CN109558806A - The detection method and system of high score Remote Sensing Imagery Change - Google Patents
The detection method and system of high score Remote Sensing Imagery Change Download PDFInfo
- Publication number
- CN109558806A CN109558806A CN201811316602.2A CN201811316602A CN109558806A CN 109558806 A CN109558806 A CN 109558806A CN 201811316602 A CN201811316602 A CN 201811316602A CN 109558806 A CN109558806 A CN 109558806A
- Authority
- CN
- China
- Prior art keywords
- image
- branch
- convolutional neural
- classification
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2135—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on approximation criteria, e.g. principal component analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/088—Non-supervised learning, e.g. competitive learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Computing Systems (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Multimedia (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses the detection methods and system of a kind of high score Remote Sensing Imagery Change.Wherein, this method comprises: obtain two when phase images in each pixel neighborhood characteristics；Neighborhood characteristics input branch's convolutional neural networks model is obtained into the classification of image, wherein classification is variation class and does not change class；Branch's convolutional neural networks model is to carry out unsupervised pre-training to image to obtain, branch's convolutional neural networks model includes input layer, branching networks, full articulamentum and classification layer, branching networks are for using neighborhood characteristics, as label and by neighborhood characteristics convolution dimensionality reduction, layer of classifying to be different classes of for dividing the image into.The present invention solves technical problem not high to Image Change Detection precision in the prior art.
Description
Technical field
The present invention relates to high score remote sensing images to change detection field, in particular to a kind of high score Remote Sensing Imagery Change
Detection method and system.
Background technique
Surface ecosystems and human social activity are all active development and constantly develop.Earth's surface is accurately acquired in real time
Change information for preferably preserve the ecological environment, manage natural resources, research social development, and understand mankind's activity with from
Relationship and reciprocation between right environment have great significance.
The many change detecting methods proposed both at home and abroad at present can substantially be divided into two classes: the variation detection of pixel-oriented
Variation with object-oriented detects.The variation detection of pixel-oriented includes traditional image algebraic approach, image converter technique etc..Image
Algebraic approach it is more representative be exactly Change vector Analysis method, that is, calculate two phase image respective pixel gray values away from
From reusing threshold value and be split.Image converter technique is more various, but principle is essentially all from original multidate image wave
Variation atural object can be protruded, distinguish variation class another characteristic by extracting in segment data, is reused the methods of Threshold segmentation and is obtained most
Variation testing result eventually, wherein more representational algorithm is PCA differential technique.The presence of the variation detection of pixel-oriented lacks
Point, first, result of variations be flooded with a large amount of noise, second, due to the method using Threshold segmentation, result of variations usually includes
The variation of as caused by video imaging tone difference etc. puppet.
The variation detection of object-oriented it is main first using partitioning algorithm by original remote sensing image be divided into similar spectral,
Space, textural characteristics region obtain final variation detection by comparing the feature and boundary information of two phase corresponding regions
As a result.The shortcomings that variation detection of object-oriented is to miss due to being first split or classifying to original remote sensing image so existing
The not high problem of variation detection accuracy caused by difference is accumulative, final variation testing result are largely dependent upon segmentation or divide
The order of accuarcy of class.
For above-mentioned problem, currently no effective solution has been proposed.
Summary of the invention
The embodiment of the invention provides the detection methods and system of a kind of high score Remote Sensing Imagery Change, existing at least to solve
The technical problem not high to Image Change Detection precision in technology.
According to an aspect of an embodiment of the present invention, a kind of detection method of high score Remote Sensing Imagery Change is provided, comprising:
When obtaining two in phase images each pixel neighborhood characteristics；Neighborhood characteristics input branch's convolutional neural networks model is obtained
The classification of described image, wherein the classification is variation class or does not change class；Branch's convolutional neural networks branch convolution mind
It is to carry out unsupervised pre-training to described image to obtain through network model, branch's convolutional neural networks branch convolutional Neural
Network model includes input layer, branching networks, full articulamentum and classification layer, and the branching networks are for by the neighborhood characteristics
As label and by the neighborhood characteristics convolution dimensionality reduction, the classification layer is different classes of for described image to be divided into.
Further, neighborhood characteristics input branch's convolutional neural networks model is obtained into the classification packet of described image
It includes: the neighborhood characteristics being inputted into the branching networks and obtain convolution feature；The convolution feature is carried out by full articulamentum
Dimensionality reduction obtains dimensionality reduction feature；The dimensionality reduction feature is inputted into the classification layer and obtains the classification of described image.
Further, the neighborhood characteristics of each pixel include: before to be calculated by Histogram Matching in phase images when obtaining two
Method is pre-processed the image after being corrected to described image, wherein the tone of image and other images after the correction
Unanimously.
Further, when taking two in phase images before the neighborhood characteristics of each pixel, described image pre-process
Include: after image after to correction using object pixel 5 × 5 Neighborhood matrixes as feature extraction standard to the correction
Image afterwards carries out feature extraction.
Further, branch's convolutional neural networks model after training, by the neighborhood characteristics of phase images when two
Position is exchanged, by the neighborhood characteristics after exchanging input described model of branch's convolutional neural networks retraining obtain it is new
Branch's convolutional neural networks model.
Further, after neighborhood characteristics input branch's convolutional neural networks model being obtained the classification of described image
It include: by the brightness of Principal Component Analysis and each phase, saturation degree and tone value component computational shadowgraph index, wherein institute
Stating the high place of shadow index indicates that shaded area, the low place of the shadow index indicate not to be shaded area；Pass through maximum
Ostu method segmentation shadow index image obtains shadow image, wherein the shadow image is bianry image；In change to be detected
The shadow image is subtracted in the image of change obtains the image of removal shadow interference.
Further, subtracted in the image of variation to be detected the shadow image obtain removal shadow interference image it
Afterwards, comprising: by mean shift algorithm by it is described removal shadow interference image segmentation be multiple subgraphs, wherein every height
Image is same colour type；Count total picture that variation pixel in the image of the removal shadow interference accounts for each colour type
The ratio of prime number；When the ratio is more than preset threshold, then all pixels of current class are variation pixel, if the ratio is not
More than the preset threshold, then all pixels of current class are not change pixel.
