WO2017161892A1 - 一种高光谱遥感图像的分类方法及其系统 - Google Patents

一种高光谱遥感图像的分类方法及其系统 Download PDF

Info

Publication number
WO2017161892A1
WO2017161892A1 PCT/CN2016/104656 CN2016104656W WO2017161892A1 WO 2017161892 A1 WO2017161892 A1 WO 2017161892A1 CN 2016104656 W CN2016104656 W CN 2016104656W WO 2017161892 A1 WO2017161892 A1 WO 2017161892A1
Authority
WO
WIPO (PCT)
Prior art keywords
remote sensing
hyperspectral remote
dimensional gabor
sensing image
quadrant
Prior art date
Application number
PCT/CN2016/104656
Other languages
English (en)
French (fr)
Inventor
贾森
沈琳琳
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
Publication of WO2017161892A1 publication Critical patent/WO2017161892A1/zh

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
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24147—Distances to closest patterns, e.g. nearest neighbour classification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/22—Matching criteria, e.g. proximity measures
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
    • G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
    • G06V10/449—Biologically inspired filters, e.g. difference of Gaussians [DoG] or Gabor filters

Definitions

  • the present invention relates to the field of image processing, and in particular, to a method and system for classifying hyperspectral remote sensing images.
  • the large difference between the hyperspectral dimension of hyperspectral data and the limited training samples is an important challenge for the classification of hyperspectral remote sensing images. Due to the interference of noise and the ubiquity of the phenomenon of "homogeneous foreign matter" (that is, the spectral similarity of different features), the traditional classification method based on spectral feature difference between features is difficult to obtain satisfactory accuracy.
  • feature extraction and band selection techniques are used to reduce the spectral dimension of hyperspectral data, alleviating the "Hughes phenomenon" (ie, given a fixed number of training samples whose prediction power decreases as the dimension increases) The problem of reduced classification accuracy of hyperspectral images.
  • the reduction of the dimension also causes the loss of effective information, which inevitably loses the classification accuracy.
  • the object of the present invention is to provide a method for classifying hyperspectral remote sensing images and a system thereof, aiming at solving the problem of low classification accuracy in the prior art.
  • the invention provides a classification method for hyperspectral remote sensing images, and the classification method comprises:
  • the hyperspectral remote sensing image is classified by a regularized Hamming distance using the encoded features.
  • the number of the plurality of three-dimensional Gabor filters is four.
  • the step of generating a plurality of three-dimensional Gabor filters specifically includes:
  • the frequency and direction of the candidate filter are set, and then a plurality of three-dimensional Gabor filters are generated according to a preset formula.
  • the step of performing image-capable coding on the three-dimensional Gabor phase feature of each pixel specifically includes:
  • the real part is 0/1 encoded according to the position of the quadrant in the three-dimensional Gabor phase, as in the first/fourth quadrant, the code is 1, otherwise the code is 0;
  • the imaginary part is 0/1 encoded according to the position of the quadrant in the three-dimensional Gabor phase, as in the first/second quadrant, the code is 1, otherwise the code is 0.
  • the step of using the coded feature to classify the hyperspectral remote sensing image by a regularized Hamming distance comprises:
  • the test sample t is divided into the p-th class, wherein the test sample t is a hyperspectral Remote Sensing Image.
  • the present invention also provides a classification system for hyperspectral remote sensing images, the system comprising:
  • An encoding module configured to perform a convolution operation on the hyperspectral remote sensing image with the generated plurality of three-dimensional Gabor filters to obtain a three-dimensional Gabor phase feature, and perform quadrant encoding on the three-dimensional Gabor phase feature of each pixel;
  • a classification module for classifying the hyperspectral remote sensing image by a regularized Hamming distance using the encoded features.
  • the number of the plurality of three-dimensional Gabor filters is four.
  • the generating module comprises:
  • a submodule configured to set a frequency and a direction of the candidate filter, and then generate a plurality of three-dimensional Gabor filters according to a preset formula.
  • the encoding module performs 0/1 encoding according to the position of the quadrant of the real part in the three-dimensional Gabor phase for each pixel, as in the first/fourth quadrant, the encoding is 1, otherwise the encoding is 0;
  • the position of the quadrant in the three-dimensional Gabor phase is 0/1 encoded, as in the first/second quadrant, the code is 1, otherwise the code is 0.
  • the classification module comprises:
  • a calculation submodule for calculating a similarity between each test sample t and any training sample s in the training set A;
