WO2020042070A1 - 提升高光谱图像分类精度的方法、装置、设备及存储介质 - Google Patents

提升高光谱图像分类精度的方法、装置、设备及存储介质 Download PDF

Info

Publication number
WO2020042070A1
WO2020042070A1 PCT/CN2018/103204 CN2018103204W WO2020042070A1 WO 2020042070 A1 WO2020042070 A1 WO 2020042070A1 CN 2018103204 W CN2018103204 W CN 2018103204W WO 2020042070 A1 WO2020042070 A1 WO 2020042070A1
Authority
WO
WIPO (PCT)
Prior art keywords
tensor
order
domain
hyperspectral image
slice
Prior art date
Application number
PCT/CN2018/103204
Other languages
English (en)
French (fr)
Inventor
李岩山
范雷东
唐浩劲
谢维信
Original Assignee
深圳大学
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 深圳大学 filed Critical 深圳大学
Priority to PCT/CN2018/103204 priority Critical patent/WO2020042070A1/zh
Publication of WO2020042070A1 publication Critical patent/WO2020042070A1/zh

Links

Images

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06FELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00Pattern recognition

Definitions

  • the invention relates to the technical field of image processing, and in particular, to a method, a device, a device, and a storage medium for improving classification accuracy of hyperspectral images.
  • the hyperspectral image contains not only the spatial structure information that reflects the morphological features of the ground features, but also the spectral structure information that reflects the attributes and state attributes of the ground features.
  • This "map-in-one" technology has greatly improved the human senses, and has been widely used in agricultural science, materials, earth science, food industry, biomedical applications, and military fields.
  • LBP Local Binary Pattern
  • the main purpose of the present invention is to provide a method, a device, a device and a storage medium for improving the classification accuracy of hyperspectral images. .
  • a first aspect of the present invention provides a method for improving classification accuracy of a hyperspectral image.
  • the method includes:
  • the feature statistical histograms of the respective front slices are connected to obtain the statistical histogram of the preset local area, and the statistical histogram of the preset local area is used to classify the hyperspectral image.
  • the step of traversing each positive slice in the third-order kernel tensor and using a preset encoding function to establish a feature statistical histogram of each positive slice includes:
  • the feature statistical histogram of the traversed front slice is determined by using the spatial spectrum domain joint coding value.
  • the step of quantizing the feature information of the domain to obtain a spatially coded joint coded value of the tensor element includes:
  • the feature information of the domain carrying the binomial coefficient is converted into a decimal number to obtain a spatially coded joint coded value of the tensor element.
  • the step of joining the feature statistical histograms of the respective front slices includes:
  • the feature statistical histograms of the respective front slices are connected.
  • a second aspect of the present invention provides a device for improving classification accuracy of hyperspectral images.
  • the device includes:
  • a establishing module configured to establish a third-order tensor of a preset local region of a hyperspectral image, and perform tensor decomposition on the third-order tensor to obtain a third-order kernel tensor corresponding to the preset local region;
  • a classification module is configured to join the statistical histograms of the features of the respective front slices to obtain the statistical histograms of the preset local regions, and use the statistical histograms of the preset local regions to process the hyperspectral image. sort.
  • the operation module includes:
  • a first determining module configured to determine domain feature information of a tensor element in the traversed front slice
  • a quantization module configured to quantify the feature information of the domain to obtain a spatially coded joint coding value of the tensor element
  • a second determining module is configured to determine a feature statistical histogram of the traversed frontal slice by using the spatial spectrum domain joint coding value.
  • the quantization module is specifically configured to:
  • the preset binomial coefficient is added to the domain feature information, and the domain feature information carrying the binomial coefficient is converted into a decimal number to obtain a spatially spectral domain joint coding value of the tensor element.
  • the classification module includes:
  • a linking module is configured to link the feature statistical histograms of the frontal slices according to the position order of the frontal slices in the third-order kernel tensor.
  • a third aspect of the present invention provides an electronic device including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor executes the computer program
  • a memory a processor
  • a computer program stored in the memory and executable on the processor, and the processor executes the computer program
  • a fourth aspect of the present invention provides a computer-readable storage medium, where the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the improvement provided by the first aspect of the present invention Various steps in the method of spectral image classification accuracy.
  • the invention provides a method for improving the classification accuracy of hyperspectral images.
  • a third-order tensor of a preset local area of a hyperspectral image is first established, and the third-order tensor is tensor-decomposed to obtain The third-order kernel tensor corresponding to the preset local area, and then the feature statistical histogram of each front slice in the third-order kernel tensor is created. Finally, the feature statistical histogram of each front slice is connected to obtain the statistical histogram of the preset local area. This statistical histogram can classify hyperspectral images.
  • each of the above frontal slices can reflect the joint distribution of the spectral values of different bands in the hyperspectral image and the spatial distribution information
  • the information on the spectral and spatial domains of the hyperspectral image can be simultaneously represented by the feature statistical histogram of each frontal slice. That is, using the statistical histogram to classify the hyperspectral image can effectively improve the classification accuracy.
  • FIG. 1 is a schematic diagram of a third-order tensor model of a hyperspectral image in an embodiment of the present invention
  • FIG. 2 is a schematic flowchart of steps in a method for improving classification accuracy of a hyperspectral image according to an embodiment of the present invention
  • FIG. 3 is an exploded view of a mode 3 fiber of the third-order tensor T in the embodiment of the present invention.
  • FIG. 4 is a schematic diagram of a virtual program module of an apparatus for improving classification accuracy of a hyperspectral image according to an embodiment of the present invention
  • FIG. 5 is a schematic diagram of a detailed program module of an operation module 402 according to an embodiment of the present invention.
  • FIG. 6 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention.
  • the hyperspectral image may be expressed as I (x, y, ⁇ ), where (x, y) is a spatial domain coordinate, and ⁇ is a spectral domain coordinate.
  • the hyperspectral image can be modeled as a third-order tensor, that is, I is represented by a third-order tensor T ⁇ R X ⁇ Y ⁇ ⁇ .
  • X and Y are the size of the hyperspectral image space domain, that is, the value interval of (x, y); ⁇ is the size of the hyperspectral image spectral domain (band number), that is, the value interval of ⁇ ;
  • Each pixel I (x, y, ⁇ ) corresponds to each element T xy ⁇ of the third-order tensor T.
  • FIG. 1 is a schematic diagram of a third-order tensor model of a hyperspectral image.
  • the fibers and slices of the tensor are one of the basic elements constituting the tensor.
  • the fibers and slices of the tensor can realize the extraction of vectors or matrices in high-dimensional geometry, which effectively reduces the computational complexity.
  • a tensor's fiber is defined as an index that changes at any order, and indexes at other orders are fixed.
