WO2018192023A1 - 一种高光谱遥感图像分类方法及装置 - Google Patents

一种高光谱遥感图像分类方法及装置 Download PDF

Info

Publication number
WO2018192023A1
WO2018192023A1 PCT/CN2017/083766 CN2017083766W WO2018192023A1 WO 2018192023 A1 WO2018192023 A1 WO 2018192023A1 CN 2017083766 W CN2017083766 W CN 2017083766W WO 2018192023 A1 WO2018192023 A1 WO 2018192023A1
Authority
WO
WIPO (PCT)
Prior art keywords
remote sensing
sensing image
gabor
hyperspectral remote
features
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.)
Ceased
Application number
PCT/CN2017/083766
Other languages
English (en)
French (fr)
Inventor
贾森
邓琳
沈琳琳
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.)
Shenzhen University
Original Assignee
Shenzhen 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 Shenzhen University filed Critical Shenzhen University
Publication of WO2018192023A1 publication Critical patent/WO2018192023A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques
    • G06F18/253—Fusion techniques of extracted features

Definitions

  • the invention relates to the field of image processing, and in particular to a method and a device for classifying hyperspectral remote sensing images.
  • Hyperspectral remote sensing image refers to the hyperspectral image data obtained by imaging the sensor in different wavelengths in the visible, near-infrared, mid-infrared and thermal infrared bands of the electromagnetic spectrum. Therefore, hyperspectral remote sensing images contain a wealth of spatial, radiation and spectral triple information, which provides a possibility for the fine classification and identification of surface materials.
  • the amplitude information of three-dimensional Gabor features since the amplitude information of three-dimensional Gabor features has good stability, it is usually used directly for classification, but for hyperspectral remote sensing images, the three-dimensional Gabor features are rich. The phase characteristics, so using only the amplitude characteristics of the Gabor feature to classify the surface features will make the classification accuracy low.
  • the embodiment of the invention provides a method and a device for classifying hyperspectral remote sensing images, so as to improve the classification accuracy of the features.
  • an embodiment of the present invention provides a method for classifying a hyperspectral remote sensing image, where the method includes:
  • N is a positive integer
  • the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features are feature-fused by a preset feature fusion algorithm to determine a feature category in the hyperspectral remote sensing image.
  • an embodiment of the present invention provides a hyperspectral remote sensing image classification device, wherein the device includes:
  • An acquiring module configured to acquire N preset Gabor filters, where N is a positive integer
  • An extracting module configured to acquire N three-dimensional Gabor amplitude features and N three-dimensional Gabor phase features of the hyperspectral remote sensing image based on the N preset Gabor filters;
  • a determining module configured to perform feature fusion on the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine a feature category in the hyperspectral remote sensing image.
  • N preset Gabor filters are first acquired, and then N three-dimensional Gabor amplitude features of the hyperspectral remote sensing image are acquired based on the N preset Gabor filters.
  • N three-dimensional Gabor phase features and finally feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine a category of the features in the hyperspectral remote sensing image .
  • phase feature of the hyperspectral remote sensing image includes rich phase features
  • the three-dimensional Gabor amplitude feature is complementary to the three-dimensional Gabor phase feature
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor are merged by the hyperspectral remote sensing image in the embodiment of the present invention.
  • the phase feature is used to determine the feature categories in hyperspectral remote sensing images and improve the accuracy of feature classification for hyperspectral remote sensing images.
  • the three-dimensional Gabor phase information has extremely high sensitivity to the spatial position of the ground object, the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image are used for classification, which reduces the classification robustness.
  • FIG. 1 is a schematic flow chart of a first embodiment of a hyperspectral remote sensing image classification method according to an embodiment of the present invention
  • FIG. 2 is a schematic diagram of a frequency domain relationship of a three-dimensional Gabor feature according to an embodiment of the present invention
  • 3 is a three-dimensional Gabor filter parallel to the spectral dimension direction observed from three different viewing angles according to an embodiment of the present invention
  • FIG. 4-a is a hyperspectral remote sensing image according to an embodiment of the present invention.
  • Figure 4-b is a three-dimensional Gabor amplitude feature set provided by an embodiment of the present invention.
  • Figure 4-c is a phase feature set and a coding feature set of a hyperspectral remote sensing image according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram showing classification of a hyperspectral remote sensing image provided by an embodiment of the present invention.
  • FIG. 6 is a schematic flow chart of a second embodiment of a hyperspectral remote sensing image classification method according to an embodiment of the present invention.
  • FIG. 7 is a schematic structural diagram of a first embodiment of a hyperspectral remote sensing image classification apparatus according to an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of a second embodiment of a hyperspectral remote sensing image classification apparatus according to an embodiment of the present invention.
  • the embodiment of the invention provides a method and a device for classifying hyperspectral remote sensing images, so as to improve the classification accuracy of the features.
  • N is a positive integer
  • a fusion algorithm performs feature fusion on the N Gabor amplitude features and the N three-dimensional Gabor phase features to determine a feature category in the hyperspectral remote sensing image.
  • FIG. 1 is a schematic flowchart diagram of a first embodiment of a hyperspectral remote sensing image classification method according to an embodiment of the present invention.
  • the hyperspectral remote sensing image classification method provided by the embodiment of the present invention includes the following steps:
  • the N is a positive integer
  • the N preset Gabor filters are used for subsequent extraction of amplitude features and phase features of the hyperspectral remote sensing image.
  • the N preset Gabor filters are Gabor filters that are parallel to the spectral dimension direction of the hyperspectral remote sensing image.
  • N Gabor filters parallel to the spectral dimension direction of the hyperspectral remote sensing image can be obtained by the following formula
  • f is the frequency of the preset Gabor filter
  • I the angle between the preset Gabor filter and the ⁇ axis
  • ⁇ is the angle between the preset Gabor filter and the uv plane.
  • ⁇ is the width of the Gaussian envelope.
  • FIG. 2 is a schematic diagram of a frequency domain relationship of a three-dimensional Gabor feature according to an embodiment of the present invention.
  • the direction indicated by the frequency f is a spectrum of a hyperspectral remote sensing image. Dimension direction.
  • FIG. 3 is a three-dimensional Gabor filter parallel to the spectral dimension direction observed from three different perspectives according to an embodiment of the present invention.
  • the Gabor filter may be a Log-Gabor filter, a haar-Gabor filter.
  • a three-dimensional Gabor feature is obtained by convoluting a hyperspectral remote sensing image with N preset Gabor filters, and then encoding each pixel of the hyperspectral image to further obtain a high Amplitude and phase characteristics of the spectral image.
  • the hyperspectral remote sensing image is convoluted with N three-dimensional Gabor filters to obtain i three-dimensional Gabor features.
  • R represents the hyperspectral remote sensing image
  • G i represents the i-th three-dimensional Gabor feature, where i is an arbitrary integer between 1 and N.
  • the amplitude characteristics M i (x, y, b) and phase characteristics F i (x, y, b) of the hyperspectral remote sensing image are calculated by the following formula:
  • M i (x, y, b) abs(G i (x, y, b));
  • Re(G i (x, y, b)) and Im(G i (x, y, b)) are the real and imaginary parts of the Gabor feature, respectively.
  • the four three-dimensional Gabor filters shown in FIG. 3 are first acquired, and the hyperspectral remote sensing image and the generated four three-dimensional Gabor filters are further performed.
  • the convolution operation four three-dimensional Gabor filter features are obtained, and four three-dimensional Gabor amplitude features G i (x, y, b) and three-dimensional Gabor phase features F i (x, y, b) are further obtained by the above formula. ), where i takes any integer between 1 and 4.
  • 4-a is a hyperspectral remote sensing image according to an embodiment of the present invention, and two Gabor filters in four three-dimensional Gabor filters are taken as For example, the three-dimensional Gabor amplitude feature set of Figure 4-b and the three-dimensional Gabor phase feature set of Figure 4-c are obtained.
  • 4b is a three-dimensional Gabor amplitude feature set provided by an embodiment of the present invention;
  • FIG. 4-c is a phase feature set and a coding feature set of the hyperspectral remote sensing image provided by the embodiment of the present invention.
  • the first column represents the three-dimensional Gabor phase features obtained by the two three-dimensional Gabor filters; the second column represents the features obtained when the real part of the first column of the three-dimensional Gabor features is encoded; A feature obtained by encoding an imaginary part of a first column of three-dimensional Gabor features.
  • other methods may also be used to acquire the three-dimensional Gabor phase features, for example, based on a fusion coding method or a contention based coding method.
  • the preset feature fusion algorithm refers to an algorithm for combining three-dimensional Gabor amplitude features and phase features to simultaneously use three-dimensional Gabor amplitude features and three-dimensional Gabor phase features for classification of remote sensing images.
  • the N-dimensional Gabor amplitude feature and the N three-dimensional Gabor phase features are feature-fused by a preset feature fusion algorithm to determine the hyperspectral remote sensing.
  • the feature categories in the image including:
  • a category similarity measure p based on a Gabor phase characteristics from any of the N-dimensional Gabor phase characteristics of the acquired remote sensing images F i and any of the categories of P P is a positive integer;
  • Determining the confidence in the p categories Measure distance from the similarity
  • the category in which the sum of the squared differences under the N Gabor filters is the largest is the feature class in the hyperspectral remote sensing image.
  • confidence is between 0 and 1, which is used to reflect the probability that the hyperspectral remote sensing image belongs to the p-th category, thus the confidence level
  • the Hamming distance The value is between 0 and 1, which is used to reflect the matching degree between the hyperspectral remote sensing image and the category p, when the Hamming distance The smaller the smaller, the greater the probability that the hyperspectral remote sensing image belongs to the p-class.
  • the basis of any of the N-dimensional Gabor amplitude characteristic of a three-dimensional Gabor amplitude characteristic M i acquires the image sensing P belongs to any one of categories category p Confidence include:
  • the D is a decision matrix of the remote sensing image acquired by the support vector machine, and the n p is the number of non-zero elements of the p-th row in the decision matrix D.
  • the decision matrix P can be obtained by assuming that there is a P-type feature, and for each test sample t, P ⁇ (P-1) are established by using a one-to-one strategy.
