技术领域technical field
本发明属于图像处理技术领域，更进一步涉及图像分类技术领域中的一种基于融合多尺度多维空谱特征的高光谱图像分类方法。本发明通过对高光谱图像中地物的种类分析，可应用于灾害监测、地质勘探、城市规划、目标识别等诸多领域。The invention belongs to the technical field of image processing, and further relates to a hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral features in the technical field of image classification. The present invention can be applied to many fields such as disaster monitoring, geological exploration, urban planning, target recognition, etc. by analyzing the types of ground objects in hyperspectral images.
背景技术Background technique
高光谱以其丰富的波段信息记录了地物目标的连续光谱特征，具备了能够进行更多种类地物目标识别和更高精度地进行目标分类的可能性。高光谱图像分类技术的关键在于利用小数量的训练样本获得较高的分类精度。近期，随着深度学习在各个领域的广泛应用，高光谱分类也出现了多种深度学习分类方法，如自编码器AE(Autoencoder)、卷积神经网络CNN(Convolutional Neural Networks)、深度信念网络DBN(Deep Belief Network)等。这些方法中，CNN的分类性能最好，但是，始终达不到人们的预期。因为高光谱图像存在严重的“同谱异物，同物异谱”的现象。Hyperspectral records the continuous spectral characteristics of ground objects with its rich band information, and has the possibility of recognizing more types of ground objects and classifying objects with higher accuracy. The key to hyperspectral image classification technology is to use a small number of training samples to obtain higher classification accuracy. Recently, with the wide application of deep learning in various fields, hyperspectral classification has also emerged a variety of deep learning classification methods, such as autoencoder AE (Autoencoder), convolutional neural network CNN (Convolutional Neural Networks), deep belief network DBN (Deep Belief Network), etc. Among these methods, the classification performance of CNN is the best, but it still falls short of people's expectations. Because there is a serious phenomenon of "same spectrum and different spectrum, same spectrum and different spectrum" in hyperspectral images.
西安电子科技大学在其申请的专利文献“基于空间坐标与空谱特征融合的高光谱分类方法”(专利申请号：201710644479.6，申请公布号：CN 107451614A)提出了一种基于空间坐标与空谱特征融合的高光谱图像分类方法。该方法先对高光谱图像进行空间邻域划分采样，随后将空间坐标作为空间特征，接着将空间特征与光谱特征分别利用支持向量机SVM进行分类，将分类所得像素点属于每类的概率作为概率特征，最后将空间特征分类得到的概率特征与光谱特征所得概率特征融合，再次利用支持向量机SVM(Support VectorMachine)进行分类，得出最终分类结果。该方法存在的不足之处在于，仅利用了空间坐标作为空间特征，高光谱图像的空间信息利用不足，空谱特征融合不充分的问题。而且，使用一维特征向量作为输入，没有充分利用尺度特征，空间坐标和一维特征向量对于样本分布不集中或样本量很少的地物类别分类效果不好。In the patent document "Hyperspectral classification method based on fusion of spatial coordinates and spatial spectral features" applied by Xidian University (patent application number: 201710644479.6, application publication number: CN 107451614A), a method based on spatial coordinates and spatial spectral features is proposed. A fusion approach for hyperspectral image classification. In this method, the hyperspectral image is first divided into spatial neighborhoods and sampled, then the spatial coordinates are used as spatial features, and then the spatial features and spectral features are classified using the support vector machine SVM, and the probability that the classified pixels belong to each class is used as the probability Finally, the probability features obtained by spatial feature classification and the probability features obtained by spectral features are fused, and SVM (Support Vector Machine) is used to classify again to obtain the final classification result. The shortcomings of this method are that only spatial coordinates are used as spatial features, the spatial information of hyperspectral images is insufficiently utilized, and the fusion of spatial spectral features is insufficient. Moreover, the use of one-dimensional feature vectors as input does not make full use of scale features. Spatial coordinates and one-dimensional feature vectors are not effective for the classification of ground object categories where the sample distribution is not concentrated or the sample size is small.
