CN112149747A - Hyperspectral image classification method based on improved Ghost3D module and covariance pooling - Google Patents
Hyperspectral image classification method based on improved Ghost3D module and covariance pooling Download PDFInfo
- Publication number
- CN112149747A CN112149747A CN202011036420.7A CN202011036420A CN112149747A CN 112149747 A CN112149747 A CN 112149747A CN 202011036420 A CN202011036420 A CN 202011036420A CN 112149747 A CN112149747 A CN 112149747A
- Authority
- CN
- China
- Prior art keywords
- matrix
- module
- ghost3d
- covariance
- acquiring
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
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 Analysis (AREA)
Abstract
The invention discloses a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling, which comprises the steps of obtaining a spectral band in a hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point; acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D module stacks; acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector; and sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image. The method combines the advantages of a lightweight network formed by stacking Ghost3D modules and a covariance pooling method, and realizes accurate classification of hyperspectral images under low training samples.
Description
Technical Field
The application belongs to the technical field of image processing, and particularly relates to a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling.
Background
With the development of the spectral imager, the hyperspectral remote sensing technology has become a research hotspot in the field of remote sensing. As an important means of earth observation, the method is widely applied to the aspects of fine agriculture, environmental monitoring, military detection and the like. The hyperspectral image is an image with high dimensionality and contains rich spectral characteristic information and spatial characteristic information, so that the hyperspectral image is greatly different from a common image. By integrating the spectral band information of the reflected target radiation with the image information reflecting the target two-dimensional space, unified ground object information can be obtained. When the hyperspectral images are classified and the like, a severe Hughes phenomenon can be generated along with the increase of data dimensionality, namely, the classification precision is obviously reduced. Therefore, how to efficiently extract the spectrum and spatial information of the hyperspectral image becomes an urgent problem to be solved.
In recent years, deep learning is rapidly developed due to the intrinsic structure of a visual system, and a plurality of scholars apply the deep learning method to the classification research of hyperspectral images. The convolutional neural network is used as an important algorithm in deep learning, has strong characteristic learning capability, has structural characteristics very suitable for solving the problem of the image field, and has excellent performance in the fields of target detection and tracking, image classification and the like. In a common convolutional neural network, because the number of network layers is continuously deepened, gradients are continuously attenuated in the error back propagation process, so that information is lost, and errors are continuously increased. Therefore, how to construct an architecture with strong robustness and depth is crucial for the hyperspectral image classification task.
Disclosure of Invention
The application provides a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling, and aims to solve the problems that in the prior art, due to the fact that the number of network layers is continuously deepened, gradient can be continuously attenuated in the error back propagation process, information is lost, errors are continuously increased, and accordingly the spectrum and spatial information of a hyperspectral image cannot be efficiently extracted.
In order to achieve the purpose, the technical scheme adopted by the application is as follows:
a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling comprises the following steps:
obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D module stacks;
acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
and sending the first feature vector to a full-connection network to obtain classification data of the hyperspectral remote sensing image.
A lightweight network is constructed through a Ghost3D module, and second-order information is extracted by using a covariance pooling method, so that the hyperspectral images are accurately classified under a low training sample.
Preferably, the obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, the spectral band including cubic data with a fixed spatial size from each pixel point, includes:
acquiring input data X and an objective function, wherein the objective function is as follows: minP tr(PT BP),
Wherein, I is an identity matrix with a size of L multiplied by L, tr and T respectively represent the trace and transposition operations of the matrix, B is a covariance matrix, and P is a projection matrix;
and acquiring the spectral band of the hyperspectral remote sensing image based on the target function.
Preferably, the obtaining a third feature map according to a lightweight network and based on the cube data, where the lightweight network includes a plurality of Ghost3D module stacks, including:
acquiring a first feature map of the Ghost3D module;
acquiring a second characteristic diagram based on the three-dimensional convolution model;
and splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain the third characteristic diagram.
Preferably, the obtaining second-order information in the third feature map based on the covariance pooling method, where the second-order information is denoted as a first feature vector, includes:
acquiring a matrix Z;
based on the matrix Z, a covariance matrix C is calculated as follows:
Mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
obtaining the first eigenvector consisting of the upper triangular data of the matrix F.