  • the confidence is between 0 and 1, and when the confidence is The larger the larger, the greater the probability that the hyperspectral remote sensing image belongs to category p.
  • the addition and multiplication can also be used to calculate the confidence.
  • the similarity measure distance For Hamming distance the acquiring a similarity metric distance between the hyperspectral remote sensing image and any one of the P categories based on any of the N three-dimensional Gabor phase features F i include:
  • the similarity measure between the hyperspectral remote sensing image t and any training sample s in the training set A is obtained by the following formula:
  • B is the spectral dimension of the hyperspectral image
  • the Hamming distance The value is between 0 and 1, and when Hamming distance The smaller the time, the greater the probability that the hyperspectral remote sensing image belongs to the category p, and the Hamming distance for the optimal matching of the hyperspectral remote sensing image and the category p Zero.
  • the foregoing Hamming distance may also be calculated by using a classification based on sparse representation, K-inline classification, and the like.
  • the similarity metric distance may also be other distances, such as a Levinstein distance and a Li distance.
  • the above confidence Distance from Hamming The parameters in the calculation formula are all determined parameters, so when using the above formula to calculate, no parameter estimation is needed, which will make the calculation more accurate.
  • the E p is defined as the confidence Distance from the Hamming The sum of the squared differences under the N Gabor filters, ie Because of confidence The larger the probability, the higher the probability that the hyperspectral remote sensing image belongs to the category p, when the Hamming distance The smaller the hyperspectral imagery greater probability of belonging to the category p, whereby when the three-dimensional integration and three-dimensional Gabor Gabor phase amplitude characteristic feature, the larger E p, the hyperspectral imagery greater probability of belonging to the category p, and finally The category corresponding to the value at which E p is the largest is the feature type in the hyperspectral remote sensing image.
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image can be simultaneously used in determining the feature category in the hyperspectral remote sensing image, thereby making the determined category more accurate.
  • the fusion feature value E p may also be defined in other forms, such as a square form, an exponential form, and the like.
  • the four three-dimensional Gabor filters shown in FIG. 3 are first obtained, and then the three-dimensional Gabor amplitudes shown in FIG. 4-b are calculated by using step S102.
  • the three-dimensional Gabor amplitude features shown in Figure 4-c are calculated by using step S102.
  • the hyperspectral remote sensing image 4-a is obtained as the confidence level of each class.
  • the hyperspectral remote sensing image is obtained for each class of Hamming distance.
  • FIG. 5 is a schematic diagram showing classification of hyperspectral remote sensing images according to an embodiment of the present invention.
  • N preset Gabor filters are first acquired, and then N three-dimensional Gabor amplitude features and N three-dimensional Gabors of the hyperspectral remote sensing image are acquired based on the N preset Gabor filters. Phase features, and finally feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features using a preset feature fusion algorithm to determine feature categories in the hyperspectral remote sensing image.
  • the phase feature of the hyperspectral remote sensing image includes rich phase features
  • the three-dimensional Gabor amplitude feature is complementary to the three-dimensional Gabor phase feature
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase are obtained by fusing the hyperspectral remote sensing image in the embodiment of the present invention.
  • the feature is used to determine the classification of the features in the hyperspectral remote sensing image and improve the accuracy of the classification of the features of the hyperspectral remote sensing image.
  • the three-dimensional Gabor phase information has extremely high sensitivity to the spatial position of the ground object, the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image are used for classification, which reduces the classification robustness.
  • FIG. 6 is a schematic flowchart diagram of a second embodiment of a hyperspectral remote sensing image classification method according to an embodiment of the present invention.
  • the same or similar content as the method shown in FIG. 1 can be referred to the detailed description in FIG. 1, and details are not described herein again.
  • the hyperspectral remote sensing image classification method provided by the embodiment of the present invention includes the following steps:
  • S605. Determine the confidence level in the p categories. Measure distance from the similarity The category in which the sum of the squared differences under the N Gabor filters is the largest is the feature class in the hyperspectral remote sensing image.
  • acquiring the target image first acquiring N preset Gabor filters, and then acquiring N three-dimensional Gabor amplitude features of the hyperspectral remote sensing image based on the N preset Gabor filters N three-dimensional Gabor phase features, and finally feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine a category of the features in the hyperspectral remote sensing image .
  • the phase feature of the hyperspectral remote sensing image includes rich phase features
  • the three-dimensional Gabor amplitude feature is complementary to the three-dimensional Gabor phase feature
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase are obtained by fusing the hyperspectral remote sensing image in the embodiment of the present invention.
  • the features are used to determine the feature categories in hyperspectral remote sensing images and improve the accuracy of feature classification for hyperspectral remote sensing images.
  • the three-dimensional Gabor phase information has extremely high sensitivity to the spatial position of the ground object, the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image are used for classification, which reduces the classification robustness.
  • the embodiment of the invention further provides a hyperspectral remote sensing image classification device, comprising:
  • An acquiring module configured to acquire N preset Gabor filters, where N is a positive integer
  • An extracting module configured to acquire a hyperspectral remote sensing image based on the N preset Gabor filters Three-dimensional Gabor amplitude features and N three-dimensional Gabor phase features;
  • a determining module configured to perform feature fusion on the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine a feature category in the hyperspectral remote sensing image.
  • FIG. 7 is a schematic structural diagram of a first embodiment of a hyperspectral remote sensing image classification device according to an embodiment of the present invention, which is used to implement a hyperspectral remote sensing image classification disclosed in an embodiment of the present invention. method.
  • a hyperspectral remote sensing image classification device 700 according to an embodiment of the present invention may include:
  • the acquisition module 710, the extraction module 720, and the determination module 730 are the acquisition module 710, the extraction module 720, and the determination module 730.
  • the obtaining module 710 is configured to acquire N preset Gabor filters, where the N is a positive integer.
  • the N is a positive integer
  • the N preset Gabor filters are used for subsequent extraction of amplitude features and phase features of the hyperspectral remote sensing image.
  • the N preset Gabor filters are Gabor filters that are parallel to the spectral dimension direction of the hyperspectral remote sensing image.
  • N Gabor filters parallel to the spectral dimension direction of the hyperspectral remote sensing image can be obtained by the following formula
  • f is the frequency of the preset Gabor filter
  • I the angle between the preset Gabor filter and the ⁇ axis
  • ⁇ is the angle between the preset Gabor filter and the uv plane.
  • ⁇ is the width of the Gaussian envelope.
  • FIG. 2 is a schematic diagram of a frequency domain relationship of a three-dimensional Gabor feature according to an embodiment of the present invention.
  • the direction indicated by the frequency f is a spectrum of a hyperspectral remote sensing image. Dimension direction.
  • FIG. 3 is a three-dimensional Gabor filter parallel to the spectral dimension direction observed from three different perspectives according to an embodiment of the present invention.
  • the Gabor filter may be a Log-Gabor filter, a haar-Gabor filter.
  • the extracting module 720 is configured to acquire N three-dimensional Gabor amplitude features and N three-dimensional Gabor phase features of the hyperspectral remote sensing image based on the N preset Gabor filters.
  • a three-dimensional Gabor feature is obtained by convoluting a hyperspectral remote sensing image with N preset Gabor filters, and then encoding each pixel of the hyperspectral image to further obtain a high Amplitude and phase characteristics of the spectral image.
  • a hyperspectral remote sensing image is convoluted with N three-dimensional Gabor filters by the following formula to obtain i three-dimensional Gabor features Gi(x, y, b), Where R represents the hyperspectral remote sensing image, and G i represents the i-th three-dimensional Gabor feature, where i is an arbitrary integer between 1 and N.
  • the amplitude characteristics M i (x, y, b) and phase characteristics F i (x, y, b) of the hyperspectral remote sensing image are calculated by the following formula:
  • M i (x, y, b) abs(G i (x, y, b));
  • Re(G i (x, y, b)) and Im(G i (x, y, b)) are the real and imaginary parts of the Gabor feature, respectively.
  • the four three-dimensional Gabor filters shown in FIG. 3 are first acquired, and the hyperspectral remote sensing image and the generated four three-dimensional Gabor filters are further performed.
  • the convolution operation four three-dimensional Gabor filter features are obtained, and four three-dimensional Gabor amplitude features G i (x, y, b) and three-dimensional Gabor phase features F i (x, y, b) are further obtained by the above formula. ), where i takes any integer between 1 and 4.
  • 4-a is a hyperspectral remote sensing image according to an embodiment of the present invention, and two Gabor filters in four three-dimensional Gabor filters are taken as For example, the three-dimensional Gabor amplitude feature set of Figure 4-b and the three-dimensional Gabor phase feature set of Figure 4-c are obtained.
  • 4b is a three-dimensional Gabor amplitude feature set provided by an embodiment of the present invention;
  • FIG. 4-c is a phase feature set and a coding feature set of the hyperspectral remote sensing image provided by the embodiment of the present invention.
  • the first column represents the three-dimensional Gabor phase features obtained by the two three-dimensional Gabor filters; the second column represents the features obtained when the real part of the first column of the three-dimensional Gabor features is encoded; A feature obtained by encoding an imaginary part of a first column of three-dimensional Gabor features.
  • other methods may also be used to acquire the three-dimensional Gabor phase features, for example, based on a fusion coding method or a contention based coding method.
  • the determining module 730 is configured to perform feature fusion on the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine a feature category in the hyperspectral remote sensing image.
  • the preset feature fusion algorithm refers to an algorithm for combining three-dimensional Gabor amplitude features and phase features to simultaneously use three-dimensional Gabor amplitude features and three-dimensional Gabor phase features for classification of remote sensing images.
  • the determining module 730 includes:
  • An acquisition unit 731 based on any of the N-dimensional Gabor amplitude characteristic of a Gabor amplitude characteristic M i of the remote sensing image acquiring a category belongs to the confidence of P p to any category And a category similarity measure p based on a Gabor phase characteristics from any of the N-dimensional Gabor phase characteristics of the acquired remote sensing images F i and any of the categories of P P is a positive integer;
  • a determining unit 732 configured to determine the confidence level in the p categories Measure distance from the similarity
  • the category in which the sum of the squared differences under the N Gabor filters is the largest is the feature class in the hyperspectral remote sensing image.
  • confidence is between 0 and 1, which is used to reflect the probability that the hyperspectral remote sensing image belongs to the p-th category, thus the confidence level
  • the Hamming distance The value is between 0 and 1, which is used to reflect the matching degree between the hyperspectral remote sensing image and the category p, when the Hamming distance The smaller the smaller, the greater the probability that the hyperspectral remote sensing image belongs to the p-class.
  • the obtaining unit 731 is specifically configured to:
  • the D is a decision matrix of the remote sensing image acquired by the support vector machine, and the n p is the number of non-zero elements of the p-th row in the decision matrix D.
  • the decision matrix P can be obtained by assuming that there is a P-type feature, and for each test sample t, P ⁇ (P-1) are established by using a one-to-one strategy.
  • the confidence is between 0 and 1, and when the confidence is The larger the larger, the greater the probability that the hyperspectral remote sensing image belongs to category p.
  • the addition and multiplication can also be used to calculate the confidence.
  • the acquiring unit 731 is specifically configured to:
  • the similarity measure between the hyperspectral remote sensing image t and any training sample s in the training set A is obtained by the following formula:
  • the Hamming distance The value is between 0 and 1, and when Hamming distance The smaller the time, the greater the probability that the hyperspectral remote sensing image belongs to the category p, and the Hamming distance for the optimal matching of the hyperspectral remote sensing image and the category p Zero.
  • the foregoing Hamming distance may also be calculated by using a classification based on sparse representation, K-inline classification, and the like.
  • the similarity metric distance may also be other distances, such as a Levinstein distance and a Li distance.
  • the above confidence Distance from Hamming The parameters in the calculation formula are all determined parameters, so when using the above formula to calculate, no parameter estimation is needed, which will make the calculation more accurate.
  • the E p is defined as the confidence Distance from the Hamming The sum of the squared differences under the N Gabor filters, ie Because of confidence The larger the probability, the higher the probability that the hyperspectral remote sensing image belongs to the category p, when the Hamming distance The smaller the hyperspectral imagery greater probability of belonging to the category p, whereby when the three-dimensional integration and three-dimensional Gabor Gabor phase amplitude characteristic feature, the larger E p, the hyperspectral imagery greater probability of belonging to the category p, and finally The category corresponding to the value at which E p is the largest is the feature type in the hyperspectral remote sensing image.
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image can be simultaneously used in determining the feature category in the hyperspectral remote sensing image, thereby making the determined category more accurate.
  • the fusion feature value E p may also be defined in other forms, such as a square form, an exponential form, and the like.
  • the four three-dimensional Gabor filters shown in FIG. 3 are first obtained, and then the three-dimensional Gabor amplitudes shown in FIG. 4-b are calculated by using step S102.
  • the three-dimensional Gabor amplitude features shown in Figure 4-c are calculated by using step S102.
  • the hyperspectral remote sensing image 4-a is obtained as the confidence level of each class.
  • the hyperspectral remote sensing image is obtained for each class of Hamming distance.
  • FIG. 5 is a schematic diagram showing classification of hyperspectral remote sensing images according to an embodiment of the present invention.
  • the hyperspectral remote sensing image classification device 700 first acquires N preset Gabor filters, and then acquires N three-dimensional Gabor amplitudes of the hyperspectral remote sensing image based on the N preset Gabor filters. Value feature and N three-dimensional Gabor phase features, and finally feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine features in the hyperspectral remote sensing image
  • the category of the category is a preset Gabor filters.
  • the phase feature of the hyperspectral remote sensing image includes rich phase features
  • the three-dimensional Gabor amplitude feature is complementary to the three-dimensional Gabor phase feature
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase are obtained by fusing the hyperspectral remote sensing image in the embodiment of the present invention.
  • the features are used to determine the feature categories in hyperspectral remote sensing images and improve the accuracy of feature classification for hyperspectral remote sensing images.
  • the three-dimensional Gabor phase information has extremely high sensitivity to the spatial position of the ground object, the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image are used for classification, which reduces the classification robustness.
  • the hyperspectral remote sensing image classification device 700 is presented in the form of a unit.
  • a "unit” herein may refer to an application-specific integrated circuit (ASIC), a processor and memory that executes one or more software or firmware programs, integrated logic circuits, and/or other devices that provide the functionality described above. .
  • ASIC application-specific integrated circuit
  • FIG. 8 is a schematic structural diagram of a second embodiment of a hyperspectral remote sensing image classification apparatus according to an embodiment of the present invention, which is used to implement a hyperspectral remote sensing image classification method disclosed in an embodiment of the present invention.
  • the hyperspectral remote sensing image classification device 800 may include at least one bus 801, at least one processor 802 connected to the bus 801, and at least one memory 803 connected to the bus 801.
  • the processor 802 calls, by using the bus 801, code stored in the memory for acquiring N preset Gabor filters, where N is a positive integer; acquiring hyperspectral remote sensing images based on the N preset Gabor filters N three-dimensional Gabor amplitude features and N three-dimensional Gabor phase features; feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features using a preset feature fusion algorithm to determine the hyperspectral remote sensing The feature category in the image.
  • the N preset Gabor filters are Gabor filters that are parallel to a spectral dimension direction of the hyperspectral remote sensing image.
  • the processor 802 performs feature fusion on the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine
  • the feature categories in the hyperspectral remote sensing image include:
  • a category similarity measure p based on a Gabor phase characteristics from any of the N-dimensional Gabor phase characteristics of the acquired remote sensing images F i and any of the categories of P P is a positive integer;
  • Determining the confidence in the p categories Measure distance from the similarity
  • the category in which the sum of the squared differences under the N Gabor filters is the largest is the feature class in the hyperspectral remote sensing image.
  • the processor 802 based on any of the N-dimensional Gabor amplitude characteristic of a three-dimensional Gabor amplitude characteristic M i of the remote sensing images acquired categories in P Confidence in any category p include:
  • the D is a decision matrix of the remote sensing image acquired by the support vector machine, and the n p is the number of non-zero elements of the p-th row in the decision matrix D.
  • the processor 802 is based on any three-dimensional Gabor phase feature F of the N three-dimensional Gabor phase features. i acquires the similarity Hyperspectral image P with any of the categories in a category of distance measure p include:
  • the similarity measure between the hyperspectral remote sensing image t and any training sample s in the training set A is obtained by the following formula:
  • the hyperspectral remote sensing image classification device 800 first acquires N preset Gabor filters, and then acquires N three-dimensional Gabor amplitudes of the hyperspectral remote sensing image based on the N preset Gabor filters. Value feature and N three-dimensional Gabor phase features, and finally feature fusion of the N three-dimensional Gabor amplitude features and the N three-dimensional Gabor phase features by using a preset feature fusion algorithm to determine features in the hyperspectral remote sensing image
  • the category of the category is a preset Gabor filters.
  • the phase feature of the hyperspectral remote sensing image includes rich phase features
  • the three-dimensional Gabor amplitude feature is complementary to the three-dimensional Gabor phase feature
  • the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase are obtained by fusing the hyperspectral remote sensing image in the embodiment of the present invention.
  • the features are used to determine the feature categories in hyperspectral remote sensing images and improve the accuracy of feature classification for hyperspectral remote sensing images.
  • the three-dimensional Gabor phase information has extremely high sensitivity to the spatial position of the ground object, the three-dimensional Gabor amplitude feature and the three-dimensional Gabor phase feature of the hyperspectral remote sensing image are used for classification, which reduces the classification robustness.
  • the hyperspectral remote sensing image classification device 800 is presented in the form of a unit.
  • a "unit” herein may refer to an application-specific integrated circuit (ASIC), a processor and memory that executes one or more software or firmware programs, integrated logic circuits, and/or other devices that provide the functionality described above. .
  • ASIC application-specific integrated circuit
  • the embodiment of the present invention further provides a computer storage medium, wherein the computer storage medium can store a program, and the program includes some or all of the steps of any hyperspectral remote sensing image classification method described in the foregoing method embodiments.
  • the disclosed apparatus may be implemented in other ways.
  • the device embodiments described above are merely illustrative.
  • the division of the unit is only a logical function division.
  • there may be another division manner for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not executed.
  • the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be electrical or otherwise.
  • the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.
  • each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.
  • the above integrated unit can be implemented in the form of hardware or in the form of a software functional unit.
  • the integrated unit if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention is essential or the part contributing to the prior art or the entire technical solution.
  • the portion or portion may be embodied in the form of a software product stored in a storage medium, including instructions for causing a computer device (which may be a personal computer, server or network device, etc.) to perform various embodiments of the present invention. All or part of the steps of the method described.
  • the foregoing storage medium includes: a U disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk, and the like. .