Zilong Zhong等人在其发表的论文“Spectral-Spatial Residual Network forHyperspectral Image Classification:A 3-D Deep Learning Framework”(IEEETransactions on Geoscience and Remote Sensing,2017:1-12)中提出一种利用端到端的光谱-空间残差网络(SSRN)对高光谱图像进行分类的方法。该方法以原始三维立方体作为输入数据，且无需进行特征工程。在端到端的光谱-空间残差网络中，光谱和空间残差块连续地从高光谱图像中丰富的光谱特征和空间背景中学习识别特征，三维卷积神经网络得到的光谱特征和二维卷积神经网络得到的空间特征以级联的方式融合，最后将融合的特征输入分类层进行高光谱图像分类。该方法存在的不足之处在于，单一尺度对尺度特征的利用不够充分，仅提取了单一尺度的邻域块儿作为输入。而单一尺度特征在整体分类精度上表现不佳。Zilong Zhong et al. proposed an end-to-end spectral - Spatial Residual Network (SSRN) method for classifying hyperspectral images. The method takes raw 3D cubes as input data and does not require feature engineering. In the end-to-end spectral-spatial residual network, the spectral and spatial residual blocks continuously learn recognition features from the rich spectral features and spatial background in hyperspectral images, spectral features obtained by 3D convolutional neural networks and 2D volumetric The spatial features obtained by the product neural network are fused in a cascaded manner, and finally the fused features are input into the classification layer for hyperspectral image classification. The disadvantage of this method is that the use of scale features at a single scale is not sufficient, and only single-scale neighborhood blocks are extracted as input. While single-scale features perform poorly in overall classification accuracy.
发明内容Contents of the invention
本发明的目的在于针对上述现有技术的不足，提出一种基于融合多尺度多维空谱特征的高光谱图像分类方法，融合不同尺度特征之间的相关信息，同时实现高维特征与低维特征、空间特征与光谱特征融合，用于解决现有高光谱图像分类方法中分类精度不高，对于样本分布不集中或样本量很少的地物类别分类效果不好的问题。The purpose of the present invention is to address the deficiencies of the above-mentioned prior art, and propose a hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial-spectral features, which fuses related information between features of different scales, and realizes high-dimensional features and low-dimensional features at the same time , The fusion of spatial features and spectral features is used to solve the problem of low classification accuracy in existing hyperspectral image classification methods, and poor classification results for ground object categories with non-concentrated sample distribution or small sample size.
实现本发明目的的思路是，先构建三个结构相同的特征提取分支和一个组合分类器，然后生成多尺度空谱特征和多维特征融合网络，将多尺度训练样本输入多尺度空谱特征和多维特征融合网络，提取多尺度多维空谱联合特征并进行分类，利用损失函数对网络进行训练，最后将测试样本输入到训练好的多尺度空谱特征和多维特征融合网络中，对高光谱图像进行分类。The idea of realizing the purpose of the present invention is to first construct three feature extraction branches with the same structure and a combined classifier, then generate multi-scale spatial spectral features and multi-dimensional feature fusion network, and input multi-scale training samples into multi-scale spatial spectral features and multi-dimensional features. The feature fusion network extracts multi-scale and multi-dimensional space-spectrum joint features and classifies them, uses the loss function to train the network, and finally inputs the test samples into the trained multi-scale space-spectrum feature and multi-dimensional feature fusion network to perform hyperspectral image Classification.
为实现上述目的，本发明的具体步骤包括如下：To achieve the above object, the concrete steps of the present invention include as follows:
(1)输入高光谱图像：(1) Input hyperspectral image:
输入一幅高光谱图像，该高光谱图像是一个三维特征立方体高光谱图像中每个波段对应特征立方体中的一个二维矩阵其中，∈表示属于符号，表示实数域符号，m表示高光谱图像的长，n表示高光谱图像的宽，b表示高光谱图像的光谱波段数，i表示高光谱图像中光谱波段的序号，i＝1,2,…,b；Input a hyperspectral image, which is a 3D feature cube Each band in the hyperspectral image corresponds to a 2D matrix in the feature cube Among them, ∈ means belonging to the symbol, Represents the real number domain symbol, m represents the length of the hyperspectral image, n represents the width of the hyperspectral image, b represents the number of spectral bands in the hyperspectral image, i represents the serial number of the spectral band in the hyperspectral image, i=1,2,..., b;
(2)对待分类的高光谱图像进行预处理：(2) Preprocessing the hyperspectral image to be classified:
(2a)将m×n×b的三维高光谱图像矩阵转换成a×b二维的特征矩阵，a＝m×n，其中，二维特征矩阵中的每列表示光谱维度，每行表示每个样本的所有光谱信息；(2a) Convert the m×n×b three-dimensional hyperspectral image matrix into a×b two-dimensional feature matrix, a=m×n, where each column in the two-dimensional feature matrix represents the spectral dimension, and each row represents each All spectral information of a sample;
(2b)采用归一化公式，对二维特征矩阵进行归一化处理；(2b) using a normalization formula to normalize the two-dimensional feature matrix;
(2c)将归一化后的二维特征矩阵转换成与原始高光谱图像尺寸大小相同的归一化后三维特征矩阵；(2c) converting the normalized two-dimensional feature matrix into a normalized three-dimensional feature matrix having the same size as the original hyperspectral image;
(3)邻域取块：(3) Neighborhood blocks:
(3a)对归一化后的三维特征矩阵进行0像素边缘填充操作，边缘填充0像素的尺寸分别为3、5、7；(3a) Perform 0-pixel edge filling operation on the normalized three-dimensional feature matrix, and the size of the 0-pixel edge filling is 3, 5, and 7 respectively;
(3b)在填充处理后的高光谱图像中，以每个像素点为中心，分别选取7×7、11×11、15×15的邻域块，得到三种尺度的邻域块；(3b) In the hyperspectral image after the filling process, each pixel point is the center, and the neighborhood blocks of 7×7, 11×11, and 15×15 are respectively selected to obtain neighborhood blocks of three scales;
(4)生成训练集与测试集：(4) Generate training set and test set:
(4a)分别将三种尺度的邻域块按其中心像素点类别分配到该类别所属的集合中；(4a) respectively assign the neighborhood blocks of three scales to the set to which the category belongs according to their central pixel category;
(4b)分别将每类集合中的邻域块中已知类别标签的邻域块作为训练集，并将每个邻域块的中心像素点标签作为该邻域块的标签，分别将每类集合中剩余的邻域块作为测试集；(4b) The neighborhood blocks with known category labels in the neighborhood blocks in each class set are used as the training set, and the center pixel label of each neighborhood block is used as the label of the neighborhood block, respectively. The remaining neighborhood blocks in the set are used as the test set;
(5)构建多尺度空谱特征和多维特征融合网络：(5) Construct multi-scale spatial spectral features and multi-dimensional feature fusion network:
(5a)分别搭建三个结构相同的特征提取分支，每个分支的结构依次为：第一三维卷积层→第一规范层→第一激活函数层→第二三维卷积层→第二规范层→第二激活函数层→第三三维卷积层→第三规范层→第三激活函数层→第一二维卷积层→第四规范层→第四激活函数层→第一融合层→第二二维卷积层→第五规范层→第五激活函数层→第二融合层→第三二维卷积层→第六规范层→第六激活函数层→第一最大池化层→第四二维卷积层→第七规范层→第七激活函数层→第二最大池化层；所述第一融合层是将第一激活函数层和第四激活函数层经由加法操作相融合；所述第二融合层是将第二激活函数层和第五激活函数层经由加法操作相融合；(5a) Build three feature extraction branches with the same structure, and the structure of each branch is as follows: first three-dimensional convolution layer → first norm layer → first activation function layer → second three-dimensional convolution layer → second norm Layer → second activation function layer → third three-dimensional convolutional layer → third normalization layer → third activation function layer → first two-dimensional convolutional layer → fourth normalization layer → fourth activation function layer → first fusion layer → The second two-dimensional convolutional layer → the fifth normative layer → the fifth activation function layer → the second fusion layer → the third two-dimensional convolutional layer → the sixth normative layer → the sixth activation function layer → the first maximum pooling layer → The fourth two-dimensional convolutional layer → the seventh normative layer → the seventh activation function layer → the second maximum pooling layer; the first fusion layer is to fuse the first activation function layer and the fourth activation function layer via an addition operation ; The second fusion layer is to fuse the second activation function layer and the fifth activation function layer via an addition operation;
(5b)将三个结构相同的特征提取分支经由concatenate层合并融合后依次与全局平均池化层、输出层相连，组成多尺度空谱特征和多维特征融合网络；(5b) The three feature extraction branches with the same structure are merged and fused through the concatenate layer, and then connected to the global average pooling layer and the output layer in turn to form a multi-scale spatial spectral feature and multi-dimensional feature fusion network;