  • the fibers of the third-order tensor T of the hyperspectral image are divided into rows (mode 1 fiber), columns (mode 2 fiber), and tube fibers (mode 3 fiber), which are denoted as T : y ⁇ , Tx : ⁇ , and Txy: .
  • This model can map the pixels with spatial coordinates (x, y) to n-dimensional vectors Representation, analysis of the hyperspectral image in combination with the 3-fiber model can effectively improve the efficiency of the use of spatial spectral domain information.
  • a slice is a two-dimensional part of a tensor, which is implemented by fixing an index of any two orders, and is usually used to perform a matrix extraction operation.
  • a slice of the third-order tensor T of a hyperspectral image is defined as a horizontal slice, a side slice, and a front slice, which are respectively represented as T x :: , T : y: and T :: ⁇ , where the front slice can effectively represent the hyperspectral image. Joint distribution of spectral values in different bands.
  • the spatial spectrum domain combined probability distribution is used to represent the characteristics of hyperspectral images.
  • a method for improving the classification accuracy of hyperspectral images is proposed.
  • FIG. 2 is a schematic flowchart of steps in a method for improving classification accuracy of hyperspectral images in an embodiment of the present invention.
  • the method for improving classification accuracy of hyperspectral images includes:
  • Step 201 Establish a third-order tensor of a preset local region of a hyperspectral image, and perform tensor decomposition on the third-order tensor to obtain a third-order kernel tensor corresponding to the preset local region.
  • Tensor decomposition based on modular n multiplication is an important method for compressing and expressing the original tensor. It can represent the tensor as a form of multiplication of the kernel tensor and the projection matrix on each order, where the kernel tensor is the original tensor. A compressed representation of a tensor.
  • the embodiment of the present invention uses the tensor decomposition to decompose the hyperspectral image on the spectral dimension based on the third-order tensor model of the hyperspectral image. Can effectively reduce the spectral redundant information of hyperspectral images.
  • the hyperspectral image is reconstructed into a third-order kernel tensor for representation by tensor decomposition, so that the spatial-domain coding is invariant to the spectral resolution.
  • T represents a tensor model of hyperspectral image.
  • Tensor decomposition of T is shown as follows:
  • T T ′ ⁇ 3 L T
  • L T is the principal component on the modulo 3 fiber of T
  • ⁇ 3 represents the modulo 3 fiber multiplication of the tensor
  • T′ ⁇ R X ⁇ Y ⁇ ⁇ ′ is the nuclear tensor of T after tensor decomposition
  • X, Y, ⁇ ′ represent the sizes of the mode 1 fiber, mode 2 fiber, and mode 3 fiber of T ′
  • T is compressed by the transpose matrix L T of its projection matrix to obtain the corresponding nuclear tensor T ′, and the hyperspectral image with the number of bands ⁇ is mapped to the tensor with the fiber degree ⁇ ′ of module 3, which can be effectively
  • the correlation and redundant information between the bands are reduced, and the spectral curve of each pixel faces the more refined feature category information.
  • This can not only significantly improve the information amount and efficiency of the spatial spectral domain LBP, but also enhance the invariance of the feature's resistance to changes in the spectral domain scale.
  • FIG. 3 fiber decomposition diagram.
  • M, N are the sizes of the D 1's mode 1 and 2 fibers, and U 'is the size of the D'' s mode 3 fibers.
  • Step 202 Traverse each positive slice in the third-order kernel tensor, and use a preset encoding function to establish a feature histogram of each positive slice.
  • step 202 includes:
  • Step 1 determine the domain feature information of the tensor elements in the traversed front slice
  • Step 2 quantify the feature information of the domain to obtain a spatially coded joint coding value of the tensor element
  • Step 3 Determine the feature statistical histogram of the traversed frontal slice by using the joint coding values in the spatial spectrum domain.
  • the tensor elements in the front slice of the third-order kernel tensor Neighborhood feature information V defined as The joint probability distribution of the values of the tensor elements in the local neighborhood of the spatial spectrum domain is as follows:
  • the step of quantizing the domain characteristic information in the above step 2 to obtain a spatially coded joint coding value of the tensor element includes:
  • Step a adding a preset binomial coefficient to the domain characteristic information
  • Step b The domain feature information carrying the binomial coefficient is converted into a decimal number to obtain a spatially spectral domain joint coding value of the tensor element.
  • TSSLBP P, J, R will produce 2 P + J different output values, that is , the number of binary modes of TSSLBP P, J, R encoding is 2 P + J.
  • the rotation operation only occurs in the pixels in the spatial domain, and the pixels in the spectral domain do not change, so it has a space-invariant encoded value of the rotation invariance Is defined as follows:
  • min ( ⁇ ) represents the minimum value
  • ROR (x, r) represents the binary value of a P-bit binary number x after circular right shift r times.
  • TSSLBP (m, n, u) represents the TSSLBP encoding value of a pixel whose spatial position is (m, n) and whose modulus is u.
  • b represents any binary mode of TSSLBP encoding, b ⁇ [0, 2 P + J ] and b is an integer.
  • Step 203 Join the statistical histograms of the features of the respective front slices to obtain the statistical histogram of the preset local area, and use the statistical histogram of the preset local area to classify the hyperspectral image. .
  • the statistical histograms on U ′ front slices need to be connected as the feature statistical histogram h of the local region D ′.
  • step of joining the feature statistical histograms of the respective front slices in step 203 includes:
  • the feature statistical histograms of the respective front slices are connected.
  • the method for improving the classification accuracy of a hyperspectral image provided by the embodiment of the present invention firstly establishes a third-order tensor of a preset local area of the hyperspectral image, and performs tensor decomposition on the third-order tensor. To obtain the third-order kernel tensor corresponding to the preset local area, then establish the feature statistical histogram of each front slice in the third-order kernel tensor, and finally combine the feature statistical histograms of each front slice to obtain the statistical histogram of the preset local area Use this statistical histogram to classify hyperspectral images.
  • each of the above frontal slices can reflect the joint distribution of the spectral values of different wavebands in the hyperspectral image and the spatial distribution information, the information of the spectral and spatial domains of the hyperspectral image can be represented simultaneously by the feature statistical histogram of each frontal slice. Therefore, the present invention uses the statistical histogram to classify the hyperspectral image, which can effectively improve the classification accuracy.
  • FIG. 4 is a schematic diagram of a virtual program module of the device for improving classification accuracy of hyperspectral images according to an embodiment of the present invention.
  • the foregoing device includes:
  • the establishing module 401 is configured to establish a third-order tensor of a preset local region of a hyperspectral image, and perform tensor decomposition on the third-order tensor to obtain a third-order kernel tensor corresponding to the preset local region.
  • the operation module 402 is configured to traverse each positive slice in the third-order kernel tensor, and use a preset coding function to establish a feature histogram of each positive slice.
  • a classification module 403 is configured to join the statistical histograms of the features of the respective front slices to obtain the statistical histograms of the preset local regions, and use the statistical histograms of the preset local regions to perform the hyperspectral Images are classified.
