B (en) | Semi-supervised salient object detection method | |
CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
Morgan et al. | Automated analysis of aerial photographs and potential for historic forest mapping | |
CN103745233B (en) | The hyperspectral image classification method migrated based on spatial information | |
Ziaei et al. | A rule-based parameter aided with object-based classification approach for extraction of building and roads from WorldView-2 images | |
CN104346814B (en) | Based on the SAR image segmentation method that level vision is semantic | |
Zhang et al. | A hybrid framework for single tree detection from airborne laser scanning data: A case study in temperate mature coniferous forests in Ontario, Canada | |
CN112070079B (en) | X-ray contraband package detection method and device based on feature map weighting | |
Scholefield et al. | Estimating habitat extent and carbon loss from an eroded northern blanket bog using UAV derived imagery and topography | |
Sarmiento et al. | Methodology for classification of geographical features with remote sensing images: Application to tidal flats | |
Wu et al. | Shape-based object extraction in high-resolution remote-sensing images using deep Boltzmann machine | |
CN113111716A (en) | Remote sensing image semi-automatic labeling method and device based on deep learning | |
Xu et al. | Using linear spectral unmixing for subpixel mapping of hyperspectral imagery: A quantitative assessment | |
Karl et al. | A technique for estimating rangeland canopy-gap size distributions from high-resolution digital imagery | |
Hu et al. | Soil moisture retrieval using convolutional neural networks: Application to passive microwave remote sensing |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
WD01 | Invention patent application deemed withdrawn after publication |
Application publication date: 20140730 |
|
WD01 | Invention patent application deemed withdrawn after publication |