(5c)设置多尺度空谱特征和多维特征融合网络的参数如下：将第一、第二、第三三维卷积层的神经元个数均设置为24，卷积核尺寸依次设置为(1，1，20)、(1，1，3)、(1，1，10)，卷积步长依次设置为20、1、1；将第一、第二、第三、第四二维卷积层的神经元个数依次设置为240、240、24、24，卷积核长度均设置为3，卷积步长均设置为1；将每个最大池化层的池化长度均设置为3；将每个规范层的动量因子均设置为0.8；将全局平均池化层的池化长度设置为7；将每个激活函数层的激活函数均设置为ReLU激活函数；将输出层神经元的个数设置为类别数，激活函数选用softmax函数；(5c) Set the parameters of the multi-scale spatial-spectral feature and multi-dimensional feature fusion network as follows: set the number of neurons in the first, second, and third three-dimensional convolutional layers to 24, and set the convolution kernel size to (1 , 1, 20), (1, 1, 3), (1, 1, 10), the convolution step size is set to 20, 1, 1 in turn; the first, second, third, and fourth two-dimensional convolution The number of neurons in the product layer is set to 240, 240, 24, 24 in turn, the length of the convolution kernel is set to 3, and the convolution step is set to 1; the pooling length of each maximum pooling layer is set to 3; set the momentum factor of each specification layer to 0.8; set the pooling length of the global average pooling layer to 7; set the activation function of each activation function layer to the ReLU activation function; set the output layer neuron The number of is set to the number of categories, and the activation function uses the softmax function;
(6)训练多尺度空谱特征和多维特征融合网络：(6) Training multi-scale spatial spectral features and multi-dimensional feature fusion network:
将训练集与训练集的标签输入到多尺度空谱特征和多维特征融合网络中进行训练，得到训练好的多尺度空谱特征和多维特征融合网络；Input the training set and the label of the training set into the multi-scale spatial spectral feature and multi-dimensional feature fusion network for training, and obtain the trained multi-scale spatial spectral feature and multi-dimensional feature fusion network;
(7)对测试样本进行分类：(7) Classify the test samples:
将测试集输入到训练好的多尺度空谱特征和多维特征融合网络中，输出测试样本的类别标签，将多尺度空谱特征和多维特征融合网络的输出作为测试样本的预测标签，完成高光谱图像分类。Input the test set into the trained multi-scale spatial spectral feature and multi-dimensional feature fusion network, output the category label of the test sample, and use the output of the multi-scale spatial spectral feature and multi-dimensional feature fusion network as the predicted label of the test sample to complete the hyperspectral Image classification.
本发明与现有技术相比具有以下优点：Compared with the prior art, the present invention has the following advantages:
第一，本发明搭建三个结构相同的特征提取分支，用于提取高光谱图像中不同尺度的尺度特征，克服了现有技术中单一尺度对尺度特征利用不够充分，分类精度不高的缺点，使得本发明具有提取多尺度特征的优点，提高了对高光谱图像中地物的分类精度。First, the present invention builds three feature extraction branches with the same structure for extracting scale features of different scales in hyperspectral images, which overcomes the disadvantages of insufficient utilization of scale features and low classification accuracy in the prior art for a single scale. The present invention has the advantage of extracting multi-scale features, and improves the classification accuracy of ground objects in hyperspectral images.
第二，本发明将第一激活函数层和第四激活函数层经由加法操作相融合，将第二激活函数层和第五激活函数层经由加法操作相融合，用于将三维卷积神经网络3D-CNN(3DConvolutional Neural Networks)提取到的光谱特征和二维卷积神经网络2D-CNN(2DConvolutional Neural Networks)提取到的空间特征相融合，克服了现有技术空谱特征融合不充分的，对于样本分布不集中或样本量很少的地物类别分类效果不好的不足，使得本发明具有充分融合光谱特征和空间特征、高维特征和低维特征的优点，提高了对小数量样本类别的识别能力。Second, the present invention fuses the first activation function layer and the fourth activation function layer via an addition operation, and fuses the second activation function layer and the fifth activation function layer via an addition operation to integrate the three-dimensional convolutional neural network into a 3D -The fusion of spectral features extracted by CNN (3DConvolutional Neural Networks) and spatial features extracted by two-dimensional convolutional neural network 2D-CNN (2DConvolutional Neural Networks) overcomes the insufficient fusion of spatial spectral features in the existing technology. Due to the disadvantages of poor classification effect of ground object categories with non-concentrated distribution or small sample size, the present invention has the advantages of fully integrating spectral features and spatial features, high-dimensional features and low-dimensional features, and improves the recognition of small number of sample categories ability.
