[go: up one dir, main page]
More Web Proxy on the site http://driver.im/

CN113850769A - Hyperspectral change detection method based on Simese space spectrum joint convolution network - Google Patents

Hyperspectral change detection method based on Simese space spectrum joint convolution network Download PDF

Info

Publication number
CN113850769A
CN113850769A CN202111092212.3A CN202111092212A CN113850769A CN 113850769 A CN113850769 A CN 113850769A CN 202111092212 A CN202111092212 A CN 202111092212A CN 113850769 A CN113850769 A CN 113850769A
Authority
CN
China
Prior art keywords
hyperspectral
change
pixels
pixel
tensor
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Granted
Application number
CN202111092212.3A
Other languages
Chinese (zh)
Other versions
CN113850769B (en
Inventor
詹天明
宋博
徐超
吴泽彬
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
NANJING AUDIT UNIVERSITY
Original Assignee
NANJING AUDIT UNIVERSITY
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by NANJING AUDIT UNIVERSITY filed Critical NANJING AUDIT UNIVERSITY
Priority to CN202111092212.3A priority Critical patent/CN113850769B/en
Publication of CN113850769A publication Critical patent/CN113850769A/en
Application granted granted Critical
Publication of CN113850769B publication Critical patent/CN113850769B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • 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
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10052—Images from lightfield camera
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20081—Training; Learning
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A40/00—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production
    • Y02A40/10—Adaptation technologies in agriculture, forestry, livestock or agroalimentary production in agriculture

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a hyperspectral change detection method based on a Simese space spectrum combined convolutional network, which comprises the following steps of: collecting two hyperspectral images at different time points in the same area, and removing wave bands with low signal-to-noise ratio; sequentially extracting the spectrum of a pixel to be detected and the spectrum of a neighborhood pixel thereof from hyperspectral images of the pixel at front and rear time points to form two tensors and pairing; selecting 10% of pixels as training pixels, and manually marking the change condition of tensor pairs; inputting a tensor pair for training into a Siamese network consisting of one-dimensional convolution and two-dimensional convolution, and obtaining corresponding eigenvectors; the hyperspectral image change detection method based on the hyperspectral image retains the complete spectrum and spatial characteristics of hyperspectrum, is high in identification speed, fuses the spectrum characteristics and the spatial characteristics, reduces the influence caused by noise, can effectively identify the change area which is interested by people in two hyperspectral images at different time phases, and improves the change detection efficiency and precision.