According to another aspect of an embodiment of the present invention, a kind of high score Remote Sensing Imagery Change Detection system is additionally provided, comprising:
Obtain module, when for obtaining two in phase images each pixel neighborhood characteristics；Categorization module, for the neighborhood characteristics are defeated
Enter branch's convolutional neural networks model and obtain the classification of described image, wherein the classification is variation class or does not change class；It is described
Branch's convolutional neural networks model is to carry out unsupervised pre-training to described image to obtain, branch's convolutional neural networks mould
Type includes input layer, branching networks, full articulamentum and classification layer, and the branching networks are for using the neighborhood characteristics as mark
Sign and by the neighborhood characteristics convolution dimensionality reduction, the classification layer is different classes of for described image to be divided into.
In embodiments of the present invention, using the neighborhood characteristics of each pixel in phase images when acquisition two；The neighborhood is special
Sign input branch's convolutional neural networks model obtains the classification of described image, wherein the classification is variation class and does not change class；
Branch's convolutional neural networks model is to carry out unsupervised pre-training to described image by branch's convolutional neural networks to obtain
, branch's convolutional neural networks model includes input layer, branching networks, full articulamentum and classification layer, the branching networks
It is for using the neighborhood characteristics, as label and by the neighborhood characteristics convolution dimensionality reduction, the classification layer to be for will be described
Image is divided into inhomogeneity otherwise, solves technical problem not high to Image Change Detection precision in the prior art, improves
Detection efficiency and detection accuracy.
Detailed description of the invention
The drawings described herein are used to provide a further understanding of the present invention, constitutes part of this application, this hair
Bright illustrative embodiments and their description are used to explain the present invention, and are not constituted improper limitations of the present invention.In the accompanying drawings:
Fig. 1 is a kind of flow chart of the detection method of high score Remote Sensing Imagery Change according to an embodiment of the present invention；
Fig. 2 is a kind of specific flow chart of the detection method of high score Remote Sensing Imagery Change according to an embodiment of the present invention；
Fig. 3 is the schematic diagram of data prediction effect according to an embodiment of the present invention；
Fig. 4 is the flow chart of feature extraction phases according to an embodiment of the present invention；
Fig. 5 is the schematic diagram of branch's convolutional neural networks model according to an embodiment of the present invention；
Fig. 6 is the schematic diagram of branching networks according to an embodiment of the present invention；
Fig. 7 is the effect diagram in removal shadow interference stage according to an embodiment of the present invention；
Fig. 8 is a kind of detection method flow chart of optional high score Remote Sensing Imagery Change according to an embodiment of the present invention；
Fig. 9 is a kind of schematic diagram of the detection system of high score Remote Sensing Imagery Change according to an embodiment of the present invention.
Specific embodiment
In order to enable those skilled in the art to better understand the solution of the present invention, below in conjunction in the embodiment of the present invention
Attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is only
The embodiment of a part of the invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill people
The model that the present invention protects all should belong in member's every other embodiment obtained without making creative work
It encloses.
It should be noted that description and claims of this specification and term " first " in above-mentioned attached drawing, "
Two " etc. be to be used to distinguish similar objects, without being used to describe a particular order or precedence order.It should be understood that using in this way
Data be interchangeable under appropriate circumstances, so as to the embodiment of the present invention described herein can in addition to illustrating herein or
Sequence other than those of description is implemented.In addition, term " includes " and " having " and their any deformation, it is intended that cover
Cover it is non-exclusive include, for example, the process, method, system, product or equipment for containing a series of steps or units are not necessarily limited to
Step or unit those of is clearly listed, but may include be not clearly listed or for these process, methods, product
Or other step or units that equipment is intrinsic.
According to embodiments of the present invention, a kind of embodiment of the detection method of high score Remote Sensing Imagery Change is provided, is needed
Bright, step shown in the flowchart of the accompanying drawings can be held in a computer system such as a set of computer executable instructions
Row, although also, logical order is shown in flow charts, and it in some cases, can be to be different from sequence herein
Execute shown or described step.
Fig. 1 is the detection method of high score Remote Sensing Imagery Change according to an embodiment of the present invention, as shown in Figure 1, this method packet
Include following steps:
Step S102, obtain two when phase images in each pixel neighborhood characteristics；
Neighborhood characteristics input branch's convolutional neural networks model is obtained the classification of image, wherein classification is by step S104
Change class and does not change class；Branch's convolutional neural networks model is to carry out unsupervised pre-training to image to obtain, branch's volume
Product neural network model includes input layer, branching networks, full articulamentum and classification layer, and branching networks are for making neighborhood characteristics
For label and by neighborhood characteristics convolution dimensionality reduction, layer of classifying is different classes of for dividing the image into.
It is detected by deep learning network change (training branch's convolutional neural networks model) by two phase high score remote sensing shadows
It as the neighborhood characteristics for each pixel extracted, is input in branch's convolutional neural networks, first branching networks is carried out unsupervised
Pre-training completes the network parameter initialization of the two forked sections (branching networks).Cooperate neighborhood characteristics as input again
The variation label manually marked carries out Training.Unlike the prior art, the prior art is first to extract the neighbour of object pixel
Then neighborhood characteristics are shown laid flat in feature vector by characteristic of field, then splice the feature vector of two phases.And the present embodiment
It, can be simultaneously from two phase images directly using the neighborhood characteristics of object pixel as the defeated of network using the structure of branching networks
Out, a large amount of map function is avoided, the prior art technical problem not high to Image Change Detection precision is solved, improves
Detection efficiency.
Branch's convolutional neural networks in above-mentioned steps can make the detection high score remote sensing of the present invention more efficiently and accurately
The variation of image, the structure of branch's convolutional neural networks also unlike the prior art, the present embodiment using classification layer instead of
Output layer in the prior art obtains if directly using the output layer and loss function of the twin network of the prior art
As a result and a gray level image, the gray scale of the gray level image every bit indicate the similarity degree of two phase image respective pixels,
Being difficult to define gray value to be greater than how many is variation, and it is unchanged that gray value, which is less than how many,.And the present embodiment uses and divides
The feature of two phase respective pixels can be divided into variation class and not change two classifications of class by class device, and the image obtained in this way is just
It is in white and black bianry image, there is practical value.