  • the device for improving classification accuracy of hyperspectral images can realize: establishing a third-order tensor of a preset local area of a hyperspectral image, performing tensor decomposition on the third-order tensor, and obtaining a correspondence of the preset local area A third-order kernel tensor; traverse each positive slice in the third-order kernel tensor, and use a preset encoding function to build a feature statistical histogram of each positive slice; join the feature statistical histograms of each positive slice to obtain a preset local The statistical histogram of the region, and the statistical histogram of the preset local region is used to classify the above hyperspectral image.
  • the statistical histogram obtained in the embodiment of the present invention can not only reflect the information in the spatial domain of the hyperspectral image, but also reflect the information in the spectral domain, so it can effectively improve the classification of hyperspectral images. Classification accuracy.
  • FIG. 5 is a schematic diagram of a detailed program module of an operation module 402 according to an embodiment of the present invention.
  • the operation module 402 includes:
  • the first determining module 501 is configured to determine domain feature information of a tensor element in the traversed front slice.
  • a quantization module 502 is configured to quantize the feature information of the domain to obtain a spatial-spectrum-domain joint coding value of the tensor element.
  • a second determining module 503 is configured to determine a statistical histogram of features of the traversed frontal slice by using the spatially coded joint coding values.
  • the quantization module 502 is specifically configured to:
  • the preset binomial coefficient is added to the domain feature information, and the domain feature information carrying the binomial coefficient is converted into a decimal number to obtain a joint coding value of the spatial spectrum domain of the tensor element.
  • classification module 403 includes:
  • a linking module is configured to link the feature statistical histograms of the frontal slices according to the position order of the frontal slices in the third-order kernel tensor.
  • the device for improving the classification accuracy of hyperspectral images can realize: establishing a third-order tensor of a preset local area of a hyperspectral image, and performing expansion on the third-order tensor Quantitative decomposition to obtain the third-order kernel tensor corresponding to the preset local area, and then establish the feature statistical histogram of each front slice in the third-order kernel tensor, and finally combine the feature statistical histograms of each front slice to obtain the statistics of the preset local area. Histogram. Use this statistical histogram to classify hyperspectral images.
  • each of the above frontal slices can reflect the joint distribution of the spectral values of different wavebands in the hyperspectral image and the spatial distribution information, the information of the spectral and spatial domains of the hyperspectral image can be represented simultaneously by the feature statistical histogram of each frontal slice. Therefore, the present invention uses the statistical histogram to classify the hyperspectral image, which can effectively improve the classification accuracy.
  • An embodiment of the present invention further provides an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor.
  • the processor executes the computer program, the method for improving classification accuracy of hyperspectral images according to the present invention is implemented. Corresponds to each step in each embodiment.
  • An embodiment of the present invention also provides a readable storage medium.
  • the readable storage medium is a computer-readable storage medium having a computer program stored thereon.
  • the invention improves the classification accuracy of hyperspectral images.
  • the method corresponds to each step in each embodiment.
  • FIG. 6 is a schematic structural diagram of an electronic device provided in an embodiment of the present invention.
  • the electronic device 06 of this embodiment mainly includes a processor 60, a memory 61, and a computer program 62 stored in the memory 61 and executable on the processor 60, for example, a program for improving classification accuracy of a hyperspectral image .
  • the processor 60 executes the computer program 62
  • the steps in the embodiments of the method for improving the classification accuracy of the hyperspectral image are implemented, for example, the steps shown in any one of FIGS. 1 to 3.
  • the processor 60 executes the computer program 62
  • the functions of each module / unit in the foregoing device embodiments are implemented, for example, the functions of each module shown in FIG. 4.
  • the computer program 62 may be divided into one or more modules / units, and one or more modules / units are stored in the memory 61 and executed by the processor 60 to complete the present invention.
  • One or more modules / units may be a series of computer program instruction sections capable of performing a specific function, and the instruction sections are used to describe the execution process of the computer program 62 in the computing device 06.
  • the computer program 62 may be divided into functions of a building module 401, a computing module 402, and a classification module 403 (modules in a virtual device).
  • the computing device 06 may include, but is not limited to, a processor 60 and a memory 61. Those skilled in the art can understand that FIG. 6 is only an example of the computing device 06, and does not constitute a limitation on the computing device 06. It may include more or fewer components than shown in the figure, or combine some components or different components. For example, computing devices may also include input and output devices, network access devices, and buses.
  • the so-called processor 60 may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application specific integrated circuits (ASICs), Ready-made programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.
  • a general-purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
  • the memory 61 may be an internal storage unit of the computing device 06, such as a hard disk or a memory of the computing device 06.
  • the memory 61 may also be an external storage device of the computing device 06, such as a plug-in hard disk, a smart memory card (SMC), a secure digital (SD) card, and a flash memory card (Flash) provided on the computing device 06. Card) and so on.
  • the memory 61 may include both an internal storage unit of the computing device 06 and an external storage device.
  • the memory 61 is used to store computer programs and other programs and data required by the computing device.
  • the memory 61 may also be used to temporarily store data that has been output or is to be output.
  • the disclosed apparatus and method may be implemented in other ways.
  • the device embodiments described above are only schematic.
  • the division of the modules is only a logical function division.
  • multiple modules or components may be combined or Can be integrated into another system, or some features can be ignored or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or modules, which may be electrical, mechanical or other forms.
  • the modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical modules, may be located in one place, or may be distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the objective of the solution of this embodiment.
  • each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist separately physically, or two or more modules may be integrated into one module.
  • the above integrated modules may be implemented in the form of hardware or software functional modules.
  • the integrated module When the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.
  • the technical solution of the present invention essentially or part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium , Including a number of instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.
  • the foregoing storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes .

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Artificial Intelligence (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

本发明公开了一种提升高光谱图像分类精度的方法、装置、设备及存储介质,方法包括:建立高光谱图像预设局部区域的三阶张量,对该三阶张量进行张量分解,得到预设局部区域对应的三阶核张量;遍历该三阶核张量中的各个正面切片,利用预设的编码函数建立各个正面切片的特征统计直方图;对各个正面切片的特征统计直方图进行联接,得到预设局部区域的统计直方图,并利用预设局部区域的统计直方图,对上述高光谱图像进行分类。相较于现有技术而言,本发明实施例得到的统计直方图不仅可以反映高光谱图像空间域上的信息,而且还能反映光谱域的信息,因此在高光谱图像分类中能够有效的提升分类精度。

Description

提升高光谱图像分类精度的方法、装置、设备及存储介质 技术领域
本发明涉及图像处理技术领域,尤其涉及一种提升高光谱图像分类精度的方法、装置、设备及存储介质。
背景技术
高光谱图像不仅包含了反映地物形态特征的空间结构信息,而且包含了反映地物类别属性和状态属性的光谱结构信息,称为“图谱合一”。这种“图谱合一”的技术极大地提高了人类的感官能力,在农业科学、材料、地球科学、食品工业、生物医学应用以及军事等领域得到了广泛的应用。
LBP(Local Binary Pattern,局部二值模式)算法是一种简单有效,计算复杂度低的图像纹理描述算法,能够深入挖掘图像中蕴含的空间信息,目前已经被广泛用于与分类相关的任务,包括纹理分析、目标检测、目标识别等。然而,现有的应用在高光谱图像上的LBP算法着重于空间信息的描述,而忽略了光谱信息,导致高光谱图像的分类精度较低。
发明内容
本发明的主要目的在于提供一种提升高光谱图像分类精度的方法、装置、设备及存储介质,旨在解决现有技术中利用LBP算法对高光谱图像进行分类时,分类精度较低的技术问题。
为实现上述目的,本发明第一方面提供一种提升高光谱图像分类精度的方法,该方法包括:
建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进行张量分解, 得到所述预设局部区域对应的三阶核张量;
遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图;
对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
可选的,所述遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图的步骤包括:
确定遍历到的正面切片中的张量元素的领域特征信息;
对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值;
利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
可选的,所述对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值的步骤包括:
将预设的二项式系数添加至所述领域特征信息中;
将携带有所述二项式系数的领域特征信息转换为十进制数,得到所述张量元素的空谱域联合编码值。
可选的,所述对所述各个正面切片的特征统计直方图进行联接的步骤包括:
按照所述各个正面切片在所述三阶核张量中的位置顺序,对所述各个正面切片的特征统计直方图进行联接。
为实现上述目的,本发明第二方面提供一种提升高光谱图像分类精度的装置,该装置包括:
建立模块,用于建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进行张量分解,得到所述预设局部区域对应的三阶核张量;
运算模块,用于遍历所述三阶核张量中的各个正面切片,利用预设的编码 函数建立所述各个正面切片的特征统计直方图;
分类模块,用于对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
可选的,所述运算模块包括:
第一确定模块,用于确定遍历到的正面切片中的张量元素的领域特征信息;
量化模块,用于对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值;
第二确定模块,用于利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
可选的,所述量化模块具体用于:
将预设的二项式系数添加至所述领域特征信息中,并将携带有所述二项式系数的领域特征信息转换为十进制数,得到所述张量元素的空谱域联合编码值。可选的,所述分类模块包括:
联接模块,用于按照所述各个正面切片在所述三阶核张量中的位置顺序,对所述各个正面切片的特征统计直方图进行联接。
为实现上述目的,本发明第三方面提供一种电子设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,所述处理器执行所述计算机程序时实现本发明第一方面提供的提升高光谱图像分类精度的方法中的各个步骤。
为实现上述目的,本发明第四方面提供一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,所述计算机程序被处理器执行时实现本发明第一方面提供的提升高光谱图像分类精度的方法中的各个步骤。
本发明提供一种提升高光谱图像分类精度的方法,相较于现有技术而言,先建立高光谱图像预设局部区域的三阶张量,对该三阶张量进行张量分解,得 到预设局部区域对应的三阶核张量,然后建立三阶核张量中各个正面切片的特征统计直方图,最后对各个正面切片的特征统计直方图进行联接得到预设局部区域的统计直方图,利用该统计直方图即可对高光谱图像进行分类。由于上述各个正面切片能够反映出高光谱图像中不同波段的光谱值联合分布情况以及空间分布信息,故可以通过各个正面切片的特征统计直方图来同时表示高光谱图像光谱域与空间域上的信息,即利用上述统计直方图对高光谱图像进行分类,能够有效的提升分类精度。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案,下面将对实施例或现有技术描述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为本发明实施例中高光谱图像的三阶张量模型示意图;
图2为本发明实施例中提升高光谱图像分类精度的方法的步骤流程示意图;
图3为本发明实施例中三阶张量T的模3纤的分解示意图;
图4为本发明实施例中提升高光谱图像分类精度的装置的虚拟程序模块示意图;
图5为本发明实施例中运算模块402的细化程序模块示意图;