Preferably, the three-layer fully-connected network comprises three fully-connected layers, wherein the first two of the three fully-connected layers use a Dropout method in a normalized form, and the last fully-connected layer uses Softmax to output a prediction classification result.
A hyperspectral image classification device based on an improved Ghost3D module and covariance pooling, comprising:
an acquisition module: obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
a processing module: acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D stacks;
a pooling module: acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
a sending module: and sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image.
Preferably, the acquiring module includes:
a first acquisition unit: acquiring input data X and an objective function, wherein the objective function is as follows: minP tr(PTBP),
Wherein, I is an identity matrix with a size of L multiplied by L, tr and T respectively represent the trace and transposition operations of the matrix, B is a covariance matrix, and P is a projection matrix;
a second acquisition unit: and acquiring the spectral band of the hyperspectral remote sensing image based on the target function.
Preferably, the processing module includes:
a third acquisition unit: acquiring a first feature map of the Ghost3D module;
a fourth acquisition unit: acquiring a second characteristic diagram based on the three-dimensional convolution model;
a fifth acquisition unit: and splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain the third characteristic diagram.
Preferably, the pooling module comprises:
a sixth acquisition unit: acquiring a matrix Z;
a pooling subunit: based on the matrix Z, a covariance matrix C is calculated as follows:
wherein z isi∈[z1,z2,...,zN]，ziThe ith column vector along the second dimension for matrix Z,and is
A mapping unit: mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
a seventh acquisition unit: obtaining the first eigenvector consisting of the upper triangular data of the matrix F.
The invention has the following beneficial effects:
according to the hyperspectral image classification method based on the improved Ghost3D module and the covariance pooling, the spectrum dimensionality reduction is carried out on a hyperspectral image by utilizing principal component analysis, the Ghost3D module is formed by improving the nonlinear mapping part in the Ghost module into three-dimensional convolution, the Ghost3D module is stacked to form a lightweight network, second-order information is extracted by utilizing the covariance pooling method, more distinguishing features are obtained by fully utilizing spectrum and space feature information, and the accuracy of hyperspectral image classification is remarkably improved.
Drawings
FIG. 1 is a first flowchart of a method for classifying hyperspectral images based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 2 is a second flowchart of a method for classifying hyperspectral images based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 3 is a third flowchart of a method for classifying hyperspectral images based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 4 is a fourth flowchart of a method for classifying hyperspectral images based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
fig. 5 is a flowchart of an embodiment of the present invention for implementing a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling.
FIG. 6 is a schematic diagram of a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 7 is a schematic diagram of an acquisition module of a hyperspectral image classification device based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 8 is a schematic diagram of a processing module of a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 9 is a schematic diagram of an output module of a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling according to an embodiment of the invention;
FIG. 10 is a flowchart of an embodiment of the present invention for implementing a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling;
fig. 11 is a schematic diagram of an electronic device implementing a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling according to an embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application, and it is obvious that the described embodiments are only a part of the embodiments of the present application, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present application.
The terms "first," "second," and the like in the claims and in the description of the application are used for distinguishing between similar elements and not necessarily for describing a particular sequential or chronological order. It is to be understood that the terms so used are interchangeable under appropriate circumstances and are merely descriptive of the various embodiments of the application and how objects of the same nature can be distinguished. Furthermore, the terms "comprises," "comprising," and "having," and any variations thereof, are intended to cover a non-exclusive inclusion, such that a process, method, system, article, or apparatus that comprises a list of elements is not necessarily limited to those elements, but may include other elements not expressly listed or inherent to such process, method, article, or apparatus.
Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description of the present application herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.
Example 1
As shown in fig. 1, a hyperspectral image classification method based on improved Ghost3D module and covariance pooling includes:
s110, obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
s120, acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D module stacks;
s130, acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
s140, sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image.
According to the embodiment 1, the lightweight network is constructed through the Ghost3D module, and the second-order information is extracted by using the covariance pooling method, so that the hyperspectral images are accurately classified under the condition of low training samples.
Example 2
As shown in fig. 2, a hyperspectral image classification method based on improved Ghost3D module and covariance pooling includes:
s210, acquiring input data X and an objective function, wherein the objective function is as follows: minP tr(PT BP),
Wherein, I is an identity matrix with a size of L multiplied by L, tr and T respectively represent the trace and transposition operations of the matrix, B is a covariance matrix, and P is a projection matrix;
s220, acquiring a spectral band of the hyperspectral remote sensing image based on the target function;
s230, acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D module stacks;
s240, acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
and S250, sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image.