In a kind of optional embodiment, neighborhood characteristics input branch's convolutional neural networks model is obtained into the class of image
It does not include: that neighborhood characteristics input branching networks are obtained into convolution feature；Dimensionality reduction is carried out to convolution feature by full articulamentum to obtain
Dimensionality reduction feature；The dimensionality reduction feature is inputted into classification layer and obtains the classification of image.
Branching networks in above-mentioned steps are different from the branching networks in twin network in the prior art, the present embodiment
Branching networks use first convolution, then the structure of deconvolution, the advantage of doing so is that can the training stage with input picture from
Body is as label, and by the image elder generation convolution dimensionality reduction inputted, and dimensionality reduction can lose the information of part, is risen and is tieed up by deconvolution again, benefit
The output obtained with lossy residual, information keeps output result and input picture even more like as a result, passing through constantly training, leads to
Crossing above-mentioned training first the parameter to branching networks can complete initialization, use manual tag training whole network so as to subsequent
When, it can more rapid convergence, faster completion.
In a kind of optional embodiment, the neighborhood characteristics of each pixel include: before logical in phase images when obtaining two
It crosses Histogram Matching algorithm and is pre-processed the image after being corrected to image, wherein image and other images after correction
Tone it is consistent.
By using Histogram Matching algorithm on the basis of two phase images are registered and absolute radiation correction into
Line number Data preprocess weakens so that the image-forming condition of two phase Image Change Detections interferes.
It needs to carry out feature extraction after pre-processing image the image after being corrected, in feature extraction phases
Extraction feature using the Neighborhood matrix feature of object pixel as current pixel.If the coordinate of object pixel is (i, j) adjacent
Domain matrix is all pixels that (i+n, j+n) is arrived from (i-n, j-n), and wherein n is natural number, by experiment, optimal n value
It is 2, i.e., 5 × 5 adjacent pixel, right if object pixel is located at high score Remote Sensing Image Edge position around extraction object pixel
Zero padding is taken to handle in edge.In a kind of optional embodiment, mentioned using 5 × 5 Neighborhood matrixes of object pixel as feature
The standard taken carries out feature extraction to the image after correction.
In order to increase the generalization of network model, need after to branch's convolutional neural networks model training one time by branch
Two inputs (two phase image neighborhood characteristics) of convolutional neural networks change over, and retraining is primary.In a kind of optional reality
It applies in mode, the position of the neighborhood characteristics of phase images when two is exchanged, the neighborhood characteristics after exchanging input branch's convolution
Model of neural network retraining obtains new branch's convolutional neural networks model.
Construction variation sample imbalance feelings more diverse than dismounting in training sample can be so offset through the above steps
Condition.
The embodiment of the present invention obtains neighborhood characteristics input branch's convolutional neural networks model after the classification of image can be with
Shadow interference is removed, the higher buildings object shade length as caused by two phase shooting times are different is different and what is generated retouch side effect
Answer, produce a large amount of pseudo- variation, in order to remove it is this retouch side effect, in a kind of optional embodiment, pass through principal component
Brightness, saturation degree and the tone value component computational shadowgraph index of analytic approach and each phase, wherein the high local table of shadow index
Show that shaded area, the low place of shadow index indicate not to be shaded area；Divide shadow index figure by maximum variance between clusters
As obtaining shadow image, wherein shadow image is bianry image；Shadow image is subtracted in the image of variation to be detected to be gone
Except the image of shadow interference.
The process of above-mentioned removal shade is illustrated with an optional embodiment below:
The single channel principal component analysis result of each phase image is extracted with Principle components analysis first.Then by two phases
Image is converted to HIS color model, computational shadowgraph index from RGB color model
Wherein LPCA is single channel principal component analysis as a result, I and S are HIS color model I channel components and channel S component.
Binarization threshold segmentation is carried out to shadow index finally by maximum variance between clusters, the big pixel of shadow index is direct-shadow image
Element.The shadows pixels of two phases are extracted, and ignore the variation occurred in these shadows pixels in tentatively variation testing result, i.e.,
Removal shadow interference be can get as a result, to improve detection accuracy.
Subtracting in the image of variation to be detected can be to image after shadow image obtains the image of removal shadow interference
It is denoised, by the image segmentation for removing shadow interference is more by mean shift algorithm in an optional embodiment
A subgraph, wherein each subgraph is same colour type；Variation pixel accounts for often in the image of statistics removal shadow interference
The ratio of the total pixel number of a colour type；When ratio is more than preset threshold, then all pixels of current class are variation pixel,
If ratio is less than preset threshold, all pixels of current class are not change pixel.
Above-mentioned denoising process is described in detail with an optional embodiment below:
First using phase images when mean shift algorithm segmentation two, the pixel in regulation color radius and space radius is gathered
For one kind, then all pixels in this kind are filled again with the RGB average gray of current class pixel, that is, are divided
The identical each piece of closing figure spot of color represents a classification in image after cutting.After dividing the image into, previous step is counted
Removal shadow interference result in variation pixel account for the ratio of each classification total pixel number, if ratio is more than a certain threshold value, that is, recognize
All pixels for current class are variation pixel, if not exceeded, then thinking that all pixels of current class are not change
Pixel.Enhance algorithm by above-mentioned denoising respectively to handle phase images when two, and the processing result of two phases taken into union,
Finally obtain the final result figure of denoising enhancing.For example, thering is the square area of a 5*5 size to be recognized by image segmentation
To be a classification, the total pixel number of this classification is 5*5=25, but may only have 13 pixels to be detected in this square
It is variation out, is indicated in black and white binary image with white.If defined threshold is 50% at this time, 13 are greater than 25*50%
=12.5, so that it may think all pixels in this classification all and be variation pixel, it all should be with white in black and white binary image
Color table shows, at this time can be by incomplete detection completion.If defined threshold is 60%, then 13 are less than 25*60%=15, just
Think that all pixels are all not change pixel in this classification, should all be indicated with black.Just that 13 are detected
White pixel is considered noise data, it should blacking.The value range of the threshold value is adjusted according to the actual situation.Threshold value is chosen
It can get relatively good denoising effect 60% to 90%, but different picture features is different, in practical applications, Ke Yixuan
It takes the sub-fraction for needing to change detection zone first to be tested, determines that is suitable for an optimal threshold for current locale, then answer
Use whole distract.
The noise jamming for being difficult to remove in the prior art is eliminated by the above method, keeps result more accurate.