Description

Hyperspectral change detection method based on Simese space spectrum joint convolution network
Technical Field
The invention relates to the technical field of image processing, in particular to a hyperspectral change detection method based on a Simese space spectrum joint convolution network.
Background
With the development of remote sensing technology, it becomes possible to obtain hyperspectral images of different time phases in the same area; the method has important application value in the fields of disaster assessment, terrain change analysis, city change detection analysis and the like by utilizing the multi-temporal remote sensing image data to carry out change detection; the hyperspectral images containing hundreds of wave bands have abundant spectral information and spatial information, so that the ground object observation has a stronger data source.
At present, many scholars have studied the multispectral change detection task with low number of bands, and proposed some excellent change detection algorithms. However, in the hyperspectral change detection problem, the change detection algorithm adapted to the low-dimensional space has an insignificant effect in the hyperspectral high-dimensional space; in addition, because the relevance of adjacent wave bands in the hyperspectral image is strong, a large amount of unnecessary redundant information also provides challenges for feature extraction; noise in the hyperspectral image is inevitable, wherein the noise comes from internal noise brought by the hyperspectral imager and noise from external factors such as atmospheric scattering and the like; if the algorithm only mines the spectral characteristics of the pixels, the effect of change detection will be diminished; therefore, a hyperspectral change detection method based on the Siamese space spectrum joint convolution network needs to be designed.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, provides a hyperspectral change detection method based on a Simese spatial spectrum combined convolution network for better and effectively solving the related problems, effectively reduces the parameter amount of network learning, improves the change detection speed and also improves the change detection accuracy.
In order to achieve the purpose, the invention adopts the technical scheme that:
a hyperspectral change detection method based on a Simese spatial spectrum joint convolution network comprises the following steps,
collecting hyperspectral images of two different time points in the same area, and removing wave bands with low signal-to-noise ratio;
sequentially extracting the spectrums of the pixel and the adjacent pixels thereof from the hyperspectral images of the pixel to be detected at front and back time points to form two tensors and matching;
selecting 10% of pixels as training pixels, and artificially marking the change condition of tensor pairs;
inputting a tensor pair for training into a Simese network consisting of one-dimensional convolution and two-dimensional convolution, and obtaining a corresponding feature vector;
step (E), calculating Euclidean distance of the two eigenvectors as the similarity of the two tensors corresponding to the pixel point;
step (F), optimizing the Siamese network by comparing loss functions;
step (G), calculating the variation degree of the tensor of the pixel to the input optimized Siamese network;
and (H) binarizing the similarity through a threshold value method to generate a final change detection result.
In the method for detecting hyperspectral change based on the siemese space spectrum joint convolution network, step (B) is to respectively extract the spectra of the pixel and its neighborhood pixels from hyperspectral images of the pixel to be detected at two time points in front and at the back in sequence to form two tensors and pair the two tensors, wherein the specific steps are as follows,
step (B1), let
Figure BDA0003267795240000021
Representing a hyperspectral image of a t-th time point, wherein h and w respectively correspond to the height and the width of a hyperspectral space, c represents the number of wave bands of the hyperspectral image, and t belongs to {1,2 };
step (B2) of converting X(t)The spectra of all the pixels and the spectra of the corresponding neighborhood pixels in the image are extracted to form a hyperspectral tensor
Figure BDA0003267795240000022
And
Figure BDA0003267795240000023
wherein b represents the spatial dimension of the hyperspectral tensor, b ═ 3, i ∈ {1,2, …, hw };