Landscapes

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

Abstract

一种高光谱遥感图像分类方法及装置，所述方法包括：获取N个预设Gabor滤波器(S101)；基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征(S102)；利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中地物的类别(S103)。所述方法通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中的地物类别，从而提高对高光谱遥感图像的地物分类准确度。

Description

一种高光谱遥感图像分类方法及装置
本申请要求于2017年4月21日递交国家知识产权局、申请号为201710266000.X，发明名称为“一种高光谱遥感图像分类方法及装置”的国内专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本发明涉及图像处理领域，具体涉及一种高光谱遥感图像分类方法及装置。
背景技术
高光谱遥感图像是指由传感器在电磁波谱的可见光，近红外，中红外和热红外波段范围内，在不同波段成像获得的高光谱图像数据。因此，高光谱遥感图像包含了丰富的空间、辐射和光谱三重信息，为地表物质的精细分类和识别提供了可能。
目前，在对地表物质进行分类时，由于三维Gabor特征的幅值信息具有稳定性好的优点，所以通常被直接用于分类，但由于对于高光谱遥感图像来说，三维Gabor特征中包含了丰富的相位特征，所以仅采用Gabor特征的幅值特征对地表特征进行分类将使得分类准确率不高。
发明内容
本发明实施例提供了一种高光谱遥感图像分类方法及装置，以期可以提高地物分类准确率。
第一方面，本发明实施例提供一种高光谱遥感图像分类方法，其特征在于，所述方法包括：
获取N个预设Gabor滤波器，所述N为正整数；
基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征；
利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
第二方面，本发明实施例提供一种高光谱遥感图像分类装置，其特征在于，所述装置包括：
获取模块，用于获取N个预设Gabor滤波器，所述N为正整数；
提取模块，用于基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征；
确定模块，用于利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
可以看出，本发明实施例所提供的技术方案中，首先获取N个预设Gabor滤波器，然后再基于该N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征，最后利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中地物的所属类别。由于高光谱遥感图像的相位特征包括了丰富的相位特征，并且三维Gabor幅值特征与三维Gabor相位特征互补，所以在本发明实施例中通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中的地物类别，提高对高光谱遥感图像的地物分类准确度。
进一步地，由于三维Gabor相位信息对地物的空间位置具有极高的敏感上，所以通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于分类，降低了分类鲁棒性。
附图说明
为了更清楚地说明本发明实施例中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种高光谱遥感图像分类方法的第一实施例流程示意图；
图2是本发明实施例提供的一种三维Gabor特征的频率域关系示意图；
图3是本发明实施例提供的从三个不同视角观察到的平行于光谱维度方向的三维Gabor滤波器；
图4-a为本发明实施例提供的高光谱遥感图像；
图4-b是本发明实施例提供的三维Gabor幅值特征集；
图4-c是本发明实施例提供的高光谱遥感图像的相位特征集及编码特征集；
图5示出了本发明实施例所提供的高光谱遥感图像的分类示意图；
图6是本发明实施例提供的一种高光谱遥感图像分类方法的第二实施例流程示意图；
图7是本发明实施例提供的一种高光谱遥感图像分类装置的第一实施例的结构示意图；
图8是本发明实施例提供的一种高光谱遥感图像分类装置的第二实施例的结构示意图。
具体实施方式
本发明实施例提供了一种高光谱遥感图像分类方法及装置，以期可以提高地物分类准确率。
为了使本技术领域的人员更好地理解本发明方案，下面将结合本发明实施例中的附图，对本发明实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本发明一部分的实施例，而不是全部的实施例。基于本发明中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都应当属于本发明保护的范围。
本发明的说明书和权利要求书及上述附图中的术语“第一”、“第二”和“第三”等是用于区别不同对象，而非用于描述特定顺序。此外，术语“包括”以及它们任何变形，意图在于覆盖不排他的包含。例如包含了一系列步骤或单元的过程、方法、系统、产品或设备没有限定于已列出的步骤或单元，而是可选地还包括没有列出的步骤或单元，或可选地还包括对于这些过程、方法、产 品或设备固有的其它步骤或单元。
本发明实施例提供的一种高光谱遥感图像分类方法，包括：
获取N个预设Gabor滤波器，所述N为正整数；基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个Gabor幅值特征与N个三维Gabor相位特征；利用预设特征融合算法对所述N个Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
参见图1，图1是本发明实施例提供的一种高光谱遥感图像分类方法的第一实施例流程示意图。如图1所示，本发明实施例提供的高光谱遥感图像分类方法包括以下步骤：
S101、获取N个预设Gabor滤波器。
其中，所述N为正整数，该N个预设Gabor滤波器用于后续进行高光谱遥感图像的幅值特征和相位特征的提取。
可选地，在本发明的一个实施例中，该N个预设Gabor滤波器为平行于所述高光谱遥感图像的光谱维度方向的Gabor滤波器。
可以理解，由于相位信息对空间位置的敏感性，不同Gabor滤波器获得的Gabor相位特征对于地物的分类能力存在巨大差异，不能全部用于后续的编码和分类过程，并且如果不加选择地利用所有的Gabor滤波器用于高光谱遥感图像的分类，将使得计算量非常大，所以仅选择平行于高光谱遥感图像的光谱维度方向的Gabor滤波器提取的特征用于高光谱遥感图像的分类将极大地提高算法的计算效率。
具体地，在本发明的一个实施例中，可以通过以下公式获取N个平行于高光谱遥感图像的光谱维度方向的Gabor滤波器
Figure PCTCN2017083766-appb-000001
Figure PCTCN2017083766-appb-000002
其中，f是预设Gabor滤波器的频率，
Figure PCTCN2017083766-appb-000003
是所述预设Gabor滤波器与ω轴的夹角，θ是预设Gabor滤波器与u-v平面的夹角，
Figure PCTCN2017083766-appb-000004
表示Gabor滤波器的方向，(x,y,b)分别表示高光谱遥感图像的像素的x坐标、y坐标以及光谱坐标， σ是高斯包络的宽度。
具体地，在本发明的一个示例中，若
Figure PCTCN2017083766-appb-000005
θ＝π/2，参见图2，图2是本发明实施例提供的一种三维Gabor特征的频率域关系示意图，由图2可见，该频率f所指的方向即为高光谱遥感图像的光谱维度方向。
具体地，在本发明的一个示例中，若设置频率fj＝[0.5,0.25,0.125,0.0625]，则可以得到4个平行于高光谱遥感图像的三维Gabor滤波器{Ψi,i＝1,…,4}，用于后续三维Gabor特征提取。具体可参见图3，图3是本发明实施例提供的从三个不同视角观察到的平行于光谱维度方向的三维Gabor滤波器。
可选地，在本发明的一些实施例中，该Gabor滤波器可以为Log-Gabor滤波器，haar-Gabor滤波器。
S102、基于所述N个预设Gabor滤波器提取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征。
可选地，在本发明的一个实施例中，通过将高光谱遥感图像与N个预设Gabor滤波器进行卷积运算得到三维Gabor特征，然后对高光谱图像的每一个像素进行编码进一步得到高光谱图像的幅值特征和相位特征。
具体地，首先通过如下公式将高光谱遥感图像与N个三维Gabor滤波器进行卷积操作得到i个三维Gabor特征
Figure PCTCN2017083766-appb-000006
其中，R表示该高光谱遥感图像，Gi表示第i个三维Gabor特征，其中，i的取值为1至N之间的任意整数。
当得到N个三维Gabor特征后，再利用如下公式计算该高光谱遥感图像的幅值特征Mi(x,y,b)与相位特征Fi(x,y,b)：
Mi(x,y,b)＝abs(Gi(x,y,b))；
Figure PCTCN2017083766-appb-000007
Figure PCTCN2017083766-appb-000008
其中，Re(Gi(x,y,b))和Im(Gi(x,y,b))分别是Gabor特征的实部和虚部。
举例说明，在本发明的一个示例中，若N的取值为4，首先获取到图3所 示的4个三维Gabor滤波器，再将高光谱遥感图像与生成的4个三维Gabor滤波器进行卷积运算后，将得到4个三维Gabor滤波器特征，并进一步通过上述公式得到4个三维Gabor幅值特征Gi(x,y,b)以及三维Gabor相位特征Fi(x,y,b)，其中，i的取值为1至4之间的任一整数。具体可参见图4-a，图4-b，以及图4-c，图4-a为本发明实施例提供的高光谱遥感图像，取4个三维Gabor滤波器中的两个Gabor滤波器为例，得到图4-b的三维Gabor幅值特征集以及图4-c的三维Gabor相位特征集。其中，图4-b是本发明实施例提供的三维Gabor幅值特征集；图4-c是本发明实施例提供的高光谱遥感图像的相位特征集及编码特征集。在图4-c中，第一列表示该两个三维Gabor滤波器得到的三维Gabor相位特征；第二列表示当对第一列三维Gabor特征中的实部编码得到的特征；第三列表示对第一列三维Gabor特征中的虚部编码得到的特征。
可以理解，上述三维Gabor幅值特征与三维Gabor相位特征在计算过程中不需要训练样本的参与，所以可以提高该方案的实用性。
可选地，在本发明的另一些实施例中，也可以使用其它方式来获取三维Gabor相位特征，例如，基于融合编码的方式或基于竞争编码的方式。
S103、利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
其中，该预设特征融合算法是指用于对三维Gabor幅值特征和相位特征进行融合，以能同时将三维Gabor幅值特征与三维Gabor相位特征用于遥感图像的分类的一种算法。
可选地，在本发明的一个实施例中，所述利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别，包括：
基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000009
以及基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的相似性度量距离
Figure PCTCN2017083766-appb-000010
P为正整数；
确定所述p个类别中所述置信度
Figure PCTCN2017083766-appb-000011
与所述相似性度量距离
Figure PCTCN2017083766-appb-000012
在所述N个 Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
其中，置信度
Figure PCTCN2017083766-appb-000013
取值在0至1之间，用于反映该高光谱遥感图像属于第p类别的概率，从而当置信度
Figure PCTCN2017083766-appb-000014
越大，该高光谱遥感图像属于第p类的概率越大，汉明距离
Figure PCTCN2017083766-appb-000015
取值在0至1之间，用于反映高光谱遥感图像与类别p之间的匹配度，当汉明距离
Figure PCTCN2017083766-appb-000016
越小，该高光谱遥感图像属于第p类的概率越大。
可选地，在本发明的一个实施例中，所述基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000017
包括：
利用以下公式基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000018
Figure PCTCN2017083766-appb-000019
其中，所述D为基于支持向量机获取的所述遥感图像的决策矩阵，所述np为所述决策矩阵D中第p行非零元素的个数。
具体地，在本发明的一个实施例中，可以通过如下方式获取该决策矩阵P：假设有P类地物，针对每一个测试样本t，利用一对一策略建立P×(P-1)个支持向量机的分类器。其中，对于任意两个类别C1和C2，通过投票可以得到决策值δ；进一步建立决策矩阵D，其中当δ>0时，DC1C2＝δ，否则DC1C2＝-δ，决策矩阵D中的其他元素为零。由于支持向量机在用于分类上的良好效果，所以基于支持向量机的方式来确定置信度
Figure PCTCN2017083766-appb-000020
有效提升了融合决策的精度。
可以看出，该置信度
Figure PCTCN2017083766-appb-000021
的取值为0至1之间，并且当置信度
Figure PCTCN2017083766-appb-000022