The above method belongs to the change detecting method of pixel-oriented, as shown in Fig. 2, counting first to high score remote sensing images
Then Data preprocess carries out feature extraction for each pixel, next using deep learning algorithm to two phase respective pixels
Feature carry out classification and complete variation detection, obtained result is removed shadow interference, but the denoising enhancing stage due to
The image segmentation thought for having used for reference object-oriented change detecting method, obtains the boundary information of each ground class, is finally become
Change testing result.By this method can two width to areal difference phase or multiple image analyze, detect it
In changing unit and non-changing unit.Specifically, the present embodiment is two phase images are registered and absolute radiation correction
On the basis of used Histogram Matching algorithm so that two phase remote sensing imagery change detections image-forming condition interfere weaken.Then it carries out
The experiment of feature extraction has determined optimal feature extraction strategy, and the neighborhood characteristics extracted and manual tag are input to point
It is trained in branch convolutional neural networks.Two phase remote sensing images can be tested to obtain just after obtaining trained model
Step variation testing result retouches side effect by using principal component analysis and the conversion computational shadowgraph index removal of HIS color model.Most
The denoising enhancing algorithm for having used the present embodiment to propose afterwards handles result, obtains final variation testing result.
The variation detection process of above-mentioned whole image is illustrated with a complete embodiment below:
It in data preprocessing phase high score remote sensing image is shot when around earth rotation by high score satellite, due to defending
Star is long operation cycle, and cloud cover situation when operation to areal overhead is different, so usually needing nearest difference
Period shooting synthesizes an image without cloud cover image joint, and the present embodiment is referred to as the high score remote sensing shadow an of phase
Picture.The image of different phases can be usually very different due to the difference in imaging season, imaging angle and Image registration.
This be unfavorable for very much pixel-oriented variation detection, in order to mitigate this species diversity bring influence, current embodiment require that image into
Row pretreatment.
As shown in figure 3, since No. two remote sensing images of high score that the present embodiment uses have already been through registration and absolute radiation
Correction, it is possible to find out that the image pixel of areal corresponds to substantially, but since imaging season difference causes the colour temperature of image
Difference, the present embodiment use Histogram Matching algorithm and are handled, and make the tone of piece image Yu another width image to be matched
It is consistent as far as possible, eliminates and adversely affected caused by image-forming condition difference.Analyze the image the results show that after Histogram Matching
With the raising for having biggish matter before image rectification, illustrate that it is relatively good for carrying out relative detector calibration effect using this method
's.
In feature extraction phases, since the present embodiment is that pixel-oriented is changed detection, so by variation detection conversion
Respective pixel feature is divided into variation class and does not change class two by the problem of carrying out two classification for the feature to two phase pixels
Classification.Feature extraction is carried out using the Neighborhood matrix feature of object pixel.Neighborhood matrix is characterized in referring to n × n around object pixel
The region of size, this method tested 3 × 3,5 × 5,7 × 7, sparse 5 × 5 Neighborhood matrix as feature to variation testing result
Influence, analysis result it is as follows.The reconstruct loss in pre-training stage can be made smaller using 3 × 3 Neighborhood matrixes, show 3 × 3 neighbours
Domain matrix information content is small, it is easier to be characterized.But 3 × 3 Neighborhood matrixes are simultaneously also more like the processing of traditional Change vector Analysis method
Result images, this shows that 3 × 3 neighborhood information can not fully consider the context relation around pixel, i.e., 3 × 3 neighbour
Domain matrix size or small.In addition to this 3 × 3 also have a small amount of noise, to sum up, 3 × 3 than 5 × 5 training that behave oneself best and
Faster (data volume is small), but false detection rate is higher for the speed of forecast period.It is available best using 5 × 5 Neighborhood matrixes as a result,
After having fully considered context relation, shared memory headroom is smaller when avoiding the erroneous detection of large area, while pre-processing, training
Speed is moderate.Reconstruct the largest loss using 7 × 7 Neighborhood matrixes in the pre-training stage, reason are exactly the opposite with 3 × 3.Use 7
× 7 Neighborhood matrixes extremely consume memory in Data Preparation Process, and processing speed is very slow.It can be using 7 × 7 Neighborhood matrixes
It promotes the promotion for slightly inhibiting false detection rate while variation detects positive inspection rate, especially water area changes and whole details is promoted, but
Consider that data volume is huge, it is higher to promote details cost.
In order to more fully consider wider neighborhood information, and avoid increasing the data volume increase bring calculating time
Increase and occupy the problem of memory space increases, the present embodiment has attempted sparse 5 × 5 Neighborhood matrix.Sparse 5 × 5 Neighborhood matrix
Big region as 7 × 7 Neighborhood matrixes can be covered, but only with the memory space of 5 × 5 Neighborhood matrixes.Use sparse 5 × 5
Simultaneously indifference, operation result also indicate that sparse 5 × 5 matrix to Neighborhood matrix with 5 × 5 matrixes on memory consumption and processing speed
Final effect is between 3 × 3 and 5 × 5 matrixes.Comprehensively consider hardware spending etc. of variation detection effect and whole flow process because
Element finally has chosen standard of 5 × 5 Neighborhood matrixes as final feature extraction.Extract process such as Fig. 4 institute of Neighborhood matrix feature
Show.
Pass through the neighborhood spy for each pixel that two phase high score remote sensing images are extracted in the detection of deep learning network change
Sign, is input in branch's convolutional neural networks, and branch's convolutional neural networks include 4 parts, i.e. input layer, branching networks, Quan Lian
Connect layer and classification layer.Wherein input layer is the dimension for the neighborhood characteristics extracted, and classification layer uses softmax classifier, tool
Volume grid structure is as shown in Figure 5.The structure of branching networks is the self-encoding encoder based on convolutional neural networks in fact, works as branch
When network receives input picture, convolution is 1. first carried out using convolutional layer, convolution nuclear volume is 64, and convolution mode is same；Then
Without pond layer, then carry out convolutional layer 2., convolution nuclear volume is 32, and convolution mode is same；Over-fitting in order to prevent, connects down
To introduce Dropout layers；It is then input to warp lamination 1., convolution nuclear volume is 32, and convolution mode is same；Next warp
Cross warp lamination 2., convolution nuclear volume is 64, and convolution mode is same；Finally be linked into output layer, the data dimension of output with
The data of input are consistent, specific branching networks structure such as Fig. 6.In the training process, first input picture is input to point
In branch network, unsupervised pre-training is carried out using input picture itself as label, so that the weight of two branching networks carries out initially
Change (training is slow, can not train classification layer)；Then the output by branching networks convolutional layer 2., flattening are input to the full connection of Fig. 4
In layer, then classifier is trained using manual tag, finally obtains trained network model.Network model can pass through
The convolution feature in Neighborhood matrix image block is automatically extracted, is classified using these more abstract convolution features.Wherein it is
The generalization for increasing network model, needs to exchange two phase image neighborhood characteristics, retraining is primary.Instruction can so be offset
Practice construction variation sample imbalance situation more diverse than dismounting in sample.The network model that training is completed can be to each pixel
Classify, be divided into variation class and do not change class, obtains preliminary variation testing result.