step (B3), the hyperspectral tensors of the same pixel are paired to form a tensor pair set
Figure BDA0003267795240000031
In the hyperspectral change detection method based on the siemese space spectrum joint convolution network, in the step (C), 10% of pixels are selected as training pixels, the change condition of a tensor pair is marked artificially, wherein 10% of pixels are selected from hw pixels, and the tensor pair corresponding to the training pixels is marked artificially according to the actual change condition of the pixels to obtain YiAnd Y isi0 means pixel unchanged, Yi1 denotes a pixel change.
The hyperspectral change detection method based on the Simese spatial spectrum joint convolution network comprises the following specific steps of (D) inputting a tensor pair for training into the Simese network consisting of one-dimensional convolution and two-dimensional convolution to obtain a corresponding feature vector,
a step (D1) of performing spectral feature extraction on 2 hyperspectral tensors of 3 × c in the tensor pair by 3D convolution of 5 layers of sizes 1 × 3, to form 2 feature tensors of 3 × 80;
step (D2) of extracting spatial features from 2 feature tensors of 3 × 80 by 3-D convolution with a size of 3 × 1 to form 2 feature vectors
Figure BDA0003267795240000032
And
Figure BDA0003267795240000033
where d denotes the dimension of the feature vector and d is 80.
The hyperspectral change detection method based on the Simese spatial spectrum joint convolution network comprises the step (E) of calculating two featuresThe Euclidean distance of the eigenvector is used as the similarity of the two tensors corresponding to the pixel point, wherein the specific content of calculation is to calculate 2 eigenvectors
Figure BDA0003267795240000034
And
Figure BDA0003267795240000035
euclidean distance of
Figure BDA0003267795240000036
And the formula of the calculation is shown as formula (1),
Figure BDA0003267795240000037
the method comprises the following steps of (A) optimizing a Simese network through a comparison Loss function, wherein the specific steps are that a variation degree metric value corresponding to a training pixel and an artificially marked actual variation mark value are input into the comparison Loss function to calculate a Loss function value, parameters in the Simese network are updated and optimized reversely according to the Loss function value, and the comparison Loss function Loss is as shown in a formula (2),
Figure BDA0003267795240000041
where m denotes a feature vector distance threshold, and m is 1.5.
The hyperspectral change detection method based on the Simese spatial spectrum joint convolution network comprises the step (G) of calculating the change degree of the tensor of the pixel to the input optimized Simese network, wherein the change degree is calculated
Figure BDA0003267795240000042
As a measure of the degree of variation of the pixel.
In the foregoing method for detecting hyperspectral variation based on the Siamese space spectrum joint convolution network, in step (H), the similarity is binarized by a threshold method to generate a final variation detection result, which includes the following specific steps,
and step (H1) of determining the change threshold T by using the manual marking and the change degree metric of the training pixel through a traversal method, which comprises the following steps,
step (H11) of obtaining the minimum and maximum degree of change metric t in the training pixelsminAnd tmax；
Step (H12) of setting a candidate change determination threshold Tj＝tmin+0.0001j, wherein
Figure BDA0003267795240000043
And Tj∈[tmin,tmax]Wherein T isjBased on a measure of the variation of the training pixels
Figure BDA0003267795240000044
Predict the change if
Figure BDA0003267795240000045
Then the prediction is 0 and the prediction is,
Figure BDA0003267795240000046
the prediction is 1;
step H13, the prediction result and the manually marked YiComparing, calculating the number of pixels with correct prediction as
Figure BDA0003267795240000047
The variation threshold T is as shown in equation (3),
Figure BDA0003267795240000048
a step (H2) of judging the variation of all the pixels based on the variation threshold T and generating a variation detection result map CM, which includes the steps of,
a step (H21) of measuring the degree of change of the pixel
Figure BDA0003267795240000051
And the change threshold T generates a change detection result
Figure BDA0003267795240000052
As shown in the formula (4),
Figure BDA0003267795240000053
step (H22) of converting CM intoiDeformation-resulting change detection map