附图说明Description of drawings
图1是本发明的实现流程图；Fig. 1 is the realization flowchart of the present invention;
图2是本发明多尺度空谱特征与多维特征融合网络的结构图。Fig. 2 is a structural diagram of the multi-scale spatial-spectral feature and multi-dimensional feature fusion network of the present invention.
具体实施方式Detailed ways
下面结合附图对本发明作进一步的详细描述。The present invention will be described in further detail below in conjunction with the accompanying drawings.
参照图1，对本发明的具体实施作进一步的详细描述。Referring to Fig. 1, the specific implementation of the present invention will be further described in detail.
步骤1，输入高光谱图像。Step 1, input hyperspectral image.
输入一幅高光谱图像，该高光谱图像是一个三维特征立方体高光谱图像中每个波段对应特征立方体中的一个二维矩阵其中，∈表示属于符号，表示实数域符号，m表示高光谱图像的长，n表示高光谱图像的宽，b表示高光谱图像的光谱波段数，i表示高光谱图像中光谱波段的序号，i＝1,2,…,b。Input a hyperspectral image, which is a 3D feature cube Each band in the hyperspectral image corresponds to a 2D matrix in the feature cube Among them, ∈ means belonging to the symbol, Represents the real number domain symbol, m represents the length of the hyperspectral image, n represents the width of the hyperspectral image, b represents the number of spectral bands in the hyperspectral image, i represents the serial number of the spectral band in the hyperspectral image, i=1,2,..., b.
步骤2，对待分类的高光谱图像进行预处理。Step 2, preprocessing the hyperspectral image to be classified.
将m×n×b的三维高光谱图像矩阵转换成a×b二维的特征矩阵，a＝m×n，其中，二维特征矩阵中的每列表示光谱维度，每行表示每个样本的所有光谱信息。Convert the m×n×b three-dimensional hyperspectral image matrix into a×b two-dimensional feature matrix, a=m×n, where each column in the two-dimensional feature matrix represents the spectral dimension, and each row represents the All spectral information.
采用归一化公式，对二维特征矩阵进行归一化处理。The normalization formula is used to normalize the two-dimensional feature matrix.
所述的归一化公式如下：The normalization formula is as follows:
其中，表示归一化后的二维特征矩阵中第i个光谱波段中的第j个地物目标，表示二维特征矩阵中第i个光谱波段中所有地物目标的平均像素值，表示二维特征矩阵中第i个光谱波段中所有地物目标像素值的方差值。in, Indicates the j-th object in the i-th spectral band in the normalized two-dimensional feature matrix, Indicates the average pixel value of all ground objects in the i-th spectral band in the two-dimensional feature matrix, Indicates the variance value of all surface object pixel values in the i-th spectral band in the two-dimensional feature matrix.
将归一化后的二维特征矩阵转换成与原始高光谱图像尺寸大小相同的归一化后三维特征矩阵。Convert the normalized two-dimensional feature matrix into a normalized three-dimensional feature matrix with the same size as the original hyperspectral image.
步骤3，邻域取块。Step 3, take blocks in the neighborhood.
对归一化后的三维特征矩阵进行0像素边缘填充操作，边缘填充0像素的尺寸分别为3、5、7。Perform 0-pixel edge padding operation on the normalized three-dimensional feature matrix, and the sizes of the 0-pixel edge padding are 3, 5, and 7, respectively.
在填充处理后的高光谱图像中，以每个像素点为中心，分别选取7×7、11×11、15×15的邻域块，得到三种尺度的邻域块。In the filled hyperspectral image, each pixel point is the center, and the neighborhood blocks of 7×7, 11×11, and 15×15 are respectively selected to obtain neighborhood blocks of three scales.
步骤4，生成训练集与测试集。Step 4, generate training set and test set.
分别将三种尺度的邻域块按其中心像素点类别分配到该类别所属的集合中。The neighborhood blocks of the three scales are assigned to the set to which the category belongs according to their central pixel category.
分别将每类集合中的邻域块中已知类别标签的邻域块作为训练集，并将每个邻域块的中心像素点标签作为该邻域块的标签，分别将每类集合中剩余的邻域块作为测试集。The neighborhood blocks with known category labels in the neighborhood blocks in each type of set are used as the training set, and the center pixel label of each neighborhood block is used as the label of the neighborhood block, and the remaining The neighborhood blocks of are used as the test set.
步骤5，构建多尺度空谱特征和多维特征融合网络。Step 5, build multi-scale spatial spectral features and multi-dimensional feature fusion network.