The prediction result of deep learning network is not final result, it is also necessary to by the perfect and enhancing of algorithm.Into
Row Urban Areas variation detection during the present embodiment find, Urban Areas variation it is less, but due to downtown areas building compared with
It is more, so the shadow length that each phase building is shined upon generation is different, in this way caused by shade change it is very more.It is this
The pseudo- variation of the generations such as building surrounding shadow, the present embodiment are referred to as " retouching side effect ".
Side effect is retouched when causing variation detection as shown in fig. 7, the shadow in three buildings is different in size in phasor when two.For
Mitigation is this to retouch side effect, and the present embodiment has used principal component analysis (Principal Component Analysis, PCA)
It is converted into HIS color model with RGB, and the method for computational shadowgraph index (Shadow Index, SI) is extracted and removes shade.It is main
Constituent analysis is a kind of dimension reduction method, can be the RGB triple channel image dimensionality reduction of the present embodiment at single channel gray level image, this reality
Applying example, treated that image is denoted as LPCA by principal component analysis, he can ignore the present embodiment during extracting shade and not feel
The colouring information of interest, and retain shadow information as much as possible.HIS color model is as RGB color model, and one kind
The color model being widely adopted.Unlike RGB color model, RGB color model is with red-green-blue channel ash
Angle value characterizes each pixel, and HIS characterizes each picture using three brightness (I), saturation degree (S) and tone value (H) components
Element.It is converted into HIS color model by RGB, the present embodiment has obtained brightness, saturation degree and the tone value component of each phase.
Above-mentioned data can be used in last the present embodiment, computational shadowgraph index SI:
The high place of shadow index is considered as shaded area, and the low place of shadow index then indicates it is not shade
Area.Divide shadow index image, available binary map, the shadow image as extracted using maximum variance between clusters.?
These shades extracted are subtracted in the modified-image of prediction, as remove the final result for retouching side effect.
It is still flooded with much noise in the present embodiment discovery variation testing result after removing the interference of shade, and
The accuracy compared with original image of the marginal information of variation testing result is not too high.If being denoised using simple median filtering,
A large amount of boundary information can be then lost, denoising result edge is allowed to become round and smooth, in order to make final variation testing result purer
Only, also for that can obtain more accurate boundary information, the present embodiment proposes a kind of denoising enhancing algorithm.Denoising enhancing
Phase images when algorithm uses average drifting (Mean Shift) algorithm to divide two first.Mean shift algorithm can will provide color
Pixel in radius and space radius is gathered for one kind, then by all pixels in this kind again with current class pixel
RGB average gray is filled, that is, the identical each piece of closing figure spot of color represents a classification in the image after dividing.It will
After image segmentation, the ratio that variation pixel in the variation detection prediction result of previous step accounts for each classification total pixel number is counted
Value thinks that all pixels of current class are variation pixel, if not exceeded, then thinking to work as if ratio is more than a certain threshold value
The all pixels of preceding classification are not change pixel.By this method, tiny noise pixel can be removed, can also be incited somebody to action
The incomplete part completion of boundary information in original prediction result, so that the accuracy rate of final result is further promoted, Fig. 8 is shown
The effect image of denoising enhancing algorithm and final variation testing result.
The embodiment of the invention also provides a kind of high score Remote Sensing Imagery Change Detection system, which can pass through mould of classifying
Block realizes its function.It should be noted that a kind of high score Remote Sensing Imagery Change Detection system of the embodiment of the present invention can be used for
A kind of high score method for detecting change of remote sensing image provided by the embodiment of the present invention is executed, a kind of high score of the embodiment of the present invention is distant
Feel image change detection method can also through the embodiment of the present invention provided by a kind of high score Remote Sensing Imagery Change Detection system
To execute.Fig. 9 is a kind of schematic diagram of high score Remote Sensing Imagery Change Detection system according to an embodiment of the present invention.As shown in figure 9,
A kind of high score Remote Sensing Imagery Change Detection system includes:
Obtain module 92, when for obtaining two in phase images each pixel neighborhood characteristics；
Categorization module 94, for neighborhood characteristics input branch's convolutional neural networks model to be obtained the classification of image, wherein
Classification is variation class and does not change class；Branch's convolutional neural networks model is to carry out nothing to image by branch's convolutional neural networks
Supervision pre-training obtains, and branch's convolutional neural networks include input layer, branching networks, full articulamentum and classification layer, branched network
Network is for using neighborhood characteristics, as label and by neighborhood characteristics convolution dimensionality reduction, layer of classifying to be for dividing the image into inhomogeneity
It is other.
A kind of above-mentioned high score Remote Sensing Imagery Change Detection system embodiment is and a kind of high score Remote Sensing Imagery Change Detection side
Method is corresponding, so repeating no more for beneficial effect.
The serial number of the above embodiments of the invention is only for description, does not represent the advantages or disadvantages of the embodiments.
In the above embodiment of the invention, it all emphasizes particularly on different fields to the description of each embodiment, does not have in some embodiment
The part of detailed description, reference can be made to the related descriptions of other embodiments.
In several embodiments provided herein, it should be understood that disclosed technology contents can pass through others
Mode is realized.Wherein, the apparatus embodiments described above are merely exemplary, such as the division of the unit, Ke Yiwei
A kind of logical function partition, there may be another division manner in actual implementation, for example, multiple units or components can combine or
Person is desirably integrated into another system, or some features can be ignored or not executed.Another point, shown or discussed is mutual
Between coupling, direct-coupling or communication connection can be through some interfaces, the INDIRECT COUPLING or communication link of unit or module
It connects, can be electrical or other forms.