As can be seen from embodiment 2, the high-dimensional data includes too much redundant information, which makes the calculation amount very large, and therefore, the principal component analysis is used to perform spectral dimensionality reduction on the hyperspectral image, thereby reducing the dimensionality of the spectral band and retaining complete spatial information.
Example 3
As shown in fig. 3, a hyperspectral image classification method based on improved Ghost3D module and covariance pooling includes:
s310, acquiring a first characteristic diagram of the Ghost3D module;
s320, acquiring a second characteristic diagram based on the three-dimensional convolution model;
s330, splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain a third characteristic diagram.
As can be seen from embodiment 3, cube data of a fixed spatial size is taken for each pixel point and input to the lightweight network, and the generalization error is reduced by reducing the dimensionality of the data.
Example 4
As shown in fig. 4, a hyperspectral image classification method based on improved Ghost3D module and covariance pooling includes:
s410, acquiring a matrix Z;
s420, based on the matrix Z, calculating a covariance matrix C as follows:
S430, mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
s440, obtaining the first feature vector consisting of the upper triangular data of the matrix F.
As can be seen from the embodiment 4, the second-order statistic information is obtained by calculating the covariance matrix among the feature maps, so that the dimension of the third-order information is reduced to the second-order information, the calculation error is reduced, and the calculation process is more controllable.
Example 5
As shown in fig. 5, one specific embodiment may be:
s510, acquiring a hyperspectral image to be classified;
in order to improve the data processing speed and accuracy, the acquired hyperspectral image needs to be preprocessed first.
S520, performing spectrum dimensionality reduction on the hyperspectral image by using a principal component analysis method;
the size of the obtained hyperspectral remote sensing image is W multiplied by H multiplied by L;
reshaping the hyperspectral remote sensing image to input data X with the size of L multiplied by N, wherein N is W multiplied by H;
calculating a covariance matrix B of the input data X;
acquiring k eigenvectors corresponding to the largest k eigenvalues in the input data X, and taking the k eigenvectors as column vectors to obtain a projection matrix P;
the optimization objective function for principal component analysis is established as follows:
wherein, I is a unit matrix with L multiplied by L size, tr and T respectively represent the trace of the matrix and the transposition operation;
calculating an optimization objective function, and outputting a dimensionality-reduced hyperspectral remote sensing image Y with the size of kxL by principal component analysis when P with the size of Lxk is obtained through calculation, wherein Y is PTX, the spectral band size of the hyperspectral remote sensing image is reduced from the original L to k.
S530, taking cubic data with a fixed space size (such as 25 multiplied by 25) for each pixel point in the spectral band, forming a Ghost3D module by improving nonlinear mapping in the Ghost module into three-dimensional convolution, and extracting spectrum-space characteristics by using a lightweight network stacked by the Ghost3D module to obtain a final characteristic diagram;
acquiring a first characteristic diagram, wherein the size of the first characteristic diagram is H multiplied by W multiplied by K;
and acquiring a second feature map with the size of H multiplied by W multiplied by M through three-dimensional convolution, and splicing the second feature map with the first feature map along the channel direction to form a third feature map with the size of H multiplied by W multiplied by (M + K).
S540, extracting second-order information from the feature map by using a covariance pooling method, and outputting a feature vector;
let the size of the fourth feature map be H × W × D;
reshaping the fourth feature map as a matrix Z of size D × N, and N ═ H × W;
the covariance matrix C is calculated as follows:
wherein the content of the first and second substances,zi∈[z1,z2,...,zN]，zian ith column vector along a second dimension for the matrix Z, an
Mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
and taking the upper triangular data of the matrix F to form a feature vector F for output.
And S550, inputting the feature vectors into the three-layer full-connection network to obtain a prediction classification result aiming at the hyperspectral image to be classified.
The three-layer full-connection network comprises three full-connection layers, wherein the first two full-connection layers in the three full-connection layers use a Dropout method in a normalized form, and the last full-connection layer uses Softmax to output a prediction classification result.