图6为本发明实施例中提供的电子设备的结构示意图。
具体实施方式
为使得本发明的发明目的、特征、优点能够更加的明显和易懂,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实施例,而非全部实施例。基 于本发明中的实施例,本领域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,都属于本发明保护的范围。
本发明实施例中,高光谱图像可以表示为I(x,y,λ),其中,(x,y)是空间域坐标,λ是光谱域坐标。其中,可以将高光谱图像建模为三阶张量,即用三阶张量T∈R X×Y×Λ表示I。其中,X和Y是高光谱图像空间域的大小,即(x,y)的取值区间;Λ是高光谱图像光谱域的大小(波段数),即λ的取值区间;高光谱图像的每个像元I(x,y,λ)对应着三阶张量T的每个元素T xyλ。具体可参照图1,图1为高光谱图像的三阶张量模型示意图。
其中,张量的纤和切片是构成张量的基本元素之一,通过张量的纤和切片可实现高维几何体中向量或矩阵的提取,有效地降低了计算复杂度。张量的纤定义为变化任意一阶的索引,固定其他阶的索引。高光谱图像的三阶张量T的纤分为行(模1纤)、列(模2纤)和管纤(模3纤),分别记为T :yλ、T x:λ和T xy:。通过该模型可将空域坐标为(x,y)的像元映射为n维矢量
Figure PCTCN2018103204-appb-000001
进行表示,结合模3纤模型对高光谱图像进行分析能够有效提高对空谱域信息利用的效率。
其中,切片是张量的二维部分,它通过固定任意两个阶的索引来实现,通常用于进行矩阵抽取的操作。高光谱图像的三阶张量T的切片定义为水平切片、侧面切片和正面切片,分别表示为T x::、T :y:和T ::λ,其中正面切片能够有效表示高光谱图像中不同波段的光谱值联合分布。通过对T中三个不同方向的张量元素进行矩阵抽取,能够有效结合空间域和光谱域的信息,提高空谱域编码的效率。
本发明实施例中,在高光谱图像三阶张量模型的基础上,采用空谱域联合概率分布来表示高光谱图像的特征,提出了一种提升高光谱图像分类精度的方法,具体请参阅图2,图2为本发明实施例中提升高光谱图像分类精度的方法的步骤流程示意图,本发明实施例中,上述提升高光谱图像分类精度的方法包括:
步骤201、建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进 行张量分解,得到所述预设局部区域对应的三阶核张量。
首先,由于高光谱成像仪的光谱成像间隔小,所以相邻波段的图像相关性高,波段间存在着严重的冗余数据,为了降低高光谱图像的冗余信息,提取到更加精炼的有用信息并提高特征编码的效率,需要剔除相关性较大的冗余信息,压缩数据量。由张量数学可知,张量的模n乘法是指张量
Figure PCTCN2018103204-appb-000002
与一个阵
Figure PCTCN2018103204-appb-000003
相乘,乘积结果仍为一个张量,其模的大小为:
I 1×I 2×…I n-1×J×I n-1×…×I N
当J<I n时,张量的模n乘法可看作是一个降维的过程。
基于模n乘法的张量分解是对原张量进行压缩表示的重要方法,其可将张量表示为核张量与各个阶上的投影矩阵连乘的形式,其中,核张量是对原有张量的压缩表示。为了能够充分考虑不同波段间的相关性,并能够消除频谱冗余,本发明实施例在高光谱图像三阶张量模型的基础上,利用张量分解对高光谱图像在光谱维上进行分解,可以有效的降低高光谱图像的光谱冗余信息。通过张量分解将高光谱图像重构为三阶核张量进行表示,使得空谱域编码具有光谱分辨率不变性。
其中,假设T表示高光谱图像的张量模型,对T进行张量分解如下式所示:
T=T′× 3L T
其中,L T是T的模3纤上的主分量,× 3表示张量的模3纤乘法,T′∈R X×Y×Λ′是T经过张量分解后的核张量,X,Y,Λ′分别表示T′的模1纤、模2纤和模3纤的大小,Λ′≤Λ。上式中,T经过其投影矩阵的转置矩阵L T的压缩得到了相应的核张量T′,波段数为Λ的高光谱图像被映射到模3纤维度为Λ′的张量中,能够有效地降低波段间的相关性和冗余信息,每个像元的光谱曲线都对着更加精炼的地物类别信息。这不仅能够明显提高空谱域LBP的信息量和效率,而且可以增强特征的抗光谱域尺度变化的不变性,具体可参照图3,图3是本发明实施例中三阶张量T的模3纤分解示意图。
具体的,设高光谱图像预设局部区域的三阶张量用D∈R M×N×U表示,其经过 张量分解后的三阶核张量为D′∈R M×N×U′,且D′是三阶核张量T′的子集。其中,M,N是D′的模1纤和模2纤的大小,U′则是D′的模3纤大小。
步骤202、遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图。
具体的,上述步骤202包括:
步骤一、确定遍历到的正面切片中的张量元素的领域特征信息;
步骤二、对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值;
步骤三、利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
其中,设
Figure PCTCN2018103204-appb-000004
为T′的正面切片,设
Figure PCTCN2018103204-appb-000005
Figure PCTCN2018103204-appb-000006
则借鉴灰度图像的LBP思想,将三阶核张量的正面切片中的张量元素
Figure PCTCN2018103204-appb-000007
的邻域特征信息V,定义为
Figure PCTCN2018103204-appb-000008
的空谱域局部邻域内的张量元素值的联合概率分布,如下所示:
Figure PCTCN2018103204-appb-000009
其中,S k(k=0,1,…,P-1)是在正面切片
Figure PCTCN2018103204-appb-000010
上,以
Figure PCTCN2018103204-appb-000011
为圆心,半径为R(R>0)的圆形邻域上等间隔采样的P个采样点的张量元素值,S k(k=0,1,…,P-1)的坐标为(x c+Rsin(2πi/P),y c-Rcos(2πi/P),λ′ c),i∈{1,2,…,P}。S k(k=P,P+1,…,P+J-1)则是
Figure PCTCN2018103204-appb-000012
所在的模3纤
Figure PCTCN2018103204-appb-000013
上的J个采样点的张量元素值,其坐标为(x c,y c,λ′ c+j),j∈{1,2,…,J}。
其中,为了实现光照不变性,将每个采样点的张量元素值S k