The unit as illustrated by the separation member may or may not be physically separated, aobvious as unit
The component shown may or may not be physical unit, it can and it is in one place, or may be distributed over multiple
On unit.It can some or all of the units may be selected to achieve the purpose of the solution of this embodiment according to the actual needs.
It, can also be in addition, the functional units in various embodiments of the present invention may be integrated into one processing unit
It is that each unit physically exists alone, can also be integrated in one unit with two or more units.Above-mentioned integrated list
Member both can take the form of hardware realization, can also realize in the form of software functional units.
If the integrated unit is realized in the form of SFU software functional unit and sells or use as independent product
When, it can store in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially
The all or part of the part that contributes to existing technology or the technical solution can be in the form of software products in other words
It embodies, which is stored in a storage medium, including some instructions are used so that a computer
Equipment (can for personal computer, server or network equipment etc.) execute each embodiment the method for the present invention whole or
Part steps.And storage medium above-mentioned includes: that USB flash disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
Reservoir (RAM, Random Access Memory), mobile hard disk, magnetic or disk etc. be various to can store program code
Medium.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (8)
1. a kind of detection method of high score Remote Sensing Imagery Change characterized by comprising
When obtaining two in phase images each pixel neighborhood characteristics；
Neighborhood characteristics input branch's convolutional neural networks model is obtained into the classification of described image, wherein the classification is
Variation class does not change class；Branch's convolutional neural networks model is to carry out unsupervised pre-training to described image to obtain,
Branch's convolutional neural networks model includes input layer, branching networks, full articulamentum and classification layer, and the branching networks are to use
In using the neighborhood characteristics, as label and by the neighborhood characteristics convolution dimensionality reduction, the classification layer is for by described image
It is divided into different classes of.
2. the method according to claim 1, wherein the neighborhood characteristics are inputted branch's convolutional neural networks mould
The classification that type obtains described image includes:
The neighborhood characteristics are inputted into the branching networks and obtain convolution feature；
Dimensionality reduction is carried out to the convolution feature by full articulamentum and obtains dimensionality reduction feature；
The dimensionality reduction feature is inputted into the classification layer and obtains the classification of described image.
3. the method according to claim 1, wherein obtain two when phase images in each pixel neighborhood characteristics it
Before include:
The image after being corrected is pre-processed to described image by Histogram Matching algorithm, wherein after the correction
Image is consistent with the tone of other images.
4. according to the method described in claim 3, it is characterized in that, when taking two in phase images each pixel neighborhood characteristics it
Before, include: after being pre-processed the image after being corrected to described image
Feature is carried out to the image after the correction as the standard of feature extraction using 5 × 5 Neighborhood matrixes of object pixel to mention
It takes.
5. the method according to claim 1, wherein branch's convolutional neural networks model is after training,
The position of the neighborhood characteristics of phase images when two is exchanged, the neighborhood characteristics after exchanging input branch's convolutional Neural net
Model of network retraining obtains new branch's convolutional neural networks model.
6. the method according to claim 1, wherein the neighborhood characteristics are inputted branch's convolutional neural networks mould
Type obtains
Pass through the brightness of Principal Component Analysis and each phase, saturation degree and tone value component computational shadowgraph index, wherein described
The high place of shadow index indicates that shaded area, the low place of the shadow index indicate not to be shaded area；
Divide shadow index image by maximum variance between clusters and obtain shadow image, wherein the shadow image is binary map
Picture；
The shadow image is subtracted in the image of variation to be detected obtains the image of removal shadow interference.
7. according to the method described in claim 6, it is characterized in that, subtracting the shadow image in the image of variation to be detected
Obtain removal shadow interference image after, comprising:
By mean shift algorithm by it is described removal shadow interference image segmentation be multiple subgraphs, wherein each subgraph
For same colour type；
Count the ratio that variation pixel in the image of the removal shadow interference accounts for the total pixel number of each colour type；
When the ratio is more than preset threshold, then all pixels of current class are variation pixel, if the ratio is less than institute
Preset threshold is stated, then all pixels of current class are not change pixel.
8. a kind of high score Remote Sensing Imagery Change Detection system characterized by comprising
Obtain module, when for obtaining two in phase images each pixel neighborhood characteristics；
Categorization module, for neighborhood characteristics input branch's convolutional neural networks model to be obtained the classification of described image,
In, the classification is variation class or does not change class；Branch's convolutional neural networks model is unsupervised to described image progress
What pre-training obtained, branch's convolutional neural networks model includes input layer, branching networks, full articulamentum and classification layer, institute
Stating branching networks is for using the neighborhood characteristics, as label and by the neighborhood characteristics convolution dimensionality reduction, the classification layer to be