越大时，该高光谱遥感图像属于类别p的概率越大。
可选地，在本发明的另一些实施例中，也可以使用相加、相乘的方式来计算置信度
Figure PCTCN2017083766-appb-000023
可选地，在本发明的一个实施例中，该相似性度量距离
Figure PCTCN2017083766-appb-000024
为汉明距离，所述基于所述N个三维Gabor相位特征中的任一三维Gabor相位特征Fi获取所述高光谱遥感图像与所述P个类别中的任一类别p的相似性度量距离
Figure PCTCN2017083766-appb-000025
包括：
利用以下公式获取所述高光谱遥感图像t和训练集合A中任意训练样本s之间的相似性度量：
Figure PCTCN2017083766-appb-000026
其中，B为高光谱图像的光谱维度；
获取所述高光谱遥感图像t与所述任一类别p之间的相似性度量距离
Figure PCTCN2017083766-appb-000027
为：
Figure PCTCN2017083766-appb-000028
可以看出，该汉明距离
Figure PCTCN2017083766-appb-000029
的取值为0至1之间，并且当汉明距离
Figure PCTCN2017083766-appb-000030
越小时，该高光谱遥感图像属于类别p的概率越大，对于高光谱遥感图像与类别p的最优匹配下，汉明距离
Figure PCTCN2017083766-appb-000031
为零。
可选地，在本发明的另一些实施例中，也可以使用基于稀疏表示的分类、K紧邻分类等方式来计算上述汉明距离
Figure PCTCN2017083766-appb-000032
可选地，在本发明的另一些实施例中，该相似性度量距离也可以其它距离，例如，莱文斯坦距离，李距离。
可以理解，上述置信度
Figure PCTCN2017083766-appb-000033
与汉明距离
Figure PCTCN2017083766-appb-000034
的计算公式中各参数均为确定的参数，所以在利用上述公式计算时，不需要进行参数估计，将使得计算更为准确。
可以看出，若定义融合特征值为Ep，该Ep定义为所述置信度
Figure PCTCN2017083766-appb-000035
与所述汉明距离
Figure PCTCN2017083766-appb-000036
在所述N个Gabor滤波器下的平方差之和，也即
Figure PCTCN2017083766-appb-000037
由于当置信度
Figure PCTCN2017083766-appb-000038
越大时，该高光谱遥感图像属于类别p的概率越大，当汉明距离
Figure PCTCN2017083766-appb-000039
越小，该高光谱遥感图像属于类别p的概率越大，从而当融合三维Gabor幅值特征和三维Gabor相位特征后，Ep越大，该高光谱遥感图像属于类别p的概率越大，最后将得到Ep最大的值所对应的类别即为该高光谱遥感图像中的地物类别。可以理解，通过上述方式，可以在确定高光谱遥感图像中的地物类别时，同时使用高光谱遥感图像的三维Gabor幅值特征与三维Gabor 相位特征，从而使得所确定的类别更为准确。
可选地，在本发明的另一些实施例中，该融合特征值Ep也可以定义为其它形式，例如，平方形式、指数形式等。
具体地，在本发明的一个实施例中，若取N＝4，首先获得图3所示的4个三维Gabor滤波器，然后再利用步骤S102计算得到图4-b所示的三维Gabor幅值特征与图4-c所示的三维Gabor相位特征。进一步地，若通过支持向量机对图4-b所示的Gabor幅值特征进行分类，获得高光谱遥感图像4-a属于每一类的置信度
Figure PCTCN2017083766-appb-000040
通过正则化的汉明距离，利用最近邻策略对图4-c所示的Gabor相位编码特征进行分类，获得高光谱遥感图像属于每一类的汉明距离
Figure PCTCN2017083766-appb-000041
然后再利用Ep融合置信度
Figure PCTCN2017083766-appb-000042
和汉明距离
Figure PCTCN2017083766-appb-000043
以对高光谱遥感图像进行分类，确定该高光谱遥感图像属于哪个地物类别。参见图5，图5示出了本发明实施例所提供的高光谱遥感图像的分类示意图。
可以看出，本实施例的方案中，首先获取N个预设Gabor滤波器，然后再基于该N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征，最后利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。由于高光谱遥感图像的相位特征包括了丰富的相位特征，并且三维Gabor幅值特征与三维Gabor相位特征互补，所以在本发明实施例通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中地物的所属类别，提高对高光谱遥感图像的地物分类准确度。
进一步地，由于三维Gabor相位信息对地物的空间位置具有极高的敏感上，所以通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于分类，降低了分类鲁棒性。
参见图6，图6是本发明实施例提供的一种高光谱遥感图像分类方法的第二实施例流程示意图。图6所示的方法中，与图1所示方法相同或类似的内容可以参考图1中的详细描述，此处不再赘述。如图6所示，本发明实施例提供的高光谱遥感图像分类方法包括以下步骤：
S601、获取N个预设Gabor滤波器。
具体地，可以获取到4个平行于高光谱遥感图像的光谱维度方向的Gabor滤波器。
S602、基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征。
S603、基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000044
S604、基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的汉明距离
Figure PCTCN2017083766-appb-000045
S605、确定所述p个类别中所述置信度
Figure PCTCN2017083766-appb-000046
与所述相似性度量距离
Figure PCTCN2017083766-appb-000047
在所述N个Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
需要说明，上述步骤S603和S604之间没有严格的先后顺序。
可以看出，本实施例的方案中，获取目标图像，首先获取N个预设Gabor滤波器，然后再基于该N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征，最后利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中地物的所属类别。由于高光谱遥感图像的相位特征包括了丰富的相位特征，并且三维Gabor幅值特征与三维Gabor相位特征互补，所以在本发明实施例通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中的地物类别，提高对高光谱遥感图像的地物分类准确度。
进一步地，由于三维Gabor相位信息对地物的空间位置具有极高的敏感上，所以通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于分类，降低了分类鲁棒性。
本发明实施例还提供一种高光谱遥感图像分类装置，包括：
获取模块，用于获取N个预设Gabor滤波器，所述N为正整数；
提取模块，用于基于所述N个预设Gabor滤波器获取高光谱遥感图像的N 个三维Gabor幅值特征与N个三维Gabor相位特征；
确定模块，用于利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
具体地，请参见图7，图7是本发明实施例提供的一种高光谱遥感图像分类装置的第一实施例的结构示意图，用于实现本发明实施例公开的一种高光谱遥感图像分类方法。其中，如图7所示，本发明实施例提供的一种高光谱遥感图像分类装置700可以包括：
获取模块710、提取模块720和确定模块730。
其中，获取模块710，用于获取N个预设Gabor滤波器，所述N为正整数。
其中，所述N为正整数，该N个预设Gabor滤波器用于后续进行高光谱遥感图像的幅值特征和相位特征的提取。
可选地，在本发明的一个实施例中，该N个预设Gabor滤波器为平行于所述高光谱遥感图像的光谱维度方向的Gabor滤波器。
可以理解，由于相位信息对空间位置的敏感性，不同Gabor滤波器获得的Gabor相位特征对于地物的分类能力存在巨大差异，不能全部用于后续的编码和分类过程，并且如果不加选择地利用所有的Gabor滤波器用于高光谱遥感图像的分类，将使得计算量非常大，所以仅选择平行于高光谱遥感图像的光谱维度方向的Gabor滤波器提取的特征用于高光谱遥感图像的分类将极大地提高算法的计算效率。
具体地，在本发明的一个实施例中，可以通过以下公式获取N个平行于高光谱遥感图像的光谱维度方向的Gabor滤波器
Figure PCTCN2017083766-appb-000048
Figure PCTCN2017083766-appb-000049
其中，f是预设Gabor滤波器的频率，
Figure PCTCN2017083766-appb-000050
是所述预设Gabor滤波器与ω轴的夹角，θ是预设Gabor滤波器与u-v平面的夹角，
Figure PCTCN2017083766-appb-000051
表示Gabor滤波器的方向，(x,y,b)分别表示高光谱遥感图像的像素的x坐标、y坐标以及光谱坐标， σ是高斯包络的宽度。
具体地，在本发明的一个示例中，若
Figure PCTCN2017083766-appb-000052
θ＝π/2，参见图2，图2是本发明实施例提供的一种三维Gabor特征的频率域关系示意图，由图2可见，该频率f所指的方向即为高光谱遥感图像的光谱维度方向。
具体地，在本发明的一个示例中，若设置频率fj＝[0.5,0.25,0.125,0.0625]，则可以得到4个平行于高光谱遥感图像的三维Gabor滤波器{Ψi,i＝1,…,4}，用于后续三维Gabor特征提取。具体可参见图3，图3是本发明实施例提供的从三个不同视角观察到的平行于光谱维度方向的三维Gabor滤波器。
可选地，在本发明的一些实施例中，该Gabor滤波器可以为Log-Gabor滤波器，haar-Gabor滤波器。
提取模块720，用于基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征。
可选地，在本发明的一个实施例中，通过将高光谱遥感图像与N个预设Gabor滤波器进行卷积运算得到三维Gabor特征，然后对高光谱图像的每一个像素进行编码进一步得到高光谱图像的幅值特征和相位特征。
具体地，首先通过如下公式将高光谱遥感图像与N个三维Gabor滤波器进行卷积操作得到i个三维Gabor特征Gi(x,y,b)，
Figure PCTCN2017083766-appb-000053
其中，R表示该高光谱遥感图像，Gi表示第i个三维Gabor特征，其中，i的取值为1至N之间的任意整数。
当得到N个三维Gabor特征后，再利用如下公式计算该高光谱遥感图像的幅值特征Mi(x,y,b)与相位特征Fi(x,y,b)：
Mi(x,y,b)＝abs(Gi(x,y,b))；
Figure PCTCN2017083766-appb-000054
Figure PCTCN2017083766-appb-000055
其中，Re(Gi(x,y,b))和Im(Gi(x,y,b))分别是Gabor特征的实部和虚部。
举例说明，在本发明的一个示例中，若N的取值为4，首先获取到图3所 示的4个三维Gabor滤波器，再将高光谱遥感图像与生成的4个三维Gabor滤波器进行卷积运算后，将得到4个三维Gabor滤波器特征，并进一步通过上述公式得到4个三维Gabor幅值特征Gi(x,y,b)以及三维Gabor相位特征Fi(x,y,b)，其中，i的取值为1至4之间的任一整数。具体可参见图4-a，图4-b，以及图4-c，图4-a为本发明实施例提供的高光谱遥感图像，取4个三维Gabor滤波器中的两个Gabor滤波器为例，得到图4-b的三维Gabor幅值特征集以及图4-c的三维Gabor相位特征集。其中，图4-b是本发明实施例提供的三维Gabor幅值特征集；图4-c是本发明实施例提供的高光谱遥感图像的相位特征集及编码特征集。在图4-c中，第一列表示该两个三维Gabor滤波器得到的三维Gabor相位特征；第二列表示当对第一列三维Gabor特征中的实部编码得到的特征；第三列表示对第一列三维Gabor特征中的虚部编码得到的特征。
可以理解，上述三维Gabor幅值特征与三维Gabor相位特征在计算过程中不需要训练样本的参与，所以可以提高该方案的实用性。
可选地，在本发明的另一些实施例中，也可以使用其它方式来获取三维Gabor相位特征，例如，基于融合编码的方式或基于竞争编码的方式。
确定模块730，用于利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
其中，该预设特征融合算法是指用于对三维Gabor幅值特征和相位特征进行融合，以能同时将三维Gabor幅值特征与三维Gabor相位特征用于遥感图像的分类的一种算法。
可选地，在本发明的一个实施例中，所述确定模块730包括：
获取单元731，用于基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000056
以及基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的相似性度量距离
Figure PCTCN2017083766-appb-000057
P为正整数；
确定单元732，用于确定所述p个类别中所述置信度
Figure PCTCN2017083766-appb-000058
与所述相似性度量距离
Figure PCTCN2017083766-appb-000059