Figure BDA0003267795240000054
The invention has the beneficial effects that: according to the hyperspectral change detection method based on the Simese spatial spectrum joint convolution network, a high-dimensional tensor is processed by utilizing a deep learning theory, and a change detection problem is converted into a problem of measuring the similarity of two tensors by utilizing the Simese network, so that compared with a difference method, the hyperspectral complete spectrum and spatial characteristics are reserved, the spectral characteristics of the tensor are extracted through a one-dimensional convolution network in the Simese network, the dimension is reduced, the parameter quantity of network learning is effectively reduced, and the change detection speed is improved; the two-dimensional convolution network extracts spatial features from the tensor after dimensionality reduction, the two features are fused to reduce the influence caused by noise, the accuracy of change detection is improved, and the method has the advantages of being scientific and reasonable, strong in applicability, good in effect and the like.
Drawings
FIG. 1 is a flow chart of a hyperspectral change detection method based on a Simese spatial spectrum joint convolution network according to the invention;
FIG. 2 is a schematic view of a first time-phase hyperspectral image of the invention;
FIG. 3 is a second phase hyperspectral representation of the invention;
FIG. 4 is a schematic diagram of a hyperspectral change detection result of the invention;
FIG. 5 is a schematic diagram of ground truth of an actual change area marked by an artificial marker according to the present invention.
Detailed Description
The invention will be further described with reference to the accompanying drawings.
As shown in FIG. 1, the hyperspectral change detection method based on the Simese spatial spectrum joint convolution network of the invention comprises the following steps,
collecting hyperspectral images of two different time points in the same area, and removing wave bands with low signal-to-noise ratio;
step (B), respectively extracting the spectra of the pixel and the adjacent pixels thereof from the hyperspectral images of the pixel to be detected at front and back time points in sequence to form two tensors and matching the two tensors, wherein the specific steps are as follows,
step (B1), let
Figure BDA0003267795240000061
Representing a hyperspectral image of a t-th time point, wherein h and w respectively correspond to the height and the width of a hyperspectral space, c represents the number of wave bands of the hyperspectral image, and t belongs to {1,2 };
step (B2) of converting X(t)The spectra of all the pixels and the spectra of the corresponding neighborhood pixels in the image are extracted to form a hyperspectral tensor
Figure BDA0003267795240000062
And
Figure BDA0003267795240000063
wherein b represents the spatial dimension of the hyperspectral tensor, b ═ 3, i ∈ {1,2, …, hw };
step (B3), the hyperspectral tensors of the same pixel are paired to form a tensor pair set
Figure BDA0003267795240000064
And (C) selecting 10% of pixels as training pixels, and artificially marking the change condition of tensor pairs, wherein 10% of pixels are selected from hw pixels, and the tensor corresponding to the training pixels is subjected to actual change condition of the pixelsThe quantity pairs are manually marked to obtain YiAnd Y isi0 means pixel unchanged, Yi1 denotes a pixel change.
Step (D), inputting tensor pairs for training into a Simese network consisting of one-dimensional convolution and two-dimensional convolution, and obtaining corresponding eigenvectors,
a step (D1) of performing spectral feature extraction on 2 hyperspectral tensors of 3 × c in the tensor pair by 3D convolution of 5 layers of sizes 1 × 3, to form 2 feature tensors of 3 × 80;
step (D2) of extracting spatial features from 2 feature tensors of 3 × 80 by 3-D convolution with a size of 3 × 1 to form 2 feature vectors
Figure BDA0003267795240000071
And
Figure BDA0003267795240000072
where d denotes the dimension of the feature vector and d is 80.
Step (E), calculating Euclidean distance of two eigenvectors as the similarity of two tensors corresponding to the pixel point, wherein the specific content of calculation is to calculate 2 eigenvectors
Figure BDA0003267795240000073
And
Figure BDA0003267795240000074
euclidean distance of
Figure BDA0003267795240000075
And the formula of the calculation is shown as formula (1),
Figure BDA0003267795240000076