It is different classes of for described image to be divided into.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811316602.2A CN109558806B (en) | 2018-11-07 | 2018-11-07 | Method for detecting high-resolution remote sensing image change |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811316602.2A CN109558806B (en) | 2018-11-07 | 2018-11-07 | Method for detecting high-resolution remote sensing image change |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109558806A true CN109558806A (en) | 2019-04-02 |
| CN109558806B CN109558806B (en) | 2021-09-14 |
Family
ID=65865987
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811316602.2A Active CN109558806B (en) | 2018-11-07 | 2018-11-07 | Method for detecting high-resolution remote sensing image change |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109558806B (en) |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163294A (en) * | 2019-05-29 | 2019-08-23 | 广东工业大学 | Remote Sensing Imagery Change method for detecting area based on dimensionality reduction operation and convolutional network |
| CN110298366A (en) * | 2019-07-05 | 2019-10-01 | 北华航天工业学院 | Crops are distributed extracting method and device |
| CN110378224A (en) * | 2019-06-14 | 2019-10-25 | 香港理工大学深圳研究院 | A kind of detection method of feature changes, detection system and terminal |
| CN110969088A (en) * | 2019-11-01 | 2020-04-07 | 华东师范大学 | Remote sensing image change detection method based on significance detection and depth twin neural network |
| CN111274892A (en) * | 2020-01-14 | 2020-06-12 | 北京科技大学 | Robust remote sensing image change detection method and system |
| CN111597377A (en) * | 2020-04-08 | 2020-08-28 | 广东省国土资源测绘院 | Deep learning technology-based field investigation method and system |
| CN111626409A (en) * | 2020-07-30 | 2020-09-04 | 江西高创保安服务技术有限公司 | Data generation method for image quality detection |
| CN113449690A (en) * | 2021-07-21 | 2021-09-28 | 华雁智科(杭州)信息技术有限公司 | Method and system for detecting image scene change and electronic equipment |
| CN113505256A (en) * | 2021-07-02 | 2021-10-15 | 北京达佳互联信息技术有限公司 | Feature extraction network training method, image processing method and device |
| TWI752478B (en) * | 2020-04-27 | 2022-01-11 | 台達電子工業股份有限公司 | Image processing method and image processing system |
| CN113989652A (en) * | 2021-12-27 | 2022-01-28 | 中国测绘科学研究院 | Method and system for detecting farmland change under layered multiple judgment rules |
| CN115294461A (en) * | 2022-10-10 | 2022-11-04 | 中国电建集团山东电力建设第一工程有限公司 | Power facility collapse assessment method and system based on BIM and remote sensing image |
| US11715283B2 (en) | 2020-04-27 | 2023-08-01 | Delta Electronics, Inc. | Image processing method and image processing system |
Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850837A (en) * | 2015-05-18 | 2015-08-19 | 西南交通大学 | Handwritten character recognition method |
| CN105608698A (en) * | 2015-12-25 | 2016-05-25 | 西北工业大学 | Remote image change detection method based on SAE |
| CN106023154A (en) * | 2016-05-09 | 2016-10-12 | 西北工业大学 | Multi-temporal SAR image change detection method based on dual-channel convolutional neural network (CNN) |
| CN106326925A (en) * | 2016-08-23 | 2017-01-11 | 南京邮电大学 | Apple disease image identification method based on deep learning network |
| CN106651886A (en) * | 2017-01-03 | 2017-05-10 | 北京工业大学 | Cloud image segmentation method based on superpixel clustering optimization CNN |
| CN107122796A (en) * | 2017-04-01 | 2017-09-01 | 中国科学院空间应用工程与技术中心 | A kind of remote sensing image sorting technique based on multiple-limb network integration model |
| CN107180248A (en) * | 2017-06-12 | 2017-09-19 | 桂林电子科技大学 | Strengthen the hyperspectral image classification method of network based on associated losses |
| CN107229904A (en) * | 2017-04-24 | 2017-10-03 | 东北大学 | A kind of object detection and recognition method based on deep learning |
| CN107240076A (en) * | 2017-05-31 | 2017-10-10 | 北京小米移动软件有限公司 | Image processing method and device |
| CN107316061A (en) * | 2017-06-22 | 2017-11-03 | 华南理工大学 | A kind of uneven classification ensemble method of depth migration study |
| US20180061046A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Skin lesion segmentation using deep convolution networks guided by local unsupervised learning |
| CN107944353A (en) * | 2017-11-10 | 2018-04-20 | 西安电子科技大学 | SAR image change detection based on profile ripple BSPP networks |
| CN108133173A (en) * | 2017-11-24 | 2018-06-08 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on semi-supervised ladder network |
| CN108447057A (en) * | 2018-04-02 | 2018-08-24 | 西安电子科技大学 | SAR image change detection based on conspicuousness and depth convolutional network |
| CN108573276A (en) * | 2018-03-12 | 2018-09-25 | 浙江大学 | A kind of change detecting method based on high-resolution remote sensing image |
| CN108596108A (en) * | 2018-04-26 | 2018-09-28 | 中国科学院电子学研究所 | Method for detecting change of remote sensing image of taking photo by plane based on the study of triple semantic relation |
-
2018
- 2018-11-07 CN CN201811316602.2A patent/CN109558806B/en active Active
Patent Citations (16)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850837A (en) * | 2015-05-18 | 2015-08-19 | 西南交通大学 | Handwritten character recognition method |
| CN105608698A (en) * | 2015-12-25 | 2016-05-25 | 西北工业大学 | Remote image change detection method based on SAE |
| CN106023154A (en) * | 2016-05-09 | 2016-10-12 | 西北工业大学 | Multi-temporal SAR image change detection method based on dual-channel convolutional neural network (CNN) |
| CN106326925A (en) * | 2016-08-23 | 2017-01-11 | 南京邮电大学 | Apple disease image identification method based on deep learning network |
| US20180061046A1 (en) * | 2016-08-31 | 2018-03-01 | International Business Machines Corporation | Skin lesion segmentation using deep convolution networks guided by local unsupervised learning |
| CN106651886A (en) * | 2017-01-03 | 2017-05-10 | 北京工业大学 | Cloud image segmentation method based on superpixel clustering optimization CNN |
| CN107122796A (en) * | 2017-04-01 | 2017-09-01 | 中国科学院空间应用工程与技术中心 | A kind of remote sensing image sorting technique based on multiple-limb network integration model |
| CN107229904A (en) * | 2017-04-24 | 2017-10-03 | 东北大学 | A kind of object detection and recognition method based on deep learning |
| CN107240076A (en) * | 2017-05-31 | 2017-10-10 | 北京小米移动软件有限公司 | Image processing method and device |
| CN107180248A (en) * | 2017-06-12 | 2017-09-19 | 桂林电子科技大学 | Strengthen the hyperspectral image classification method of network based on associated losses |
| CN107316061A (en) * | 2017-06-22 | 2017-11-03 | 华南理工大学 | A kind of uneven classification ensemble method of depth migration study |