在所述N个Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
其中，置信度
Figure PCTCN2017083766-appb-000060
取值在0至1之间，用于反映该高光谱遥感图像属于第p类别的概率，从而当置信度
Figure PCTCN2017083766-appb-000061
越大，该高光谱遥感图像属于第p类的概率越大，汉明距离
Figure PCTCN2017083766-appb-000062
取值在0至1之间，用于反映高光谱遥感图像与类别p之间的匹配度，当汉明距离
Figure PCTCN2017083766-appb-000063
越小，该高光谱遥感图像属于第p类的概率越大。
可选地，在本发明的一个实施例中，所述获取单元731具体用于：
利用以下公式基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000064
Figure PCTCN2017083766-appb-000065
其中，所述D为基于支持向量机获取的所述遥感图像的决策矩阵，所述np为所述决策矩阵D中第p行非零元素的个数。
具体地，在本发明的一个实施例中，可以通过如下方式获取该决策矩阵P：假设有P类地物，针对每一个测试样本t，利用一对一策略建立P×(P-1)个支持向量机的分类器。其中，对于任意两个类别C1和C2，通过投票可以得到决策值δ；进一步建立决策矩阵D，其中当δ>0时，DC1C2＝δ，否则DC1C2＝-δ，决策矩阵D中的其他元素为零。由于支持向量机在用于分类上的良好效果，所以基于支持向量机的方式来确定置信度
Figure PCTCN2017083766-appb-000066
有效提升了融合决策的精度。
可以看出，该置信度
Figure PCTCN2017083766-appb-000067
的取值为0至1之间，并且当置信度
Figure PCTCN2017083766-appb-000068
越大时，该高光谱遥感图像属于类别p的概率越大。
可选地，在本发明的另一些实施例中，也可以使用相加、相乘的方式来计算置信度
Figure PCTCN2017083766-appb-000069
可选地，在本发明的一个实施例中，若所述相似性度量距离包括汉明距离，所述获取单元731具体用于：
利用以下公式获取所述高光谱遥感图像t和训练集合A中任意训练样本s之间的相似性度量：
Figure PCTCN2017083766-appb-000070
其中，B为原始高光谱数据的光谱维度；
获取所述高光谱遥感图像t与所述任一类别p之间的汉明距离
Figure PCTCN2017083766-appb-000071
为：
Figure PCTCN2017083766-appb-000072
可以看出，该汉明距离
Figure PCTCN2017083766-appb-000073
的取值为0至1之间，并且当汉明距离
Figure PCTCN2017083766-appb-000074
越小时，该高光谱遥感图像属于类别p的概率越大，对于高光谱遥感图像与类别p的最优匹配下，汉明距离
Figure PCTCN2017083766-appb-000075
为零。
可选地，在本发明的另一些实施例中，也可以使用基于稀疏表示的分类、K紧邻分类等方式来计算上述汉明距离
Figure PCTCN2017083766-appb-000076
可选地，在本发明的另一些实施例中，该相似性度量距离也可以其它距离，例如，莱文斯坦距离，李距离。
可以理解，上述置信度
Figure PCTCN2017083766-appb-000077
与汉明距离
Figure PCTCN2017083766-appb-000078
的计算公式中各参数均为确定的参数，所以在利用上述公式计算时，不需要进行参数估计，将使得计算更为准确。
可以看出，若定义融合特征值为Ep，该Ep定义为所述置信度
Figure PCTCN2017083766-appb-000079
与所述汉明距离
Figure PCTCN2017083766-appb-000080
在所述N个Gabor滤波器下的平方差之和，也即
Figure PCTCN2017083766-appb-000081
由于当置信度
Figure PCTCN2017083766-appb-000082
越大时，该高光谱遥感图像属于类别p的概率越大，当汉明距离
Figure PCTCN2017083766-appb-000083
越小，该高光谱遥感图像属于类别p的概率越大，从而当融合三维Gabor幅值特征和三维Gabor相位特征后，Ep越大，该高光谱遥感图像属于类别p的概率越大，最后将得到Ep最大的值所对应的类别即为该高光谱遥感图像中的地物类别。可以理解，通过上述方式，可以在确定高光谱遥感图像中的地物类别时，同时使用高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征，从而使得所确定的类别更为准确。
可选地，在本发明的另一些实施例中，该融合特征值Ep也可以定义为其它形式，例如，平方形式、指数形式等。
具体地，在本发明的一个实施例中，若取N＝4，首先获得图3所示的4个三维Gabor滤波器，然后再利用步骤S102计算得到图4-b所示的三维Gabor幅值特征与图4-c所示的三维Gabor相位特征。进一步地，若通过支持向量机对图4-b所示的Gabor幅值特征进行分类，获得高光谱遥感图像4-a属于每一类的置信度
Figure PCTCN2017083766-appb-000084
通过正则化的汉明距离，利用最近邻策略对图4-c所示的Gabor 相位编码特征进行分类，获得高光谱遥感图像属于每一类的汉明距离
Figure PCTCN2017083766-appb-000085
然后再利用Ep融合置信度
Figure PCTCN2017083766-appb-000086
和汉明距离
Figure PCTCN2017083766-appb-000087
以对高光谱遥感图像进行分类，确定该高光谱遥感图像属于哪个地物类别。参见图5，图5示出了本发明实施例所提供的高光谱遥感图像的分类示意图。
可以看出，本实施例的方案中，高光谱遥感图像分类装置700首先获取N个预设Gabor滤波器，然后再基于该N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征，最后利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中地物的所属类别。由于高光谱遥感图像的相位特征包括了丰富的相位特征，并且三维Gabor幅值特征与三维Gabor相位特征互补，所以在本发明实施例通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中的地物类别，提高对高光谱遥感图像的地物分类准确度。
进一步地，由于三维Gabor相位信息对地物的空间位置具有极高的敏感上，所以通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于分类，降低了分类鲁棒性。
在本实施例中，高光谱遥感图像分类装置700是以单元的形式来呈现。这里的“单元”可以指特定应用集成电路(application-specific integrated circuit，ASIC)，执行一个或多个软件或固件程序的处理器和存储器，集成逻辑电路，和/或其他可以提供上述功能的器件。
可以理解的是，本实施例的高光谱遥感图像分类装置700的各功能单元的功能可根据上述方法实施例中的方法具体实现，其具体实现过程可以参照上述方法实施例的相关描述，此处不再赘述。
参见图8，图8是本发明实施例提供的一种高光谱遥感图像分类装置的第二实施例的结构示意图，用于实现本发明实施例公开的高光谱遥感图像分类方法。其中，该高光谱遥感图像分类装置800可以包括：至少一个总线801、与总线801相连的至少一个处理器802以及与总线801相连的至少一个存储器803。
其中，处理器802通过总线801，调用存储器中存储的代码以用于获取N个预设Gabor滤波器，所述N为正整数；基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征；利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
可选地，在本发明的一些可能的实施方式中，所述N个预设Gabor滤波器为平行于所述高光谱遥感图像的光谱维度方向的Gabor滤波器。
可选地，在本发明的一些可能的实施方式中，所述处理器802利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别，包括：
基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000088
以及基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的相似性度量距离
Figure PCTCN2017083766-appb-000089
P为正整数；
确定所述p个类别中所述置信度
Figure PCTCN2017083766-appb-000090
与所述相似性度量距离
Figure PCTCN2017083766-appb-000091
在所述N个Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
可选地，在本发明的一些可能的实施方式中，所述处理器802基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000092
包括：
利用以下公式基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
Figure PCTCN2017083766-appb-000093
Figure PCTCN2017083766-appb-000094
其中，所述D为基于支持向量机获取的所述遥感图像的决策矩阵，所述np为所述决策矩阵D中第p行非零元素的个数。
可选地，在本发明的一些可能的实施方式中，若所述相似性度量距离为汉明距离，所述处理器802基于所述N个三维Gabor相位特征中的任一三维Gabor相位特征Fi获取所述高光谱遥感图像与所述P个类别中的任一类别p的相似性 度量距离
Figure PCTCN2017083766-appb-000095
包括：
利用以下公式获取所述高光谱遥感图像t和训练集合A中任意训练样本s之间的相似性度量：
Figure PCTCN2017083766-appb-000096
其中，B为原始高光谱数据的光谱维度；
获取所述高光谱遥感图像t与所述任一类别p之间的汉明距离
Figure PCTCN2017083766-appb-000097
为：
Figure PCTCN2017083766-appb-000098
可以看出，本实施例的方案中，高光谱遥感图像分类装置800首先获取N个预设Gabor滤波器，然后再基于该N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征，最后利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中地物的所属类别。由于高光谱遥感图像的相位特征包括了丰富的相位特征，并且三维Gabor幅值特征与三维Gabor相位特征互补，所以在本发明实施例通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于确定高光谱遥感图像中的地物类别，提高对高光谱遥感图像的地物分类准确度。
进一步地，由于三维Gabor相位信息对地物的空间位置具有极高的敏感性，所以通过融合高光谱遥感图像的三维Gabor幅值特征与三维Gabor相位特征用于分类，降低了分类鲁棒性。
在本实施例中，高光谱遥感图像分类装置800是以单元的形式来呈现。这里的“单元”可以指特定应用集成电路(application-specific integrated circuit，ASIC)，执行一个或多个软件或固件程序的处理器和存储器，集成逻辑电路，和/或其他可以提供上述功能的器件。
可以理解的是，本实施例的高光谱遥感图像分类装置800的各功能单元的功能可根据上述方法实施例中的方法具体实现，其具体实现过程可以参照上述方法实施例的相关描述，此处不再赘述。
本发明实施例还提供一种计算机存储介质，其中，该计算机存储介质可存储有程序，该程序执行时包括上述方法实施例中记载的任何高光谱遥感图像分类方法的部分或全部步骤。
需要说明的是，对于前述的各方法实施例，为了简单描述，故将其都表述为一系列的动作组合，但是本领域技术人员应该知悉，本发明并不受所描述的动作顺序的限制，因为依据本发明，某些步骤可以采用其他顺序或者同时进行。其次，本领域技术人员也应该知悉，说明书中所描述的实施例均属于优选实施例，所涉及的动作和模块并不一定是本发明所必须的。
在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详述的部分，可以参见其他实施例的相关描述。
在本申请所提供的几个实施例中，应该理解到，所揭露的装置，可通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本发明的各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件功能单元的形式实现。
所述集成的单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本发明的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全 部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可为个人计算机、服务器或者网络设备等)执行本发明各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、移动硬盘、磁碟或者光盘等各种可以存储程序代码的介质。
以上所述，以上实施例仅用以说明本发明的技术方案，而非对其限制；尽管参照前述实施例对本发明进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本发明各实施例技术方案的范围。