  • test sample t is a hyperspectral remote sensing image.
  • the technical solution provided by the invention is based on the three-dimensional Gabor phase feature coding, and selects the feature subset with the most discriminative ability among a large number of three-dimensional Gabor phase features, which not only improves the classification accuracy, but also reduces the time and space complexity of the algorithm.
  • FIG. 1 is a flow chart of a method for classifying hyperspectral remote sensing images according to an embodiment of the present invention
  • step S11 shown in FIG. 1 according to an embodiment of the present invention
  • FIG. 3 is a schematic diagram of a frequency domain relationship of a three-dimensional Gabor feature according to an embodiment of the present invention.
  • FIG. 4 is a schematic diagram of an encoding strategy for each pixel according to an embodiment of the present invention.
  • FIG. 5 is a detailed flowchart of step S13 shown in FIG. 1 according to an embodiment of the present invention.
  • FIG. 6 is a flow chart of a specific example shown in FIG. 1 according to an embodiment of the present invention.
  • FIG. 7 is a schematic diagram showing the internal structure of a classification system 10 for hyperspectral remote sensing images according to an embodiment of the present invention.
  • FIG. 8 is a schematic structural diagram of the generating module 11 shown in FIG. 7 according to an embodiment of the present invention.
  • FIG. 9 is a schematic structural diagram of the classification module 13 shown in FIG. 7 according to an embodiment of the present invention.
  • a specific embodiment of the present invention provides a method for classifying hyperspectral remote sensing images, and the method mainly includes the following steps:
  • the classification method of hyperspectral remote sensing image provided by the invention is based on three-dimensional Gabor phase feature coding, and selects the feature subset with the most discriminative ability among a large number of three-dimensional Gabor phase features, which not only improves the classification accuracy, but also reduces the algorithm. Time and space complexity.
  • FIG. 1 is a flowchart of a method for classifying hyperspectral remote sensing images according to an embodiment of the present invention.
  • step S11 a plurality of three-dimensional Gabor filters parallel to the spectral direction are generated.
  • the number of the plurality of three-dimensional Gabor filters is four.
  • the number of the plurality of three-dimensional Gabor filters may also be adjusted according to actual requirements, for example, multiple three dimensions.
  • the number of Gabor filters can be designed as 2, 3, 5, 6, 7 and so on, which are not limited herein.
  • the step S11 of generating a plurality of three-dimensional Gabor filters specifically includes S111. -S112, as shown in Figure 2.
  • FIG. 2 is a detailed flowchart of step S11 shown in FIG. 1 according to an embodiment of the present invention.
  • step S111 a Gabor filter parallel to the spectral direction is selected as the candidate filter.
  • the hyperspectral image is hyperspectral image data obtained by imaging the sensor in different wavelength bands in the visible, near-infrared, mid-infrared and thermal infrared bands of the electromagnetic spectrum. Therefore, hyperspectral remote sensing images contain a wealth of spatial, radiation and spectral triple information, which provides a possibility for the fine classification and identification of surface materials.
  • step S112 the frequency and direction of the candidate filter are set, and then a plurality of three-dimensional Gabor filters are generated according to a preset formula.
  • step S12 the hyperspectral remote sensing image is convoluted with the generated three-dimensional Gabor filter to obtain a three-dimensional Gabor phase feature, and the three-dimensional Gabor phase characteristic of each pixel is obtained. Perform quadrant coding.
  • the hyperspectral remote sensing image is convoluted with the generated four three-dimensional Gabor filters, and the hyperspectral remote sensing image is represented by R, and G i represents the i-th three-dimensional Gabor feature, and the volume is taken.
  • the product operation is expressed as:
  • the three-dimensional Gabor phase feature of each pixel is subjected to quadrant encoding.
  • the encoding of each pixel has a real part coding and an imaginary part coding, according to the quadrant of the real part in the three-dimensional Gabor phase of the pixel.
  • the position is 0/1 encoded, as in the first/fourth quadrant, the code is 1, otherwise the code is 0; according to the position of the quadrant of the imaginary part of the three-dimensional Gabor phase of the pixel, 0/1 encoding, as in The first/second quadrant is encoded as 1, otherwise the code is 0, that is, the encoding of the result after the convolution operation is as follows:
  • Re(G i (x, y, b)) and Im(G i (x, y, b)) are the real and imaginary parts of the Gabor feature, respectively.
  • FIG. 4 is a schematic diagram of an encoding strategy for each pixel according to an embodiment of the present invention.
  • step S13 the hyperspectral remote sensing image is classified by the regularized Hamming distance using the encoded features.