Figure PCTCN2018103204-appb-000014
相减,得到:
Figure PCTCN2018103204-appb-000015
其中,由于差值
Figure PCTCN2018103204-appb-000016
Figure PCTCN2018103204-appb-000017
的值是相互独立的,所以上式可以因式分解为:
Figure PCTCN2018103204-appb-000018
在实际的高光谱图像中,并不能保证差值
Figure PCTCN2018103204-appb-000019
Figure PCTCN2018103204-appb-000020
的值是相互独立的,所以上式是对空谱域内张量元素值的联合概率分布的一种近似。而由于联合概率分布的大部分信息都包含在差分分布里面,则上述公式还可表示为:
Figure PCTCN2018103204-appb-000021
通过仅统计中心点的空谱域局部邻域内像元的符号信息,而不是像元的真实光谱值以实现联合差分分布的光照不变性,如下所示:
Figure PCTCN2018103204-appb-000022
其中,
Figure PCTCN2018103204-appb-000023
进一步的,上述步骤二中对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值的步骤包括:
步骤a、将预设的二项式系数添加至所述领域特征信息中;
步骤b、将携带有所述二项式系数的领域特征信息转换为十进制数,得到所述张量元素的空谱域联合编码值。
其中,通过对每个符号
Figure PCTCN2018103204-appb-000024
赋予二项式系数2 k,把V转化成一个可以完整有效地描述以
Figure PCTCN2018103204-appb-000025
为中心的局部邻域内的空谱域联合编码值TSSLBP P,J,R,如以下公式所示:
Figure PCTCN2018103204-appb-000026
由上述公式可知,TSSLBP P,J,R将产生2 P+J个不同的输出值,即TSSLBP P,J,R编码的二值模式个数为2 P+J
由于高光谱图像的特殊性,旋转操作只发生在空间域的像元,光谱域的像元并不会改变,因此具备有旋转不变性的空谱域编码值
Figure PCTCN2018103204-appb-000027
的定义如下:
Figure PCTCN2018103204-appb-000028
上式中,min(·)表示求取最小值,ROR(x,r)表示对一个P位的二进制数x进 行圆形右位移r次后的二进制数值。最终,进行直方图统计,统计所得到的特征矢量作为高光谱图像局部区域的空谱域特征描述。
其中,对于D′而言,令第u个切片的特征统计直方图为h u,其中u的取值范围是u=1,2,3,…,U′,统计直方图h u的每个柱用h u(b)表示,可由下式计算得到:
Figure PCTCN2018103204-appb-000029
其中,
Figure PCTCN2018103204-appb-000030
TSSLBP(m,n,u)代表空间位置为(m,n),模3纤为u的像元的TSSLBP编码值。b代表TSSLBP编码的任意一种二值模式,b∈[0,2 P+J]且b为整数。
步骤203、对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
本发明实施例中,在得到每个正面切片的局部特征统计直方图后,需要将U′个正面切片上的统计直方图进行联接,作为局部区域D′的特征统计直方图h。
其中,上述步骤203中对所述各个正面切片的特征统计直方图进行联接的步骤包括:
按照所述各个正面切片在所述三阶核张量中的位置顺序,对所述各个正面切片的特征统计直方图进行联接。
具体的,h的计算方法如下公式所示:
h=[h 1,h 2,…h u,…,h U′]
进一步的,为了更好的理解本发明实施例,本发明实施例中,提供一种实现本发明提升高光谱图像分类精度的方法的伪代码:
输入:高光谱图像I;
建立高光谱图像I的三阶张量T∈R X×Y×Λ
对三阶张量T进行张量分解,得到三阶张量T的三阶核张量T′∈R X×Y×Λ′
for第u个正面切片;in局部区域D′∈R M×N×U′
for空间位置为(m,n)的像元;in第u个正面切片;
利用特征编码函数计算TSSLBP编码;
End for;
建立第u个正面切片的特征统计直方图h u
End for;
for第u个正面切片;in局部区域D′∈R M×N×U′
h=[h,h u];//联接每个正面切片的特征统计直方图;
End for;
输出:局部区域D′∈R M×N×U′的统计直方图h。
本发明实施例所提供的提升高光谱图像分类精度的方法,相较于现有技术而言,先建立高光谱图像预设局部区域的三阶张量,对该三阶张量进行张量分解,得到预设局部区域对应的三阶核张量,然后建立三阶核张量中各个正面切片的特征统计直方图,最后对各个正面切片的特征统计直方图进行联接得到预设局部区域的统计直方图,利用该统计直方图即可对高光谱图像进行分类。由于上述各个正面切片能够反映出高光谱图像中不同波段的光谱值联合分布情况以及空间分布信息,因此可以通过各个正面切片的特征统计直方图来同时表示高光谱图像光谱域与空间域上的信息,故本发明利用上述统计直方图对高光谱图像进行分类,能够有效的提升分类精度。
进一步地,本发明实施例还提供一种提升高光谱图像分类精度的装置,参照图4,图4为本发明实施例中提升高光谱图像分类精度的装置的虚拟程序模块示意图,上述装置包括:
建立模块401,用于建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进行张量分解,得到所述预设局部区域对应的三阶核张量。
运算模块402,用于遍历所述三阶核张量中的各个正面切片,利用预设的 编码函数建立所述各个正面切片的特征统计直方图。
分类模块403,用于对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
本发明实施例所提供的提升高光谱图像分类精度的装置,可以实现:建立高光谱图像预设局部区域的三阶张量,对该三阶张量进行张量分解,得到预设局部区域对应的三阶核张量;遍历该三阶核张量中的各个正面切片,利用预设的编码函数建立各个正面切片的特征统计直方图;对各个正面切片的特征统计直方图进行联接,得到预设局部区域的统计直方图,并利用预设局部区域的统计直方图,对上述高光谱图像进行分类。相较于现有技术而言,本发明实施例得到的统计直方图不仅可以反映高光谱图像空间域上的信息,而且还能反映光谱域的信息,因此在高光谱图像分类中能够有效的提升分类精度。
进一步地,参照图5,图5为本发明实施例中运算模块402的细化程序模块示意图,运算模块402包括:
第一确定模块501,用于确定遍历到的正面切片中的张量元素的领域特征信息。