Claims (10)

  1. 一种高光谱遥感图像分类方法，其特征在于，所述方法包括：
    获取N个预设Gabor滤波器，所述N为正整数；
    基于所述N个预设Gabor滤波器提取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征；
    利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
  2. 根据权利要求1所述的方法，其特征在于，所述N个预设Gabor滤波器为平行于所述高光谱遥感图像的光谱维度方向的Gabor滤波器。
  3. 根据权利要求2所述的方法，其特征在于，所述利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别，包括：
    基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
    Figure PCTCN2017083766-appb-100001
    以及基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的相似性度量距离
    Figure PCTCN2017083766-appb-100002
    P为正整数；
    确定所述p个类别中所述置信度
    Figure PCTCN2017083766-appb-100003
    与所述相似性度量距离
    Figure PCTCN2017083766-appb-100004
    在所述N个Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
  4. 根据权利要求3所述的方法，其特征在于，所述基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
    Figure PCTCN2017083766-appb-100005
    包括：
    利用以下公式基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
    Figure PCTCN2017083766-appb-100006
    Figure PCTCN2017083766-appb-100007
    其中，所述D为基于支持向量机获取的所述遥感图像的决策矩阵，所述np为所述决策矩阵D中第p行非零元素的个数。
  5. 根据权利要求3所述的方法，其特征在于，所述相似性度量距离包括汉明距离，所述基于所述N个三维Gabor相位特征中的任一三维Gabor相位特征Fi获取所述高光谱遥感图像与所述P个类别中的任一类别p的相似性度量距离
    Figure PCTCN2017083766-appb-100008
    包括：
    利用以下公式获取所述高光谱遥感图像t和训练集合A中任意训练样本s之间的相似性度量：
    Figure PCTCN2017083766-appb-100009
    其中，所述B为所述高光谱图像的光谱维度；
    获取所述高光谱遥感图像t与所述任一类别p之间的汉明距离
    Figure PCTCN2017083766-appb-100010
    为：
    Figure PCTCN2017083766-appb-100011
  6. 一种高光谱遥感图像分类装置，其特征在于，所述装置包括：
    获取模块，用于获取N个预设Gabor滤波器，所述N为正整数；
    提取模块，用于基于所述N个预设Gabor滤波器获取高光谱遥感图像的N个三维Gabor幅值特征与N个三维Gabor相位特征；
    确定模块，用于利用预设特征融合算法对所述N个三维Gabor幅值特征与所述N个三维Gabor相位特征进行特征融合以确定所述高光谱遥感图像中的地物类别。
  7. 根据权利要求6所述的装置，其特征在于，所述N个预设Gabor滤波器为平行于所述高光谱遥感图像的光谱维度方向的Gabor滤波器。
  8. 根据权利要求7所述的装置，其特征在于，所述确定模块包括：
    获取单元，用于基于所述N个三维Gabor幅值特征中的任一Gabor幅值特征Mi获取所述遥感图像属于P个类别中任一类别p的置信度
    Figure PCTCN2017083766-appb-100012
    以及基于所述N个三维Gabor相位特征中的任一Gabor相位特征Fi获取所述遥感图像与所述P个类别中的任一类别p的相似性度量距离
    Figure PCTCN2017083766-appb-100013
    P为正整数；
    确定单元，用于确定所述p个类别中所述置信度
    Figure PCTCN2017083766-appb-100014
    与所述相似性度量距离
    Figure PCTCN2017083766-appb-100015
    在所述N个Gabor滤波器下的平方差之和为最大的类别为所述高光谱遥感图像中的地物类别。
  9. 根据权利要求8所述的装置，其特征在于，所述获取单元具体用于：
    利用以下公式基于所述N个三维Gabor幅值特征中的任一三维Gabor幅值特征Mi获取所述高光谱遥感图像属于P个类别中任一类别p的置信度
    Figure PCTCN2017083766-appb-100016
    Figure PCTCN2017083766-appb-100017
    其中，所述D为基于支持向量机获取的所述遥感图像的决策矩阵，所述np为所述决策矩阵D中第p行非零元素的个数。
  10. 根据权利要求8所述的装置，其特征在于，若所述相似性度量距离包括汉明距离，所述获取单元具体用于：
    利用以下公式获取所述高光谱遥感图像t和训练集合A中任意训练样本s之间的相似性度量：
    Figure PCTCN2017083766-appb-100018
    其中，B为原始高光谱数据的光谱维度；
    获取所述高光谱遥感图像t与所述任一类别p之间的汉明距离
    Figure PCTCN2017083766-appb-100019
    为：
    Figure PCTCN2017083766-appb-100020
PCT/CN2017/083766 2017-04-21 2017-05-10 一种高光谱遥感图像分类方法及装置 Ceased WO2018192023A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201710266000.XA CN107256407B (zh) 2017-04-21 2017-04-21 一种高光谱遥感图像分类方法及装置
CN201710266000.X 2017-04-21