  • the step S13 of classifying the hyperspectral remote sensing image by using the normalized Hamming distance specifically includes S131-S132, as shown in FIG. 5.
  • FIG. 5 is a detailed flowchart of step S13 shown in FIG. 1 according to an embodiment of the present invention.
  • step S131 the similarity between each test sample t and any training sample s in the training set A is calculated.
  • test sample t is a hyperspectral remote sensing image.
  • test sample t (assuming its spatial position coordinates are (x tt , y tt )) and training set A, any training sample s in t and A (assuming its spatial position coordinates are (x tr , y tr ))
  • the similarity measure between the two is as follows:
  • step S132 according to the nearest neighbor criterion, if the Hamming distance between the test sample t and one of the training samples in the p-th training set is the smallest, the test sample t is divided into the p-th class, wherein the test Sample t is a hyperspectral remote sensing image.
  • FIG. 6 is a flowchart of a specific example shown in FIG. 1 according to an embodiment of the present invention.
  • a classification system 10 for hyperspectral remote sensing images which mainly includes:
  • Generating module 11 for generating a plurality of three-dimensional Gabor filters parallel to the spectral direction;
  • the encoding module 12 is configured to perform a convolution operation on the hyperspectral remote sensing image with the generated three-dimensional Gabor filter to obtain a three-dimensional Gabor phase feature, and perform quadrant encoding on the three-dimensional Gabor phase feature of each pixel; as well as
  • the classification module 13 is configured to classify the hyperspectral remote sensing image by a regularized Hamming distance using the encoded features.
  • the classification system 10 for hyperspectral remote sensing images provided by the invention is based on three-dimensional Gabor phase feature coding, and selects the feature subset with the most discriminative ability among a large number of three-dimensional Gabor phase features, which not only improves the classification accuracy but also reduces The time and space complexity of the algorithm.
  • the classification system 10 of the hyperspectral remote sensing image mainly includes a generating module 11, an encoding module 12, and a classification module 13.
  • the generating module 11 is configured to generate a plurality of three-dimensional Gabor filters parallel to the spectral direction.
  • the number of the plurality of three-dimensional Gabor filters is four.
  • the number of the plurality of three-dimensional Gabor filters may also be adjusted according to actual requirements, for example, multiple three dimensions.
  • the number of Gabor filters can be designed as 2, 3, 5, 6, 7 and so on, which are not limited herein.
  • the generating module 11 specifically includes a selecting submodule 111 and a setting submodule 112, as shown in FIG. 8.
  • FIG. 8 shows the structure of the generation module 11 shown in FIG. 7 according to an embodiment of the present invention. intention.
  • a sub-module 111 is selected for selecting a Gabor filter parallel to the spectral direction as a candidate filter.
  • the selection method of the candidate filter is described in detail in the foregoing related step S111, and the description is not repeated here.
  • the setting sub-module 112 is configured to set a frequency and a direction of the candidate filter, and then generate a plurality of three-dimensional Gabor filters according to a preset formula.
  • the setting method of the frequency and the direction is as described in the related description in the foregoing step S112, and the description is not repeated here.
  • the encoding module 12 is configured to convolute the hyperspectral remote sensing image with the generated three-dimensional Gabor filter to obtain a three-dimensional Gabor phase feature and a three-dimensional Gabor phase for each pixel.
  • the feature is subjected to quadrant coding.
  • FIG. 9 a schematic structural diagram of the classification module 13 shown in FIG. 7 according to an embodiment of the present invention is shown.
  • the calculation sub-module 131 is configured to calculate the similarity between each test sample t and any training sample s in the training set A.
  • the categorization sub-module 132 is configured to divide the test sample t into the p-th class according to the nearest neighbor criterion, if the Hamming distance between the test sample t and one of the training samples in the p-th training set is the smallest, wherein The test sample t is a hyperspectral remote sensing image.
  • each unit included is only performed according to functional logic.
  • the divisions are not limited to the above-mentioned divisions, as long as the corresponding functions can be implemented; in addition, the specific names of the respective functional units are only for the purpose of facilitating mutual differentiation, and are not intended to limit the scope of protection of the present invention.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Multimedia (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Computation (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Biomedical Technology (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Image Analysis (AREA)