wherein the distance serves as a measure of the degree of change of the pixel at two points in time.
Step (F), optimizing the Siamese network by a contrast Loss function, wherein the specific steps are that the variation degree metric value corresponding to the training pixel and the actual variation marking value marked by manual work are input into the contrast Loss function to calculate the Loss function value, and then parameters in the Siamese network are updated and optimized reversely according to the Loss function value, and the contrast Loss function Loss is shown as a formula (2),
Figure BDA0003267795240000077
where m denotes a feature vector distance threshold, and m is 1.5.
Step (G) of calculating the degree of change of the tensor of the pixel to the input optimized Siamese network, wherein the degree of change is calculated
Figure BDA0003267795240000078
As a measure of the degree of variation of the pixel.
And (H) binarizing the similarity by a threshold method to generate a final change detection result, which comprises the following steps,
and step (H1) of determining the change threshold T by using the manual marking and the change degree metric of the training pixel through a traversal method, which comprises the following steps,
step (H11) of obtaining the minimum and maximum degree of change metric t in the training pixelsminAnd tmax；
Step (H12) of setting a candidate change determination threshold Tj＝tmin+0.0001j, wherein
Figure BDA0003267795240000081
And Tj∈[tmin,tmax]Wherein T isjBased on a measure of the variation of the training pixels
Figure BDA0003267795240000082
Predict the change if
Figure BDA0003267795240000083
Then the prediction is 0 and the prediction is,
Figure BDA0003267795240000084
the prediction is 1;
step H13, the prediction result and the manually marked YiComparing, calculating the number of pixels with correct prediction as
Figure BDA0003267795240000085
The variation threshold T is as shown in equation (3),
Figure BDA0003267795240000086
a step (H2) of judging the variation of all the pixels based on the variation threshold T and generating a variation detection result map CM, which includes the steps of,
a step (H21) of measuring the degree of change of the pixel
Figure BDA0003267795240000087
And the change threshold T generates a change detection result
Figure BDA0003267795240000088
As shown in the formula (4),
Figure BDA0003267795240000089
step (H22) of converting CM intoiDeformation-resulting change detection map
Figure BDA00032677952400000810
A specific embodiment of the hyperspectral change detection method based on the Simese space spectrum joint convolution network can show the effect of hyperspectral change detection, such as fig. 2-4, wherein fig. 2 and 3 are hyperspectral images of two different time points in the same area, fig. 4 is a change detection result of the method, and fig. 5 is a truth diagram of actual change ground change; as can be seen from the detection result and the actual ground change true value graph, the method can effectively identify the change area.
In summary, according to the hyperspectral change detection method based on the Simese spatial spectrum joint convolution network, a high-dimensional tensor is processed by utilizing a deep learning theory, and a change detection problem is converted into a problem of measuring the similarity of two tensors by utilizing the Simese network, so that compared with a difference method, the hyperspectral complete spectrum and spatial features are reserved, the spectral features of the tensors are extracted and reduced in dimension through the one-dimensional convolution network in the Simese network, the parameter quantity of network learning is effectively reduced, and the change detection speed is improved; the two-dimensional convolution network extracts spatial features from the tensor after dimensionality reduction, the two features are fused to reduce the influence caused by noise, the accuracy of change detection is improved, and the method is suitable for a change detection task of a hyperspectral image.
The foregoing illustrates and describes the principles, general features, and advantages of the present invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, which are described in the specification and illustrated only to illustrate the principle of the present invention, but that various changes and modifications may be made therein without departing from the spirit and scope of the present invention, which fall within the scope of the invention as claimed. The scope of the invention is defined by the appended claims and equivalents thereof.