Example 6
As shown in fig. 6, a hyperspectral image classification apparatus based on improved Ghost3D module and covariance pooling includes:
the acquisition module 10: obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
the processing module 20: acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D stacks;
the pooling module 30: acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
the sending module 40: and sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image.
One embodiment of the above apparatus is: the method comprises the steps that an acquisition module 10 acquires a spectral band in a hyperspectral remote sensing image, a processing module 20 acquires a third feature map according to a lightweight network and based on cube data, a pooling module 30 acquires second-order information in the third feature map based on a covariance pooling method, a sending module 40 sends a first feature vector to a three-layer full-connection network to acquire classification data of the hyperspectral remote sensing image
Example 7
As shown in fig. 7, an acquisition module 10 of a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling comprises:
the first acquisition unit 12: acquiring input data X and an objective function, wherein the objective function is as follows: minP tr(PT BP),
Wherein, I is an identity matrix with a size of L multiplied by L, tr and T respectively represent the trace and transposition operations of the matrix, B is a covariance matrix, and P is a projection matrix;
the second acquisition unit 14: and acquiring the spectral band of the hyperspectral remote sensing image based on the target function.
One specific embodiment of the acquisition module 10 of the above apparatus may be: the first obtaining unit 12 obtains input data X and an objective function, and the second obtaining unit 14 obtains a spectral band of the hyperspectral remote sensing image based on the objective function.
Example 8
As shown in fig. 8, a processing module 20 of the hyperspectral image classification apparatus based on the improved Ghost3D module and covariance pooling comprises:
the third acquisition unit 22: acquiring a first feature map of the Ghost3D module;
the fourth acquiring unit 24: acquiring a second characteristic diagram based on the three-dimensional convolution model;
the fifth acquiring unit 26: and splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain the third characteristic diagram.
One embodiment of the processing module 20 of the above apparatus may be: the third obtaining unit 22 obtains a first feature map of the Ghost3D module, the fourth obtaining unit 24 obtains a second feature map based on a three-dimensional convolution model, and the fifth obtaining unit 26 splices the first feature map and the second feature map along a channel direction to obtain the third feature map.
Example 9
As shown in fig. 9, a pooling module 30 of a hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling comprises:
the sixth acquiring unit 32: acquiring a matrix Z;
pooling sub-unit 34: based on the matrix Z, a covariance matrix C is calculated as follows:
The mapping unit 36: mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
the seventh acquiring unit 38: obtaining the first eigenvector consisting of the upper triangular data of the matrix F.
One embodiment of the pooling module 30 of the above apparatus may be: the sixth obtaining unit 32 obtains a matrix Z, the pooling sub-unit 34 obtains a covariance matrix C based on the matrix Z, the mapping unit 36 maps the covariance matrix C to the euclidean space to obtain a matrix F, and the seventh obtaining unit 38 obtains the first eigenvector formed by upper triangular data of the matrix F.
Example 10
As shown in fig. 10, one specific implementation is as follows:
s1010, acquiring a hyperspectral image to be classified;
in order to improve the data processing speed and accuracy, the acquired hyperspectral image needs to be preprocessed first.
S1020, performing spectrum dimensionality reduction on the hyperspectral image by using a principal component analysis method;
the size of the obtained hyperspectral remote sensing image is W multiplied by H multiplied by L;
reshaping the hyperspectral remote sensing image to input data X with the size of L multiplied by N, wherein N is W multiplied by H;
calculating a covariance matrix B of the input data X;
acquiring k eigenvectors corresponding to the largest k eigenvalues in the input data X, and taking the k eigenvectors as column vectors to obtain a projection matrix P;
the optimization objective function for principal component analysis is established as follows:
wherein, I is a unit matrix with L multiplied by L size, tr and T respectively represent the trace of the matrix and the transposition operation;
calculating an optimization objective function, and outputting a dimensionality-reduced hyperspectral remote sensing image Y with the size of kxL by principal component analysis when P with the size of Lxk is obtained through calculation, wherein Y is PTX, the spectral band size of the hyperspectral remote sensing image is reduced from the original L to k.