Abstract

一种高光谱遥感图像的分类方法和系统，其方法包括：生成多个平行于光谱方向的三维Gabor滤波器（S11）；将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码（S12）；以及使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类（S13）。该方法是基于三维Gabor相位特征编码，在大量的三维Gabor相位特征中选择出最具有鉴别能力的特征子集，不仅提升了分类精度，而且降低了算法的时间和空间复杂度。

Description

一种高光谱遥感图像的分类方法及其系统 技术领域
本发明涉及图像处理领域，尤其涉及一种高光谱遥感图像的分类方法及其系统。
背景技术
目前，高光谱数据的高光谱维度与有限的训练样本之间的巨大差异是高光谱遥感图像分类问题的重要挑战。由于噪声的干扰以及“同谱异物”现象(即不同地物的光谱特征具有较高的相似性)的普遍存在，传统的基于地物间光谱特征差异的分类方法难以获得令人满意的精度。同时，特征提取及波段选择技术被用于降低高光谱数据的光谱维度，减轻了由于“Hughes现象”(即给定固定数量的训练样本，其预测能力随着维度的增加而减小)引起的高光谱图像分类精度降低的问题。然而，维数降低的同时也造成了有效信息的丢失，不可避免的损失了分类精度。
发明内容
有鉴于此，本发明的目的在于提供一种高光谱遥感图像的分类方法及其系统，旨在解决现有技术中分类精度不高的问题。
本发明提出一种高光谱遥感图像的分类方法，所述分类方法包括：
生成多个平行于光谱方向的三维Gabor滤波器；
将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征；对每一个像素的三维Gabor相位特征进行象限位编码；以及
使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
优选的，所述多个三维Gabor滤波器的数量为4个。
优选的，所述生成多个三维Gabor滤波器的步骤具体包括：
选择平行于光谱方向的Gabor滤波器作为候选滤波器；
设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
优选的，所述对每一个像素的三维Gabor相位特征进行象限位编码的步骤具体包括：
针对每一个像素，根据三维Gabor相位中实部所在象限的位置对实部进行0/1编码，如在第一/第四象限，则编码为1，否则编码为0；
针对每一个像素，根据三维Gabor相位中虚部所在象限的位置对虚部进行0/1编码，如在第一/第二象限，则编码为1，否则编码为0。
优选的，所述使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类的步骤具体包括：
计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度；
根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中，所述测试样本t为高光谱遥感图像。
另一方面，本发明还提供一种高光谱遥感图像的分类系统，所述系统包括：
生成模块，用于生成多个平行于光谱方向的三维Gabor滤波器；
编码模块，用于将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码；以及
分类模块，用于使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
优选的，所述多个三维Gabor滤波器的数量为4个。
优选的，所述生成模块包括：
选择子模块，用于选择平行于光谱方向的Gabor滤波器作为候选滤波器；
设置子模块，用于设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
优选的，所述编码模块具体针对每一个像素，根据三维Gabor相位中实部所在象限的位置进行0/1编码，如在第一/第四象限，则编码为1，否则编码为0；根据三维Gabor相位中虚部所在象限的位置进行0/1编码，如在第一/第二象限，则编码为1，否则编码为0。
优选的，所述分类模块包括：
计算子模块，用于计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度；
归类子模块，用于根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中，所述测试样本t为高光谱遥感图像。
本发明提供的技术方案是基于三维Gabor相位特征编码，在大量的三维Gabor相位特征中选择出最具有鉴别能力的特征子集，不仅提升了分类精度，而且降低了算法的时间和空间复杂度。
附图说明
图1为本发明一实施方式中高光谱遥感图像的分类方法流程图；
图2为本发明一实施方式中图1所示的步骤S11的详细流程图；
图3为本发明一实施方式中三维Gabor特征的频率域关系的示意图；
图4为本发明一实施方式中针对每一个像素的编码策略示意图；
图5为本发明一实施方式中图1所示的步骤S13的详细流程图；
图6为本发明一实施方式中图1所示的具体实例流程图；
图7为本发明一实施方式中高光谱遥感图像的分类系统10的内部结构示意图；
图8为本发明一实施方式中图7所示的生成模块11的结构示意图；
图9为本发明一实施方式中图7所示的分类模块13的结构示意图。
具体实施方式
为了使本发明的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本发明进行进一步详细说明。应当理解，此处所描述的具体实施例仅仅用以解释本发明，并不用于限定本发明。
本发明具体实施方式提供了一种高光谱遥感图像的分类方法，所述方法主要包括如下步骤：
S11、生成多个平行于光谱方向的三维Gabor滤波器；
S12、将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征；然后对每一个像素的三维Gabor相位特征进行象限位编码；以及
S13、使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
本发明提供的一种高光谱遥感图像的分类方法是基于三维Gabor相位特征编码，在大量的三维Gabor相位特征中选择出最具有鉴别能力的特征子集，不仅提升了分类精度，而且降低了算法的时间和空间复杂度。
以下将对本发明所提供的一种高光谱遥感图像的分类方法进行详细说明。
请参阅图1，为本发明一实施方式中高光谱遥感图像的分类方法流程图。
在步骤S11中，生成多个平行于光谱方向的三维Gabor滤波器。
在本实施方式中，所述多个三维Gabor滤波器的数量为4个，当然，在其它实施方式中，所述多个三维Gabor滤波器的数量还可以根据实际需求进行调整，例如多个三维Gabor滤波器的数量可以设计为2个、3个、5个、6个、7个等等，在此不做限定。
在本实施方式中，生成多个三维Gabor滤波器的步骤S11具体包括S111 -S112，如图2所示。
请参阅图2，为本发明一实施方式中图1所示的步骤S11的详细流程图。
在步骤S111中，选择平行于光谱方向的Gabor滤波器作为候选滤波器。
在本实施方式中，高光谱图像是由传感器在电磁波谱的可见光、近红外、中红外和热红外波段范围内，在不同波段成像获得的高光谱图像数据。因此，高光谱遥感图像包含了丰富的空间、辐射和光谱三重信息，为地表物质的精细分类和识别提供了可能。