Claims (8)

1. A hyperspectral change detection method based on a Simese spatial spectrum joint convolution network is characterized by comprising the following steps: comprises the following steps of (a) carrying out,
collecting hyperspectral images of two different time points in the same area, and removing wave bands with low signal-to-noise ratio;
sequentially extracting the spectrums of the pixel and the adjacent pixels thereof from the hyperspectral images of the pixel to be detected at front and back time points to form two tensors and matching;
selecting 10% of pixels as training pixels, and artificially marking the change condition of tensor pairs;
inputting a tensor pair for training into a Simese network consisting of one-dimensional convolution and two-dimensional convolution, and obtaining a corresponding feature vector;
step (E), calculating Euclidean distance of the two eigenvectors as the similarity of the two tensors corresponding to the pixel point;
step (F), optimizing the Siamese network by comparing loss functions;
step (G), calculating the variation degree of the tensor of the pixel to the input optimized Siamese network;
and (H) binarizing the similarity through a threshold value method to generate a final change detection result.
2. The method for detecting hyperspectral variation based on the Simese spatial spectrum joint convolutional network as claimed in claim 1, wherein: step (B), respectively extracting the spectra of the pixel and the adjacent pixels thereof from the hyperspectral images of the pixel to be detected at front and back time points in sequence to form two tensors and matching the two tensors, wherein the specific steps are as follows,
step (B1), let
Figure FDA0003267795230000011
Representing a hyperspectral image of a t-th time point, wherein h and w respectively correspond to the height and the width of a hyperspectral space, c represents the number of wave bands of the hyperspectral image, and t belongs to {1,2 };
step (B2) of converting X(t)The spectra of all the pixels and the spectra of the corresponding neighborhood pixels in the image are extracted to form a hyperspectral tensor
Figure FDA0003267795230000021
And
Figure FDA0003267795230000022
wherein b represents the spatial dimension of the hyperspectral tensor, b ═ 3, i ∈ {1,2, …, hw };
step (B3), the hyperspectral tensors of the same pixel are paired to form a tensor pair set
Figure FDA0003267795230000023
3. The method for detecting hyperspectral variation based on the siemese space spectrum joint convolutional network as claimed in claim 2, wherein: and (C) selecting 10% of pixels as training pixels, and artificially marking the change condition of tensor pairs, wherein 10% of pixels are selected from hw pixels, and artificially marking the tensor pairs corresponding to the training pixels according to the actual change condition of the pixels to obtain YiAnd Y isi0 means pixel unchanged, Yi1 denotes a pixel change.
4. The method for detecting hyperspectral variation based on the Simese spatial spectrum joint convolutional network as claimed in claim 3, wherein: step (D), inputting tensor pairs for training into a Simese network consisting of one-dimensional convolution and two-dimensional convolution, and obtaining corresponding eigenvectors,
a step (D1) of performing spectral feature extraction on 2 hyperspectral tensors of 3 × c in the tensor pair by 3D convolution of 5 layers of sizes 1 × 3, to form 2 feature tensors of 3 × 80;
step (D2) of extracting spatial features from 2 feature tensors of 3 × 80 by 3-D convolution with a size of 3 × 1 to form 2 feature vectors
Figure FDA0003267795230000024
And
Figure FDA0003267795230000025
where d denotes the dimension of the feature vector and d is 80.
5. The method for detecting hyperspectral variation based on the Simese spatial spectrum joint convolutional network as claimed in claim 4, wherein: step (E), calculating Euclidean distance of two eigenvectors as the similarity of two tensors corresponding to the pixel point, wherein the specific calculation content is 2 tensorsFeature vector
Figure FDA0003267795230000026
And
Figure FDA0003267795230000027
euclidean distance of
Figure FDA0003267795230000028
And the formula of the calculation is shown as formula (1),
Figure FDA0003267795230000029
6. the method for detecting hyperspectral variation based on the Simese spatial spectrum joint convolutional network as claimed in claim 5, wherein: step (F), optimizing the Siamese network by a contrast Loss function, wherein the specific steps are that the variation degree metric value corresponding to the training pixel and the actual variation marking value marked by manual work are input into the contrast Loss function to calculate the Loss function value, and then parameters in the Siamese network are updated and optimized reversely according to the Loss function value, and the contrast Loss function Loss is shown as a formula (2),
Figure FDA0003267795230000031
where m denotes a feature vector distance threshold, and m is 1.5.
7. The method for detecting hyperspectral variation based on the Simese spatial spectrum joint convolutional network as claimed in claim 6, wherein: step (G) of calculating the degree of change of the tensor of the pixel to the input optimized Siamese network, wherein the degree of change is calculated
Figure FDA0003267795230000032
As pixelsA measure of the degree of change.
8. The method for detecting hyperspectral variation based on the siemese space spectrum joint convolutional network as claimed in claim 7, wherein: and (H) binarizing the similarity by a threshold method to generate a final change detection result, which comprises the following steps,
and step (H1) of determining the change threshold T by using the manual marking and the change degree metric of the training pixel through a traversal method, which comprises the following steps,
step (H11) of obtaining the minimum and maximum degree of change metric t in the training pixelsminAnd tmax；
Step (H12) of setting a candidate change determination threshold Tj＝tmin+0.0001j, wherein
Figure FDA0003267795230000033
And Tj∈[tmin,tmax]Wherein T isjBased on a measure of the variation of the training pixels
Figure FDA0003267795230000034
Predict the change if
Figure FDA0003267795230000035
Then the prediction is 0 and the prediction is,
Figure FDA0003267795230000036
the prediction is 1;
step H13, the prediction result and the manually marked YiComparing, calculating the number of pixels with correct prediction as
Figure FDA0003267795230000041
The variation threshold T is as shown in equation (3),
Figure FDA0003267795230000042
a step (H2) of judging the variation of all the pixels based on the variation threshold T and generating a variation detection result map CM, which includes the steps of,
a step (H21) of measuring the degree of change of the pixel
Figure FDA0003267795230000043
And the change threshold T generates a change detection result
Figure FDA0003267795230000044
As shown in the formula (4),
Figure FDA0003267795230000045
step (H22) of converting CM intoiDeformation-resulting change detection map
Figure FDA0003267795230000046
CN202111092212.3A 2021-09-17 2021-09-17 Hyperspectral change detection method based on Simese space spectrum joint convolution network Active CN113850769B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111092212.3A CN113850769B (en) 2021-09-17 2021-09-17 Hyperspectral change detection method based on Simese space spectrum joint convolution network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111092212.3A CN113850769B (en) 2021-09-17 2021-09-17 Hyperspectral change detection method based on Simese space spectrum joint convolution network