Publications (1)

Publication Number Publication Date
WO2018192023A1 true WO2018192023A1 (zh) 2018-10-25

Family

ID=60027831

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/083766 Ceased WO2018192023A1 (zh) 2017-04-21 2017-05-10 一种高光谱遥感图像分类方法及装置

Country Status (2)

Country Link
CN (1) CN107256407B (zh)
WO (1) WO2018192023A1 (zh)

Cited By (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111079807A (zh) * 2019-12-05 2020-04-28 二十一世纪空间技术应用股份有限公司 一种地物分类方法及装置
CN111680549A (zh) * 2020-04-28 2020-09-18 肯维捷斯（武汉）科技有限公司 一种纸纹识别方法
CN112348097A (zh) * 2020-11-12 2021-02-09 上海海洋大学 一种高光谱图像分类方法
CN113344871A (zh) * 2021-05-27 2021-09-03 中国农业大学 农业遥感图像分析方法及系统
US12169974B2 (en) 2020-07-14 2024-12-17 Flir Unmanned Aerial Systems Ulc Efficient refinement neural network for real-time generic object-detection systems and methods

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109002775A (zh) * 2018-06-29 2018-12-14 华南理工大学 面向高光谱图像分类的判别性Gabor滤波方法
CN108593569B (zh) * 2018-07-02 2019-03-22 中国地质环境监测院 基于光谱形态特征的高光谱水质参数定量反演方法
CN110175638B (zh) * 2019-05-13 2021-04-30 北京中科锐景科技有限公司 一种扬尘源监测方法

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102156872A (zh) * 2010-12-29 2011-08-17 深圳大学 一种基于多光谱数据的物体识别方法和装置
CN106022391A (zh) * 2016-05-31 2016-10-12 哈尔滨工业大学深圳研究生院 一种高光谱图像特征的并行提取与分类方法

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101089874B (zh) * 2006-06-12 2010-08-18 华为技术有限公司 一种远程人脸图像的身份识别方法
CN104834909B (zh) * 2015-05-07 2018-09-21 长安大学 一种基于Gabor综合特征的图像特征描述方法
US10839510B2 (en) * 2015-08-19 2020-11-17 Colorado Seminary, Which Owns And Operates The University Of Denver Methods and systems for human tissue analysis using shearlet transforms
CN106073767B (zh) * 2016-05-26 2018-09-21 东南大学 Eeg信号的相位同步度量、耦合特征提取及信号识别方法
CN105913053B (zh) * 2016-06-07 2019-03-08 合肥工业大学 一种基于稀疏融合的单演多特征的人脸表情识别方法

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102156872A (zh) * 2010-12-29 2011-08-17 深圳大学 一种基于多光谱数据的物体识别方法和装置
CN106022391A (zh) * 2016-05-31 2016-10-12 哈尔滨工业大学深圳研究生院 一种高光谱图像特征的并行提取与分类方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
JIA, SEN ET AL.: "A 3-D Gabor Phase-Based Coding and Matching Framework for Hyperspectral Imagery Classification", IEEE TRANSACTIONS ON CYBERNETICS, 28 March 2017 (2017-03-28), XP011678866 *
WANG, FANG ET AL.: "ISAR Image Recognition with Fusion of Gabor magnitude and Phase Feature", JOURNAL OF ELECTRONICS & INFORMATION TECHNOLOGY, vol. 35, no. 8, 31 August 2013 (2013-08-31) *

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111079807A (zh) * 2019-12-05 2020-04-28 二十一世纪空间技术应用股份有限公司 一种地物分类方法及装置
CN111079807B (zh) * 2019-12-05 2023-07-07 二十一世纪空间技术应用股份有限公司 一种地物分类方法及装置
CN111680549A (zh) * 2020-04-28 2020-09-18 肯维捷斯（武汉）科技有限公司 一种纸纹识别方法
CN111680549B (zh) * 2020-04-28 2023-12-05 肯维捷斯（武汉）科技有限公司 一种纸纹识别方法
US12169974B2 (en) 2020-07-14 2024-12-17 Flir Unmanned Aerial Systems Ulc Efficient refinement neural network for real-time generic object-detection systems and methods
CN112348097A (zh) * 2020-11-12 2021-02-09 上海海洋大学 一种高光谱图像分类方法
CN113344871A (zh) * 2021-05-27 2021-09-03 中国农业大学 农业遥感图像分析方法及系统

Also Published As

Publication number Publication date
CN107256407B (zh) 2020-11-10
CN107256407A (zh) 2017-10-17

Similar Documents

Publication Publication Date Title
WO2018192023A1 (zh) 一种高光谱遥感图像分类方法及装置
JP5749394B2 (ja) 視覚探索のための堅牢な特徴マッチング
Cheng et al. Robust affine invariant feature extraction for image matching
EP2549434B1 (fr) Procédé de modélisation de bâtiments à partir d'une image géoréférencée
Campo et al. Multimodal stereo vision system: 3D data extraction and algorithm evaluation
US9665606B2 (en) Edge-based recognition, systems and methods
CN107633216B (zh) 高光谱遥感图像的三维表面空谱联合特征编码方法及装置
US20150206319A1 (en) Digital image edge detection
CN117173437A (zh) 多维定向自相似特征的多模态遥感影像混合匹配方法和系统
CN115984178B (zh) 伪造图像检测方法、电子设备和计算机可读存储介质
Quan et al. Efficient and robust: A cross-modal registration deep wavelet learning method for remote sensing images
Pang et al. Exploiting local linear geometric structure for identifying correct matches
CN112288813A (zh) 基于多目视觉测量与激光点云地图匹配的位姿估计方法
CN104680190B (zh) 目标检测方法及装置
Shahi et al. DC4Flood: A deep clustering framework for rapid flood detection using Sentinel-1 SAR imagery
Cui et al. Global propagation of affine invariant features for robust matching
CN120088516B (zh) 一种基于亚像素偏差估计的多模态图像匹配方法与系统
Lejbølle et al. Enhancing person re‐identification by late fusion of low‐, mid‐and high‐level features
Li A novel method for multi-angle SAR image matching
CN116543249B (zh) 模型训练方法、质量评估方法、装置以及电子设备
CN102156872A (zh) 一种基于多光谱数据的物体识别方法和装置
CN118552759A (zh) 一种基于机器学习的sar图像智能分类方法和系统
CN114782784B (zh) 车辆识别方法及相关装置、电子设备和存储介质
CN110222756A (zh) 一种面向高光谱复杂背景的迭代聚类异常检测算法
CN118735975B (zh) 光学图像和合成孔径雷达图像的配准方法、装置、电子设备、存储介质和计算机程序产品

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17906560

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 20/01/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 17906560

Country of ref document: EP

Kind code of ref document: A1