在步骤S112中，设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
在本实施方式中，以fj表示频率，
Figure PCTCN2016104656-appb-000001
表示方向，则设置所述候选滤波器的频率fj＝[0.5，0.25，0.125，0.0625]，设置方向
Figure PCTCN2016104656-appb-000002
θ＝π/2，然后根据预设公式
Figure PCTCN2016104656-appb-000003
生成4个三维Gabor滤波器，用于后续的三维Gabor特征提取；其中，
Figure PCTCN2016104656-appb-000004
Figure PCTCN2016104656-appb-000005
是滤波器与ω轴的夹角，θ是滤波器与μ-ν平面的夹角；(x，y，b)分别表示像素的x坐标、y坐标、光谱坐标；σ是高斯包络的宽度；由于方向参数
Figure PCTCN2016104656-appb-000006
θ＝π/2，根据图3可以看出，频率f所指的方向即为光谱方向。
在本实施方式中，由于Gabor滤波器的方向平行于光谱方向，同时仅选择了4个频率，最终得到4个三维Gabor滤波器，用{Ψi，i＝1，...，4}表示。
请继续参阅图1，在步骤S12中，将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码。
在本实施方式中，将高光谱遥感图像与所生成的4个三维Gabor滤波器进行卷积运算操作，以R表示所述高光谱遥感图像，Gi表示第i个三维Gabor特征，则取卷积运算表示为：
Figure PCTCN2016104656-appb-000007
4个三维Gabor特征表示 为：{Gi，i＝1，...，4}。
在本实施方式中，对每一个像素的三维Gabor相位特征进行象限位编码，具体地，每一个像素的编码有实部编码和虚部编码组成，根据该像素的三维Gabor相位中实部所在象限的位置进行0/1编码，如在第一/第四象限，则编码为1，否则编码为0；根据该像素的三维Gabor相位中虚部所在象限的位置进行0/1编码，如在第一/第二象限，则编码为1，否则编码为0，即，对卷积操作后的结果进行如下操作的编码：
Figure PCTCN2016104656-appb-000008
Figure PCTCN2016104656-appb-000009
Re(Gi(x，y，b))和Im(Gi(x，y，b))分别是Gabor特征的实部和虚部。
图4为本发明一实施方式中针对每一个像素的编码策略示意图。
在步骤S13中，使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
在本实施方式中，所述使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类的步骤S13具体包括S131-S132，如图5所示。
请参阅图5，为本发明一实施方式中图1所示的步骤S13的详细流程图。
在步骤S131中，计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度。
在本实施方式中，所述测试样本t为高光谱遥感图像。针对每一个测试样本t(假设其空间位置坐标为(xtt，ytt))和训练集合A，t与A中任意训练样本s(假设其空间位置坐标为(xtr，ytr))之间的相似性度量如下：
Figure PCTCN2016104656-appb-000010
其中，
Figure PCTCN2016104656-appb-000011
是异或运算，B是原始高光谱数据的光谱维度。显然，D0的取值在0和1之间。对于最优匹配，汉明距离为零。
在步骤S132中，根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中，所述测试样本t为高光谱遥感图像。
请参阅图6，为本发明一实施方式中图1所示的具体实例流程图。
如图7所示，本发明具体实施方式还提供一种高光谱遥感图像的分类系统10，主要包括：
生成模块11，用于生成多个平行于光谱方向的三维Gabor滤波器；
编码模块12，用于将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码；以及
分类模块13，用于使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
本发明提供的一种高光谱遥感图像的分类系统10，是基于三维Gabor相位特征编码，在大量的三维Gabor相位特征中选择出最具有鉴别能力的特征子集，不仅提升了分类精度，而且降低了算法的时间和空间复杂度。
请参阅图7，所示为本发明一实施方式中高光谱遥感图像的分类系统10的结构示意图。在本实施方式中，高光谱遥感图像的分类系统10主要包括生成模块11、编码模块12以及分类模块13。
生成模块11，用于生成多个平行于光谱方向的三维Gabor滤波器。
在本实施方式中，所述多个三维Gabor滤波器的数量为4个，当然，在其它实施方式中，所述多个三维Gabor滤波器的数量还可以根据实际需求进行调整，例如多个三维Gabor滤波器的数量可以设计为2个、3个、5个、6个、7个等等，在此不做限定。
在本实施方式中，生成模块11具体包括选择子模块111以及设置子模块112，如图8所示。
请参阅图8，所示为本发明一实施方式中图7所示的生成模块11的结构示 意图。
选择子模块111，用于选择平行于光谱方向的Gabor滤波器作为候选滤波器。
在本实施例中，候选滤波器的选择方法详见前述步骤S111中的相关记载，在此不做重复描述。
设置子模块112，用于设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
在本实施例中，频率和方向的设置方法详见前述步骤S112中的相关记载，在此不做重复描述。
请继续参阅图7，编码模块12，用于将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码。
在本实施例中，编码模块12具体的编码方法详见前述步骤S12中的相关记载，在此不做重复描述。
请参阅图9，所示为本发明一实施方式中图7所示的分类模块13的结构示意图。
计算子模块131，用于计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度。
在本实施例中，具体的计算方法详见前述步骤S131中的相关记载，在此不做重复描述。
归类子模块132，用于根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中，所述测试样本t为高光谱遥感图像。
在本实施例中，具体的归类方法详见前述步骤S132中的相关记载，在此不做重复描述。
值得注意的是，上述实施例中，所包括的各个单元只是按照功能逻辑进行 划分的，但并不局限于上述的划分，只要能够实现相应的功能即可；另外，各功能单元的具体名称也只是为了便于相互区分，并不用于限制本发明的保护范围。
另外，本领域普通技术人员可以理解实现上述各实施例方法中的全部或部分步骤是可以通过程序来指令相关的硬件来完成，相应的程序可以存储于一计算机可读取存储介质中，所述的存储介质，如ROM/RAM、磁盘或光盘等。
以上所述仅为本发明的较佳实施例而已，并不用以限制本发明，凡在本发明的精神和原则之内所作的任何修改、等同替换和改进等，均应包含在本发明的保护范围之内。