Publications (2)

Publication Number Publication Date
CN113850769A true CN113850769A (en) 2021-12-28
CN113850769B CN113850769B (en) 2023-04-07

Family

ID=78974540

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111092212.3A Active CN113850769B (en) 2021-09-17 2021-09-17 Hyperspectral change detection method based on Simese space spectrum joint convolution network

Country Status (1)

Country Link
CN (1) CN113850769B (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115240072A (en) * 2022-08-04 2022-10-25 南京审计大学 Hyperspectral multi-class change detection method based on multidirectional multi-scale spectrum-space residual convolution neural network
CN117765402A (en) * 2024-02-21 2024-03-26 山东科技大学 Hyperspectral image matching detection method based on attention mechanism

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190057268A1 (en) * 2017-08-15 2019-02-21 Noblis, Inc. Multispectral anomaly detection
CN109785302A (en) * 2018-12-27 2019-05-21 中国科学院西安光学精密机械研究所 A kind of empty spectrum union feature learning network and multispectral change detecting method
CN111539316A (en) * 2020-04-22 2020-08-14 中南大学 High-resolution remote sensing image change detection method based on double attention twin network
WO2021179632A1 (en) * 2020-09-23 2021-09-16 平安科技（深圳）有限公司 Medical image classification method, apparatus and device, and storage medium

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20190057268A1 (en) * 2017-08-15 2019-02-21 Noblis, Inc. Multispectral anomaly detection
CN109785302A (en) * 2018-12-27 2019-05-21 中国科学院西安光学精密机械研究所 A kind of empty spectrum union feature learning network and multispectral change detecting method
CN111539316A (en) * 2020-04-22 2020-08-14 中南大学 High-resolution remote sensing image change detection method based on double attention twin network
WO2021179632A1 (en) * 2020-09-23 2021-09-16 平安科技（深圳）有限公司 Medical image classification method, apparatus and device, and storage medium

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
王志有等: "基于深度学习算法的卫星影像变化监测", 《计算机系统应用》 *

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115240072A (en) * 2022-08-04 2022-10-25 南京审计大学 Hyperspectral multi-class change detection method based on multidirectional multi-scale spectrum-space residual convolution neural network
CN115240072B (en) * 2022-08-04 2024-03-12 南京审计大学 Hyperspectral multi-class change detection method based on multidirectional multi-scale spectrum-space residual convolution neural network
CN117765402A (en) * 2024-02-21 2024-03-26 山东科技大学 Hyperspectral image matching detection method based on attention mechanism
CN117765402B (en) * 2024-02-21 2024-05-17 山东科技大学 Hyperspectral image matching detection method based on attention mechanism

Also Published As

Publication number Publication date
CN113850769B (en) 2023-04-07

Similar Documents

Publication Publication Date Title
US11244197B2 (en) Fast and robust multimodal remote sensing image matching method and system
CN110378196B (en) Road visual detection method combining laser point cloud data
CN103400151B (en) The optical remote sensing image of integration and GIS autoregistration and Clean water withdraw method
CN104574347B (en) Satellite in orbit image geometry positioning accuracy evaluation method based on multi- source Remote Sensing Data data
CN107992891B (en) Multispectral remote sensing image change detection method based on spectral vector analysis
CN101504315B (en) Expansion morphology and orthogonal subspace projection combined end member automatic extraction method
CN111368691B (en) Unsupervised hyperspectral remote sensing image space spectrum feature extraction method
CN111008664B (en) Hyperspectral sea ice detection method based on space-spectrum combined characteristics
CN113850769B (en) Hyperspectral change detection method based on Simese space spectrum joint convolution network
CN107240130A (en) Remote Sensing Image Matching method, apparatus and system
CN114022459A (en) Multi-temporal satellite image-based super-pixel change detection method and system
Singh et al. A hybrid approach for information extraction from high resolution satellite imagery
CN117274627A (en) Multi-temporal snow remote sensing image matching method and system based on image conversion
Liu et al. Extraction of earthquake-induced collapsed buildings from bi-temporal VHR images using object-level homogeneity index and histogram
Chen et al. Spectral Query Spatial: Revisiting the Role of Center Pixel in Transformer for Hyperspectral Image Classification
Bin et al. Parcel-based change detection in land-use maps by adopting the holistic feature
Kim et al. Automatic pseudo-invariant feature extraction for the relative radiometric normalization of hyperion hyperspectral images
Tu et al. Feature extraction using multidimensional spectral regression whitening for hyperspectral image classification
CN107832793A (en) The sorting technique and system of a kind of high spectrum image
Zhang et al. Feature-band-based unsupervised hyperspectral underwater target detection near the coastline
CN111047616B (en) Remote sensing image landslide target constraint active contour feature extraction method
CN111199251B (en) Multi-scale hyperspectral image classification method based on weighted neighborhood
CN108154107B (en) Method for determining scene category to which remote sensing image belongs
Yuan et al. Graph neural network based multi-feature fusion for building change detection
Huang et al. Morphological building index (MBI) and its applications to urban areas

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