S1030, taking cubic data with a fixed space size (such as 25 multiplied by 25) for each pixel point in the spectral band, forming a Ghost3D module by improving nonlinear mapping in the Ghost module into three-dimensional convolution, and extracting spectrum-space characteristics by using a lightweight network stacked by the Ghost3D module to obtain a final characteristic diagram;
acquiring a first characteristic diagram, wherein the size of the first characteristic diagram is H multiplied by W multiplied by K;
and acquiring a second feature map with the size of H multiplied by W multiplied by M through three-dimensional convolution, and splicing the second feature map with the first feature map along the channel direction to form a third feature map with the size of H multiplied by W multiplied by (M + K).
S1040, extracting second-order information from the feature map by using a covariance pooling method, and outputting feature vectors, wherein the method comprises the following steps:
let the size of the fourth feature map be H × W × D;
reshaping the fourth feature map as a matrix Z of size D × N, and N ═ H × W;
the covariance matrix C is calculated as follows:
wherein z isi∈[z1,z2,...,zN]，ziAn ith column vector along a second dimension for the matrix Z, an
Mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
and taking the upper triangular data of the matrix F to form a feature vector F for output.
And S1050, inputting the feature vectors into a three-layer full-connection network to obtain a prediction classification result aiming at the hyperspectral image to be classified.
The three-layer full-connection network comprises three full-connection layers, wherein the first two full-connection layers in the three full-connection layers use a Dropout method in a normalized form, and the last full-connection layer uses Softmax to output a prediction classification result.
Example 11
As shown in fig. 11, an electronic device comprises a memory 1101 and a processor 1102, the memory 1101 storing one or more computer instructions, wherein the one or more computer instructions are executed by the processor 1102 to implement a hyperspectral image classification method based on a modified Ghost3D module and covariance pooling as described above.
It can be clearly understood by those skilled in the art that, for convenience and brevity of description, the specific working process of the electronic device described above may refer to the corresponding process in the foregoing method embodiment, and is not described herein again.
A computer-readable storage medium storing a computer program which, when executed by a computer, causes the computer to implement a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling as described above.
Illustratively, the computer program may be divided into one or more modules/units, which are stored in the memory 1101 and executed by the processor 1102 to implement the present invention. One or more modules/units may be a series of computer program instruction segments capable of performing certain functions, the instruction segments being used to describe the execution of a computer program in a computer device.
The computer device may be a desktop computer, a notebook, a palm computer, a cloud server, or other computing devices. The computer device may include, but is not limited to, a memory 1101, a processor 1102. Those skilled in the art will appreciate that the present embodiments are merely exemplary of a computing device and are not intended to limit the computing device, and may include more or fewer components, or some of the components may be combined, or different components, e.g., the computing device may also include input output devices, network access devices, buses, etc.
The processor 1102 may be a Central Processing Unit (CPU), other general purpose processor 1102, a digital signal processor 1102 (DSP), an Application Specific Integrated Circuit (ASIC), an off-the-shelf programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, etc. The general purpose processor 1102 may be a microprocessor 1102 or the processor 1102 may be any conventional processor 1102 or the like.
The storage 1101 may be an internal storage unit of the computer device, such as a hard disk or a memory of the computer device. The memory 1101 may also be an external storage device of the computer device, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash card (FlashCard), etc. provided on the computer device. Further, the memory 1101 may also include both an internal storage unit and an external storage device of the computer device. The memory 1101 is used to store computer programs and other programs and data required by the computer device. The memory 1101 may also be used to temporarily store data that has been output or is to be output.
The above description is only an embodiment of the present invention, but the technical features of the present invention are not limited thereto, and any changes or modifications within the technical field of the present invention by those skilled in the art are covered by the claims of the present invention.
Claims (10)
1. A hyperspectral image classification method based on an improved Ghost3D module and covariance pooling is characterized by comprising the following steps:
obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D module stacks;
acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
and sending the first feature vector to a full-connection network to obtain classification data of the hyperspectral remote sensing image.
2. The hyperspectral image classification method based on the improved Ghost3D module and covariance pooling of claim 1, wherein a spectral band in the hyperspectral remote sensing image is obtained according to a principal component analysis method, the spectral band comprises cubic data with a fixed spatial size in each pixel point, and the method comprises:
acquiring input data X and an objective function, wherein the objective function is as follows:
and acquiring a spectral band in the hyperspectral remote sensing image based on the target function.