Claims (10)

  1. 一种高光谱遥感图像的分类方法，其特征在于，所述分类方法包括：
    生成多个平行于光谱方向的三维Gabor滤波器；
    将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征；
    对每一个像素的三维Gabor相位特征进行象限位编码；以及
    使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
  2. 如权利要求1所述的高光谱遥感图像的分类方法，其特征在于，所述多个三维Gabor滤波器的数量为4个。
  3. 如权利要求1所述的高光谱遥感图像的分类方法，其特征在于，所述生成多个三维Gabor滤波器的步骤具体包括：
    选择平行于光谱方向的Gabor滤波器作为候选滤波器；
    设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
  4. 如权利要求1所述的高光谱遥感图像的分类方法，其特征在于，所述对每一个像素的三维Gabor相位特征进行象限位编码的步骤具体包括：
    针对每一个像素，根据三维Gabor相位中实部所在象限的位置对实部进行0/1编码，如在第一/第四象限，则编码为1，否则编码为0；
    针对每一个像素，根据三维Gabor相位中虚部所在象限的位置对虚部进行0/1编码，如在第一/第二象限，则编码为1，否则编码为0。
  5. 如权利要求1所述的高光谱遥感图像的分类方法，其特征在于，所述使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类的步骤具体包括：
    计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度；
    根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中，所述测试样本t 为高光谱遥感图像。
  6. 一种高光谱遥感图像的分类系统，其特征在于，所述系统包括：
    生成模块，用于生成多个平行于光谱方向的三维Gabor滤波器；
    编码模块，用于将高光谱遥感图像与所生成的所述多个三维Gabor滤波器进行卷积运算，以得到三维Gabor相位特征，并对每一个像素的三维Gabor相位特征进行象限位编码；以及
    分类模块，用于使用编码的特征通过正则化的汉明距离对所述高光谱遥感图像进行分类。
  7. 如权利要求6所述的高光谱遥感图像的分类系统，其特征在于，所述多个三维Gabor滤波器的数量为4个。
  8. 如权利要求6所述的高光谱遥感图像的分类系统，其特征在于，所述生成模块包括：
    选择子模块，用于选择平行于光谱方向的Gabor滤波器作为候选滤波器；
    设置子模块，用于设置所述候选滤波器的频率和方向，然后根据预设公式生成多个三维Gabor滤波器。
  9. 如权利要求6所述的高光谱遥感图像的分类系统，其特征在于，所述编码模块具体针对每一个像素，根据三维Gabor相位中实部所在象限的位置对实部进行0/1编码，如在第一/第四象限，则编码为1，否则编码为0；根据三维Gabor相位中虚部所在象限的位置对虚部进行0/1编码，如在第一/第二象限，则编码为1，否则编码为0。
  10. 如权利要求6所述的高光谱遥感图像的分类系统，其特征在于，所述分类模块包括：
    计算子模块，用于计算每一个测试样本t和训练集合A中任意训练样本s之间的相似度；
    归类子模块，用于根据最近邻准则，如果测试样本t与第p类训练集中的某一个训练样本之间的汉明距离最小，则将该测试样本t划分为第p类，其中， 所述测试样本t为高光谱遥感图像。
PCT/CN2016/104656 2016-03-23 2016-11-04 一种高光谱遥感图像的分类方法及其系统 WO2017161892A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201610169168.4A CN105654098B (zh) 2016-03-23 2016-03-23 一种高光谱遥感图像的分类方法及其系统
CN201610169168.4 2016-03-23