| CN107944353A (en) * | 2017-11-10 | 2018-04-20 | 西安电子科技大学 | SAR image change detection based on profile ripple BSPP networks |
| CN108133173A (en) * | 2017-11-24 | 2018-06-08 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on semi-supervised ladder network |
| CN108573276A (en) * | 2018-03-12 | 2018-09-25 | 浙江大学 | A kind of change detecting method based on high-resolution remote sensing image |
| CN108447057A (en) * | 2018-04-02 | 2018-08-24 | 西安电子科技大学 | SAR image change detection based on conspicuousness and depth convolutional network |
| CN108596108A (en) * | 2018-04-26 | 2018-09-28 | 中国科学院电子学研究所 | Method for detecting change of remote sensing image of taking photo by plane based on the study of triple semantic relation |
Non-Patent Citations (5)
| Title |
|---|
| FEI GAO 等: "dual-branch deep convolution neural network for polarimetric SAR image classification", 《APPLIED SCIENCES》 * |
| 刘本强 等: "一种利用邻域相对熵的ＳＡＲ影像变化检测方法", 《遥感信息》 * |
| 刘辉 等: "基于 ＰＣＡ与 ＨＩＳ模型的高分辨率遥感影像阴影检测研究", 《遥感技术与应用》 * |
| 徐丽华: "顾及投影差的遥感影像变化检测", 《中国优秀硕士学位论文全文数据库 基础科学辑》 * |
| 蔡晓东 等: "基于多分支卷积神经网络的车辆图像比对方法", 《视频应用与工程》 * |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110163294A (en) * | 2019-05-29 | 2019-08-23 | 广东工业大学 | Remote Sensing Imagery Change method for detecting area based on dimensionality reduction operation and convolutional network |
| CN110378224B (en) * | 2019-06-14 | 2021-01-05 | 香港理工大学深圳研究院 | Detection method and detection system for ground feature change and terminal |
| CN110378224A (en) * | 2019-06-14 | 2019-10-25 | 香港理工大学深圳研究院 | A kind of detection method of feature changes, detection system and terminal |
| CN110298366A (en) * | 2019-07-05 | 2019-10-01 | 北华航天工业学院 | Crops are distributed extracting method and device |
| CN110298366B (en) * | 2019-07-05 | 2021-05-04 | 北华航天工业学院 | Crop distribution extraction method and device |
| CN110969088A (en) * | 2019-11-01 | 2020-04-07 | 华东师范大学 | Remote sensing image change detection method based on significance detection and depth twin neural network |
| CN111274892A (en) * | 2020-01-14 | 2020-06-12 | 北京科技大学 | Robust remote sensing image change detection method and system |
| CN111597377A (en) * | 2020-04-08 | 2020-08-28 | 广东省国土资源测绘院 | Deep learning technology-based field investigation method and system |
| TWI752478B (en) * | 2020-04-27 | 2022-01-11 | 台達電子工業股份有限公司 | Image processing method and image processing system |
| US11715283B2 (en) | 2020-04-27 | 2023-08-01 | Delta Electronics, Inc. | Image processing method and image processing system |
| CN111626409A (en) * | 2020-07-30 | 2020-09-04 | 江西高创保安服务技术有限公司 | Data generation method for image quality detection |
| CN113505256A (en) * | 2021-07-02 | 2021-10-15 | 北京达佳互联信息技术有限公司 | Feature extraction network training method, image processing method and device |
| CN113449690A (en) * | 2021-07-21 | 2021-09-28 | 华雁智科(杭州)信息技术有限公司 | Method and system for detecting image scene change and electronic equipment |
| CN113989652A (en) * | 2021-12-27 | 2022-01-28 | 中国测绘科学研究院 | Method and system for detecting farmland change under layered multiple judgment rules |
| CN113989652B (en) * | 2021-12-27 | 2022-04-26 | 中国测绘科学研究院 | Method and system for detecting farmland change under layered multiple judgment rules |
| CN115294461A (en) * | 2022-10-10 | 2022-11-04 | 中国电建集团山东电力建设第一工程有限公司 | Power facility collapse assessment method and system based on BIM and remote sensing image |
| CN115294461B (en) * | 2022-10-10 | 2023-01-31 | 中国电建集团山东电力建设第一工程有限公司 | Power facility collapse and inclination assessment method and system based on BIM and remote sensing image |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109558806B (en) | 2021-09-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109558806A (en) | The detection method and system of high score Remote Sensing Imagery Change | |
| CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
| Huang et al. | Automatic building extraction from high-resolution aerial images and LiDAR data using gated residual refinement network | |
| Li et al. | Multiscale features supported DeepLabV3+ optimization scheme for accurate water semantic segmentation | |
| Rachmadi et al. | Vehicle color recognition using convolutional neural network | |
| CN104050471B (en) | Natural scene character detection method and system | |
| CN103049763B (en) | Context-constraint-based target identification method | |
| CN107133943A (en) | A kind of visible detection method of stockbridge damper defects detection | |
| CN108537239B (en) | Method for detecting image saliency target | |
| CN111797712B (en) | Remote sensing image cloud and cloud shadow detection method based on multi-scale feature fusion network | |
| CN110084108A (en) | Pedestrian re-identification system and method based on GAN neural network | |
| CN112861690B (en) | Multi-method fused remote sensing image change detection method and system | |
| CN111832443B (en) | Construction method and application of construction violation detection model | |
| CN108960404B (en) | Image-based crowd counting method and device | |
| Zuo et al. | HF-FCN: Hierarchically fused fully convolutional network for robust building extraction | |
| Zhang et al. | CNN cloud detection algorithm based on channel and spatial attention and probabilistic upsampling for remote sensing image | |
| CN107545571A (en) | A kind of image detecting method and device | |
| CN111783523A (en) | Remote sensing image rotating target detection method | |
| CN109034184A (en) | A kind of grading ring detection recognition method based on deep learning | |
| CN105931241A (en) | Automatic marking method for natural scene image | |
| Chen et al. | Object-based multi-modal convolution neural networks for building extraction using panchromatic and multispectral imagery | |
| CN111583276A (en) | CGAN-based space target ISAR image component segmentation method | |
| CN108710862A (en) | A kind of high-resolution remote sensing image Clean water withdraw method | |
| CN113128308B (en) | Pedestrian detection method, device, equipment and medium in port scene | |
| CN114627269A (en) | Virtual reality security protection monitoring platform based on degree of depth learning target detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant | ||
| CB03 | Change of inventor or designer information |
Inventor after: Zhang Dezheng Inventor after: Yang Pan Inventor after: Zhang Xuefeng Inventor after: Li Hui Inventor after: Xu Cong Inventor after: Aziguri Ulamu Inventor after: Ding Ruidong Inventor before: Zhang Dezheng Inventor before: Yang Pan Inventor before: Li Hui Inventor before: Xu Cong Inventor before: Aziguri Ulamu Inventor before: Ding Ruidong |
|
| CB03 | Change of inventor or designer information |