3. The hyperspectral image classification method based on improved Ghost3D modules and covariance pooling of claim 1, wherein a third feature map is obtained according to the cube data and a lightweight network, wherein the lightweight network comprises a plurality of stacks of Ghost3D modules, and comprises:
acquiring a first feature map of the Ghost3D module;
acquiring a second characteristic diagram based on the three-dimensional convolution model;
and splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain the third characteristic diagram.
4. The hyperspectral image classification method based on the improved Ghost3D module and covariance pooling of claim 1, wherein the method for covariance pooling obtains second-order information in the third feature map, and the second-order information is marked as a first feature vector, and comprises:
acquiring a matrix Z;
based on the matrix Z, a covariance matrix C is calculated as follows:
Mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
obtaining the first eigenvector consisting of the upper triangular data of the matrix F.
5. A hyperspectral image classification device based on an improved Ghost3D module and covariance pooling, comprising:
an acquisition module: obtaining a spectral band in the hyperspectral remote sensing image according to a principal component analysis method, wherein the spectral band comprises cubic data with a fixed space size in each pixel point;
a processing module: acquiring a third feature map according to a lightweight network and based on the cube data, wherein the lightweight network comprises a plurality of Ghost3D stacks;
a pooling module: acquiring second-order information in the third feature map based on a covariance pooling method, wherein the second-order information is marked as a first feature vector;
a sending module: and sending the first feature vector to a three-layer full-connection network to obtain classification data of the hyperspectral remote sensing image.
6. The apparatus for hyperspectral image classification based on an improved Ghost3D module and covariance pooling of claim 5, wherein the obtaining module comprises:
a first acquisition unit: acquiring input data X and an objective function, wherein the objective function is as follows:
a second acquisition unit: and acquiring the spectral band of the hyperspectral remote sensing image based on the target function.
7. The apparatus for hyperspectral image classification based on an improved Ghost3D module and covariance pooling of claim 5, wherein the processing module comprises:
a third acquisition unit: acquiring a first feature map of the Ghost3D module;
a fourth acquisition unit: acquiring a second characteristic diagram based on the three-dimensional convolution model;
a fifth acquisition unit: and splicing the first characteristic diagram and the second characteristic diagram along the channel direction to obtain the third characteristic diagram.
8. The hyperspectral image classification apparatus based on an improved Ghost3D module and covariance pooling of claim 5, wherein the pooling module comprises:
a sixth acquisition unit: acquiring a matrix Z;
a pooling subunit: based on the matrix Z, a covariance matrix C is calculated as follows:
A mapping unit: mapping the covariance matrix C to an Euclidean space to obtain a matrix F as follows:
wherein, C is UAUTA and U respectively represent the eigenvalue and the eigenvector of the covariance matrix C, and the obtained matrix F is a symmetric matrix;
a seventh acquisition unit: obtaining the first eigenvector consisting of the upper triangular data of the matrix F.
9. An electronic device comprising a memory and a processor, the memory configured to store one or more computer instructions, wherein the one or more computer instructions are executable by the processor to implement a hyperspectral image classification method based on an improved Ghost3D module and covariance pooling according to any of claims 1-4.