Publications (1)

Publication Number Publication Date
WO2017161892A1 true WO2017161892A1 (zh) 2017-09-28

Family

ID=56495442

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2016/104656 WO2017161892A1 (zh) 2016-03-23 2016-11-04 一种高光谱遥感图像的分类方法及其系统

Country Status (2)

Country Link
CN (1) CN105654098B (zh)
WO (1) WO2017161892A1 (zh)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109460712A (zh) * 2018-10-12 2019-03-12 西北工业大学 一种基于最优近邻重构的高光谱波段选择方法
CN110222773A (zh) * 2019-06-10 2019-09-10 西北工业大学 基于不对称分解卷积网络的高光谱图像小样本分类方法
CN110636307A (zh) * 2019-08-26 2019-12-31 南京理工大学 基于Gabor滤波器的确定性图像编码方法
CN111401467A (zh) * 2020-03-26 2020-07-10 遥相科技发展(北京)有限公司 遥感影像分类方法、图像处理设备和计算机可读存储装置
CN111428627A (zh) * 2020-03-23 2020-07-17 西北大学 一种山地地貌遥感提取方法及系统
CN111460966A (zh) * 2020-03-27 2020-07-28 中国地质大学（武汉） 基于度量学习和近邻增强的高光谱遥感图像分类方法
CN111783884A (zh) * 2020-06-30 2020-10-16 山东女子学院 基于深度学习的无监督高光谱图像分类方法
CN113111863A (zh) * 2021-05-13 2021-07-13 辽宁工程技术大学 基于微型批处理与级联融合的图卷积网络高光谱分类方法

Families Citing this family (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN105654098B (zh) * 2016-03-23 2019-12-27 深圳大学 一种高光谱遥感图像的分类方法及其系统
CN106469316B (zh) * 2016-09-07 2020-02-21 深圳大学 基于超像素级信息融合的高光谱图像的分类方法及系统
WO2018081929A1 (zh) * 2016-11-01 2018-05-11 深圳大学 一种高光谱遥感图像特征提取和分类方法及其系统
CN107066959B (zh) * 2017-03-29 2020-12-18 广东交通职业技术学院 一种基于导向滤波和线性空间相关性信息的高光谱图像分类方法
CN107527001B (zh) * 2017-03-29 2020-12-22 广东交通职业技术学院 一种基于导向滤波和线性空间相关性信息的高光谱图像分类方法
CN109711269B (zh) * 2018-12-03 2023-06-20 辽宁工程技术大学 一种基于3d卷积谱空特征融合的高光谱图像分类算法
WO2022109945A1 (zh) * 2020-11-26 2022-06-02 深圳大学 基于尺度自适应滤波的高光谱和LiDAR联合分类方法

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104200201A (zh) * 2014-08-28 2014-12-10 武汉虹识技术有限公司 一种应用于虹膜识别的编码方法及系统
CN104680173A (zh) * 2015-01-26 2015-06-03 河海大学 一种遥感图像场景分类方法
CN105046205A (zh) * 2015-06-24 2015-11-11 西安理工大学 一种基于局部和全局特征融合的掌纹识别方法
CN105654098A (zh) * 2016-03-23 2016-06-08 深圳大学 一种高光谱遥感图像的分类方法及其系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN104200201A (zh) * 2014-08-28 2014-12-10 武汉虹识技术有限公司 一种应用于虹膜识别的编码方法及系统
CN104680173A (zh) * 2015-01-26 2015-06-03 河海大学 一种遥感图像场景分类方法
CN105046205A (zh) * 2015-06-24 2015-11-11 西安理工大学 一种基于局部和全局特征融合的掌纹识别方法
CN105654098A (zh) * 2016-03-23 2016-06-08 深圳大学 一种高光谱遥感图像的分类方法及其系统