量化模块502,用于对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值。
第二确定模块503,用于利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
其中,量化模块502具体用于:
将预设的二项式系数添加至所述领域特征信息中,并将携带有二项式系数的领域特征信息转换为十进制数,得到上述张量元素的空谱域联合编码值。
另外,分类模块403包括:
联接模块,用于按照所述各个正面切片在所述三阶核张量中的位置顺序, 对所述各个正面切片的特征统计直方图进行联接。
本发明实施例所提供的提升高光谱图像分类精度的装置,相较于现有技术而言,可以实现:建立高光谱图像预设局部区域的三阶张量,对该三阶张量进行张量分解,得到预设局部区域对应的三阶核张量,然后建立三阶核张量中各个正面切片的特征统计直方图,最后对各个正面切片的特征统计直方图进行联接得到预设局部区域的统计直方图,利用该统计直方图即可对高光谱图像进行分类。由于上述各个正面切片能够反映出高光谱图像中不同波段的光谱值联合分布情况以及空间分布信息,因此可以通过各个正面切片的特征统计直方图来同时表示高光谱图像光谱域与空间域上的信息,故本发明利用上述统计直方图对高光谱图像进行分类,能够有效的提升分类精度。
本发明实施例还提供一种电子设备,包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序,处理器执行计算机程序时,实现本发明提升高光谱图像分类精度的方法对应各个实施例中的各个步骤。
本发明实施例还提供一种可读存储介质,该可读存储介质为计算机可读存储介质,其上存储有计算机程序,计算机程序被处理器执行时,实现本发明提升高光谱图像分类精度的方法对应各个实施例中的各个步骤。
为了更好的理解本发明,参照图6,图6为本发明实施例中提供的电子设备的结构示意图。如图6所示,该实施例的电子设备06主要包括:处理器60、存储器61以及存储在存储器61中并可在处理器60上运行的计算机程序62,例如提升高光谱图像分类精度的程序。处理器60执行计算机程序62时实现上述提升高光谱图像分类精度的方法各实施例中的步骤,例如图1至附图3任一示例所示的步骤。或者,处理器60执行计算机程序62时实现上述各装置实施例中各模块/单元的功能,例如图4所示各模块的功能。
计算机程序62可以被分割成一个或多个模块/单元,一个或者多个模块/单元被存储在存储器61中,并由处理器60执行,以完成本发明。一个或多个模块/单元可以是能够完成特定功能的一系列计算机程序指令段,该指令段用于描 述计算机程序62在计算设备06中的执行过程。例如,计算机程序62可以被分割成建立模块401、运算模块402、分类模块403(虚拟装置中的模块)的功能。
计算设备06可包括但不仅限于处理器60、存储器61。本领域技术人员可以理解,图6仅仅是计算设备06的示例,并不构成对计算设备06的限定,可以包括比图示更多或更少的部件,或者组合某些部件,或者不同的部件,例如计算设备还可以包括输入输出设备、网络接入设备、总线等。
所称处理器60可以是中央处理单元(Central Processing Unit,CPU),还可以是其他通用处理器、数字信号处理器(Digital Signal Processor,DSP)、专用集成电路(Application Specific Integrated Circuit,ASIC)、现成可编程门阵列(Field-Programmable Gate Array,FPGA)或者其他可编程逻辑器件、分立门或者晶体管逻辑器件、分立硬件组件等。通用处理器可以是微处理器或者该处理器也可以是任何常规的处理器等。
存储器61可以是计算设备06的内部存储单元,例如计算设备06的硬盘或内存。存储器61也可以是计算设备06的外部存储设备,例如计算设备06上配备的插接式硬盘,智能存储卡(Smart Media Card,SMC),安全数字(Secure Digital,SD)卡,闪存卡(Flash Card)等。进一步地,存储器61还可以既包括计算设备06的内部存储单元也包括外部存储设备。存储器61用于存储计算机程序以及计算设备所需的其他程序和数据。存储器61还可以用于暂时地存储已经输出或者将要输出的数据。
在本申请所提供的几个实施例中,应该理解到,所揭露的装置和方法,可以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,所述模块的划分,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,例如多个模块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,装置或模块的间接耦合或通信连接,可以是电性,机械或其它的形式。
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的,作为模块显示的部件可以是或者也可以不是物理模块,即可以位于一个地方,或者也可以分布到多个网络模块上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。
另外,在本发明各个实施例中的各功能模块可以集成在一个处理模块中,也可以是各个模块单独物理存在,也可以两个或两个以上模块集成在一个模块中。上述集成的模块既可以采用硬件的形式实现,也可以采用软件功能模块的形式实现。
所述集成的模块如果以软件功能模块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(可以是个人计算机,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、移动硬盘、只读存储器(ROM,Read-Only Memory)、随机存取存储器(RAM,Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
需要说明的是,对于前述的各方法实施例,为了简便描述,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它顺序或者同时进行。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都是本发明所必须的。
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,可以参见其它实施例的相关描述。
以上为对本发明所提供的一种提升高光谱图像分类精度的方法、装置、设 备及存储介质的描述,对于本领域的技术人员,依据本发明实施例的思想,在具体实施方式及应用范围上均会有改变之处,综上,本说明书内容不应理解为对本发明的限制。