10. A computer-readable storage medium storing a computer program, the computer program causing a computer to perform a method for hyperspectral image classification based on an improved Ghost3D module and covariance pooling according to any of claims 1-4.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011036420.7A CN112149747A (en) | 2020-09-27 | 2020-09-27 | Hyperspectral image classification method based on improved Ghost3D module and covariance pooling |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202011036420.7A CN112149747A (en) | 2020-09-27 | 2020-09-27 | Hyperspectral image classification method based on improved Ghost3D module and covariance pooling |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112149747A true CN112149747A (en) | 2020-12-29 |
Family
ID=73894482
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202011036420.7A Pending CN112149747A (en) | 2020-09-27 | 2020-09-27 | Hyperspectral image classification method based on improved Ghost3D module and covariance pooling |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112149747A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112819910A (en) * | 2021-01-08 | 2021-05-18 | 上海理工大学 | Hyperspectral image reconstruction method based on double-ghost attention machine mechanism network |
| CN113052254A (en) * | 2021-04-06 | 2021-06-29 | 安徽理工大学 | Multi-attention ghost residual fusion classification model and classification method thereof |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170299435A1 (en) * | 2012-05-21 | 2017-10-19 | Digimarc Corporation | Sensor-synchronized spectrally-structured-light imaging |
| CN111310598A (en) * | 2020-01-20 | 2020-06-19 | 浙江工业大学 | Hyperspectral remote sensing image classification method based on 3-dimensional and 2-dimensional mixed convolution |
| CN111488989A (en) * | 2020-04-16 | 2020-08-04 | 济南浪潮高新科技投资发展有限公司 | Method and model for realizing lightweight target detection at mobile phone end |
-
2020
- 2020-09-27 CN CN202011036420.7A patent/CN112149747A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170299435A1 (en) * | 2012-05-21 | 2017-10-19 | Digimarc Corporation | Sensor-synchronized spectrally-structured-light imaging |
| CN111310598A (en) * | 2020-01-20 | 2020-06-19 | 浙江工业大学 | Hyperspectral remote sensing image classification method based on 3-dimensional and 2-dimensional mixed convolution |
| CN111488989A (en) * | 2020-04-16 | 2020-08-04 | 济南浪潮高新科技投资发展有限公司 | Method and model for realizing lightweight target detection at mobile phone end |
Non-Patent Citations (1)
| Title |
|---|
| KAI HAN等: "GhostNet: More Features From Cheap Operations", 《2020 IEEE/CVF CONFERENCE ON COMPUTER VISION AND PATTERN RECOGNITION (CVPR)》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112819910A (en) * | 2021-01-08 | 2021-05-18 | 上海理工大学 | Hyperspectral image reconstruction method based on double-ghost attention machine mechanism network |
| CN113052254A (en) * | 2021-04-06 | 2021-06-29 | 安徽理工大学 | Multi-attention ghost residual fusion classification model and classification method thereof |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Zheng et al. | Hyperspectral image classification using mixed convolutions and covariance pooling | |
| Wu et al. | ORSIm detector: A novel object detection framework in optical remote sensing imagery using spatial-frequency channel features | |
| Gao et al. | Change detection from synthetic aperture radar images based on channel weighting-based deep cascade network | |
| EP3385919B1 (en) | Method of processing passage record and device | |
| CN111460968B (en) | Unmanned aerial vehicle identification and tracking method and device based on video | |
| WO2021022521A1 (en) | Method for processing data, and method and device for training neural network model | |
| Akey Sungheetha | Classification of remote sensing image scenes using double feature extraction hybrid deep learning approach | |
| US11132392B2 (en) | Image retrieval method, image retrieval apparatus, image retrieval device and medium | |
| Zhu et al. | Learning compact visual representation with canonical views for robust mobile landmark search | |
| CN103700099A (en) | Rotation and dimension unchanged wide baseline stereo matching method | |
| CN112580480B (en) | Hyperspectral remote sensing image classification method and device | |
| CN113486851A (en) | Hyperspectral image classification method based on double-branch spectrum multi-scale attention network | |
| CN105320764A (en) | 3D model retrieval method and 3D model retrieval apparatus based on slow increment features | |
| CN110765882A (en) | Video tag determination method, device, server and storage medium | |
| CN112529068B (en) | Multi-view image classification method, system, computer equipment and storage medium | |
| Qayyum et al. | Designing deep CNN models based on sparse coding for aerial imagery: a deep-features reduction approach | |
| CN112149747A (en) | Hyperspectral image classification method based on improved Ghost3D module and covariance pooling | |
| WO2023221790A1 (en) | Image encoder training method and apparatus, device, and medium | |
| Zhang et al. | PSNet: Perspective-sensitive convolutional network for object detection | |
| Sjahputera et al. | Clustering of detected changes in high-resolution satellite imagery using a stabilized competitive agglomeration algorithm | |
| CN111310732A (en) | High-precision face authentication method, system, computer equipment and storage medium | |
| WO2022126529A1 (en) | Positioning method and device, and unmanned aerial vehicle and storage medium | |
| El Ouariachi et al. | RGB-D feature extraction method for hand gesture recognition based on a new fast and accurate multi-channel cartesian Jacobi moment invariants | |
| Jiang et al. | IARet: A lightweight multiscale infrared aerocraft recognition algorithm | |
| CN112733807A (en) | Face comparison graph convolution neural network training method and device |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20201229 |