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
SHEN, LINLIN: "Coding 3D Gabor Features for Hyperspectral Palmprint Recognition", 2014 INTERNATIONAL CONFERENCE ON MEDICAL BIOMETRICS, 30 June 2014 (2014-06-30) *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109460712B (zh) * 2018-10-12 2024-01-12 西北工业大学 一种基于最优近邻重构的高光谱波段选择方法
CN109460712A (zh) * 2018-10-12 2019-03-12 西北工业大学 一种基于最优近邻重构的高光谱波段选择方法
CN110222773A (zh) * 2019-06-10 2019-09-10 西北工业大学 基于不对称分解卷积网络的高光谱图像小样本分类方法
CN110222773B (zh) * 2019-06-10 2023-03-24 西北工业大学 基于不对称分解卷积网络的高光谱图像小样本分类方法
CN110636307B (zh) * 2019-08-26 2022-03-18 南京理工大学 基于Gabor滤波器的确定性图像编码方法
CN110636307A (zh) * 2019-08-26 2019-12-31 南京理工大学 基于Gabor滤波器的确定性图像编码方法
CN111428627A (zh) * 2020-03-23 2020-07-17 西北大学 一种山地地貌遥感提取方法及系统
CN111428627B (zh) * 2020-03-23 2023-03-24 西北大学 一种山地地貌遥感提取方法及系统
CN111401467B (zh) * 2020-03-26 2023-12-19 苏州征图智能科技有限公司 遥感影像分类方法、图像处理设备和计算机可读存储装置
CN111401467A (zh) * 2020-03-26 2020-07-10 遥相科技发展(北京)有限公司 遥感影像分类方法、图像处理设备和计算机可读存储装置
CN111460966A (zh) * 2020-03-27 2020-07-28 中国地质大学（武汉） 基于度量学习和近邻增强的高光谱遥感图像分类方法
CN111460966B (zh) * 2020-03-27 2024-02-02 中国地质大学（武汉） 基于度量学习和近邻增强的高光谱遥感图像分类方法
CN111783884A (zh) * 2020-06-30 2020-10-16 山东女子学院 基于深度学习的无监督高光谱图像分类方法
CN111783884B (zh) * 2020-06-30 2024-04-09 山东女子学院 基于深度学习的无监督高光谱图像分类方法
CN113111863A (zh) * 2021-05-13 2021-07-13 辽宁工程技术大学 基于微型批处理与级联融合的图卷积网络高光谱分类方法
CN113111863B (zh) * 2021-05-13 2023-11-24 辽宁工程技术大学 基于微型批处理与级联融合的图卷积网络高光谱分类方法

Also Published As

Publication number Publication date
CN105654098A (zh) 2016-06-08
CN105654098B (zh) 2019-12-27

Similar Documents

Publication Publication Date Title
WO2017161892A1 (zh) 一种高光谱遥感图像的分类方法及其系统
US9542593B2 (en) Image-based feature detection using edge vectors
Ehret et al. Image anomalies: A review and synthesis of detection methods
Davarzani et al. Scale-and rotation-invariant texture description with improved local binary pattern features
Pflug et al. A comparative study on texture and surface descriptors for ear biometrics
Dong et al. Target recognition in SAR images via classification on Riemannian manifolds
Foytik et al. A two-layer framework for piecewise linear manifold-based head pose estimation
Bursuc et al. Kernel local descriptors with implicit rotation matching
Shi et al. Hyperspectral target detection with macro-micro feature extracted by 3-D residual autoencoder
Li et al. Image matching based on two-column histogram hashing and improved RANSAC
US20120328197A1 (en) Identifying matching images
Wang et al. Fake modern Chinese painting identification based on spectral–spatial feature fusion on hyperspectral image
Roy et al. A novel quaternary pattern of local maximum quotient for heterogeneous face recognition
Yu et al. Robust point cloud normal estimation via neighborhood reconstruction
Vizilter et al. Shape-based image matching using heat kernels and diffusion maps
Asokan et al. Deep Feature Extraction and Feature Fusion for Bi-Temporal Satellite Image Classification.
García et al. Efficient semantic segmentation of hyperspectral images using adaptable rectangular convolution
Shabanifard et al. Forensic detection of image manipulation using the Zernike moments and pixel‐pair histogram
Yu et al. Affine invariant fusion feature extraction based on geometry descriptor and BIT for object recognition
Zhang et al. Feature-band-based unsupervised hyperspectral underwater target detection near the coastline
Eid et al. Development of Iris Security System Using Adaptive Quality-Based Template Fusion
WO2017166156A1 (zh) 一种高光谱图像的分类方法及其系统
Davarzani et al. Robust image description with weighted and adaptive local binary pattern features
Song et al. Face anti-spoofing detection using least square weight fusion of channel-based feature classifiers
Mohammadi et al. Using nonlocal filtering and feature extraction approaches in three-dimensional face recognition by Kinect

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

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

Ref document number: 16895238

Country of ref document: EP

Kind code of ref document: A1

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

Free format text: NOTING OF LOSS OF RIGHTS (EPO FORM 1205A DATED 04.02.2019)

122 Ep: pct application non-entry in european phase

Ref document number: 16895238

Country of ref document: EP

Kind code of ref document: A1