Claims (10)

  1. 一种提升高光谱图像分类精度的方法,其特征在于,所述方法包括:
    建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进行张量分解,得到所述预设局部区域对应的三阶核张量;
    遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图;
    对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
  2. 如权利要求1所述的方法,其特征在于,所述遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图的步骤包括:
    确定遍历到的正面切片中的张量元素的领域特征信息;
    对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值;
    利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
  3. 如权利要求2所述的方法,其特征在于,所述对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值的步骤包括:
    将预设的二项式系数添加至所述领域特征信息中;
    将携带有所述二项式系数的领域特征信息转换为十进制数,得到所述张量元素的空谱域联合编码值。
  4. 如权利要求1至3任意一项所述的方法,其特征在于,所述对所述各个正面切片的特征统计直方图进行联接的步骤包括:
    按照所述各个正面切片在所述三阶核张量中的位置顺序,对所述各个正面切片的特征统计直方图进行联接。
  5. 一种提升高光谱图像分类精度的装置,其特征在于,所述装置包括:
    建立模块,用于建立高光谱图像预设局部区域的三阶张量,对所述三阶张量进行张量分解,得到所述预设局部区域对应的三阶核张量;
    运算模块,用于遍历所述三阶核张量中的各个正面切片,利用预设的编码函数建立所述各个正面切片的特征统计直方图;
    分类模块,用于对所述各个正面切片的特征统计直方图进行联接,得到所述预设局部区域的统计直方图,并利用所述预设局部区域的统计直方图,对所述高光谱图像进行分类。
  6. 如权利要求5所述的装置,其特征在于,所述运算模块包括:
    第一确定模块,用于确定遍历到的正面切片中的张量元素的领域特征信息;
    量化模块,用于对所述领域特征信息进行量化,得到所述张量元素的空谱域联合编码值;
    第二确定模块,用于利用所述空谱域联合编码值,确定所述遍历到的正面切片的特征统计直方图。
  7. 如权利要求6所述的装置,其特征在于,所述量化模块具体用于:
    将预设的二项式系数添加至所述领域特征信息中,并将携带有所述二项式系数的领域特征信息转换为十进制数,得到所述张量元素的空谱域联合编码值。
  8. 如权利要求5至7任意一项所述的装置,其特征在于,所述分类模块包括:
    联接模块,用于按照所述各个正面切片在所述三阶核张量中的位置顺序,对所述各个正面切片的特征统计直方图进行联接。
  9. 一种电子设备,包括存储器、处理器以及存储在所述存储器中并可在所述处理器上运行的计算机程序,其特征在于,所述处理器执行所述计算机程序时实现如权利要求1至4任意一项所述的提升高光谱图像分类精度的方法中的各个步骤。
  10. 一种计算机可读存储介质,所述计算机可读存储介质存储有计算机程序,其特征在于,所述计算机程序被处理器执行时实现如权利要求1至4任意一项所述的提升高光谱图像分类精度的方法中的各个步骤。
PCT/CN2018/103204 2018-08-30 2018-08-30 提升高光谱图像分类精度的方法、装置、设备及存储介质 WO2020042070A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/103204 WO2020042070A1 (zh) 2018-08-30 2018-08-30 提升高光谱图像分类精度的方法、装置、设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2018/103204 WO2020042070A1 (zh) 2018-08-30 2018-08-30 提升高光谱图像分类精度的方法、装置、设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020042070A1 true WO2020042070A1 (zh) 2020-03-05

Family

ID=69642618

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/103204 WO2020042070A1 (zh) 2018-08-30 2018-08-30 提升高光谱图像分类精度的方法、装置、设备及存储介质

Country Status (1)

Country Link
WO (1) WO2020042070A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435265A (zh) * 2021-06-09 2021-09-24 深圳大学 高光谱图像分类方法、装置、电子设备及存储介质

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005057129A1 (de) * 2003-12-15 2005-06-23 Deutsches Zentrum für Luft- und Raumfahrt e.V. Verfahren und vorrichtung zur mindestens teilweise automatisierten auswertung von fernerkundungsdaten
CN105809205A (zh) * 2016-03-31 2016-07-27 深圳大学 一种高光谱图像的分类方法及其系统
CN106845517A (zh) * 2016-12-14 2017-06-13 西北大学 一种基于Tucker分解的光谱张量降维及分类方法

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2005057129A1 (de) * 2003-12-15 2005-06-23 Deutsches Zentrum für Luft- und Raumfahrt e.V. Verfahren und vorrichtung zur mindestens teilweise automatisierten auswertung von fernerkundungsdaten
CN105809205A (zh) * 2016-03-31 2016-07-27 深圳大学 一种高光谱图像的分类方法及其系统
CN106845517A (zh) * 2016-12-14 2017-06-13 西北大学 一种基于Tucker分解的光谱张量降维及分类方法

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113435265A (zh) * 2021-06-09 2021-09-24 深圳大学 高光谱图像分类方法、装置、电子设备及存储介质

Similar Documents

Publication Publication Date Title
Wee et al. Image quality assessment by discrete orthogonal moments
KR101912748B1 (ko) 확장성을 고려한 특징 기술자 생성 및 특징 기술자를 이용한 정합 장치 및 방법
US20160048849A1 (en) Method and system for clustering and classifying online visual information
WO2018090937A1 (zh) 图像处理方法、终端及存储介质
CN108765322B (zh) 图像去噪方法及装置
Diana Andrushia et al. Saliency-based image compression using Walsh–Hadamard transform (WHT)
WO2021258699A1 (zh) 一种图像识别方法、装置、电子设备和计算机可读介质
WO2021164269A1 (zh) 基于注意力机制的视差图获取方法和装置
CN110532413B (zh) 基于图片匹配的信息检索方法、装置、计算机设备
Liang et al. Efficient hashing method using 2D-2D PCA for image copy detection
CN111553215A (zh) 人员关联方法及其装置、图卷积网络训练方法及其装置
CN110866881A (zh) 图像处理方法及装置、存储介质和电子设备
CN116188805B (zh) 海量图像的图像内容分析方法、装置和图像信息网络
US20230401756A1 (en) Data Encoding Method and Related Device
CN111783774A (zh) 图像处理方法、设备及存储介质
CN109034274B (zh) 提升高光谱图像分类精度的方法、装置、设备及存储介质
WO2022257407A1 (zh) 高光谱图像分类方法、装置、电子设备及存储介质
CN113033448B (zh) 一种基于多尺度卷积和注意力的遥感影像去云残差神经网络系统、方法、设备及存储介质
CN114049491A (zh) 指纹分割模型训练、指纹分割方法、装置、设备及介质
WO2020042070A1 (zh) 提升高光谱图像分类精度的方法、装置、设备及存储介质
Wang et al. An unsupervised heterogeneous change detection method based on image translation network and post-processing algorithm
CN113918598A (zh) 一种乘积量化搜索方法、装置、终端和存储介质
WO2023051335A1 (zh) 数据编码方法、数据解码方法以及数据处理装置
Karkishchenko et al. On the metric on images invariant with respect to the monotonic brightness transformation
CN113727050B (zh) 面向移动设备的视频超分辨率处理方法、装置、存储介质

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: 18931574

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 11/06/2021)

122 Ep: pct application non-entry in european phase

Ref document number: 18931574

Country of ref document: EP

Kind code of ref document: A1