参照图2，对本发明多尺度空谱特征和多维特征融合网络的结构作进一步的详细描述。Referring to Fig. 2, the structure of the multi-scale spatial-spectral feature and multi-dimensional feature fusion network of the present invention will be further described in detail.
分别搭建三个结构相同的特征提取分支，每个分支的结构依次为：第一三维卷积层→第一规范层→第一激活函数层→第二三维卷积层→第二规范层→第二激活函数层→第三三维卷积层→第三规范层→第三激活函数层→第一二维卷积层→第四规范层→第四激活函数层→第一融合层→第二二维卷积层→第五规范层→第五激活函数层→第二融合层→第三二维卷积层→第六规范层→第六激活函数层→第一最大池化层→第四二维卷积层→第七规范层→第七激活函数层→第二最大池化层；所述第一融合层是将第一激活函数层和第四激活函数层经由加法操作相融合；所述第二融合层是将第二激活函数层和第五激活函数层经由加法操作相融合。Build three feature extraction branches with the same structure, and the structure of each branch is as follows: first three-dimensional convolutional layer → first normalization layer → first activation function layer → second three-dimensional convolutional layer → second normalization layer → second Two activation function layer → third three-dimensional convolutional layer → third normative layer → third activation function layer → first two-dimensional convolutional layer → fourth normative layer → fourth activation function layer → first fusion layer → second two Dimensional convolutional layer→fifth normative layer→fifth activation function layer→second fusion layer→third two-dimensional convolutional layer→sixth normative layer→sixth activation function layer→first max pooling layer→fourth two Dimensional convolution layer→seventh standard layer→seventh activation function layer→second maximum pooling layer; the first fusion layer is to fuse the first activation function layer and the fourth activation function layer via an addition operation; the The second fusion layer is to fuse the second activation function layer and the fifth activation function layer through an addition operation.
将三个结构相同的特征提取分支经由concatenate层合并融合后依次与全局平均池化层、输出层相连，组成多尺度空谱特征和多维特征融合网络。The three feature extraction branches with the same structure are merged and fused through the concatenate layer, and then connected to the global average pooling layer and the output layer in turn to form a multi-scale spatial spectral feature and multi-dimensional feature fusion network.
设置多尺度空谱特征和多维特征融合网络的参数如下：将第一、第二、第三三维卷积层的神经元个数均设置为24，卷积核尺寸依次设置为(1，1，20)、(1，1，3)、(1，1，10)，卷积步长依次设置为20、1、1；将第一、第二、第三、第四二维卷积层的神经元个数依次设置为240、240、24、24，卷积核长度均设置为3，卷积步长均设置为1；将每个最大池化层的池化长度均设置为3；将每个规范层的动量因子均设置为0.8；将全局平均池化层的池化长度设置为7；将每个激活函数层的激活函数均设置为ReLU激活函数；将输出层神经元的个数设置为类别数，激活函数选用softmax函数。Set the parameters of the multi-scale spatial-spectral feature and multi-dimensional feature fusion network as follows: set the number of neurons in the first, second, and third three-dimensional convolutional layers to 24, and set the convolution kernel size to (1, 1, 20), (1, 1, 3), (1, 1, 10), the convolution step size is set to 20, 1, 1 in sequence; the first, second, third, and fourth two-dimensional convolutional layers The number of neurons is set to 240, 240, 24, 24 in turn, the length of the convolution kernel is set to 3, and the convolution step is set to 1; the pooling length of each maximum pooling layer is set to 3; The momentum factor of each specification layer is set to 0.8; the pooling length of the global average pooling layer is set to 7; the activation function of each activation function layer is set to the ReLU activation function; the number of neurons in the output layer is set to Set to the number of categories, and the activation function uses the softmax function.
步骤6，训练多尺度空谱特征和多维特征融合网络。Step 6, train multi-scale spatial spectral features and multi-dimensional feature fusion network.
将训练集与训练集的标签输入到多尺度空谱特征和多维特征融合网络中进行训练，得到训练好的多尺度空谱特征和多维特征融合网络。The training set and the labels of the training set are input into the multi-scale spatial spectral feature and multi-dimensional feature fusion network for training, and the trained multi-scale spatial spectral feature and multi-dimensional feature fusion network are obtained.
所述的将训练集与训练集的标签输入到多尺度空谱特征和多维特征融合网络中进行训练的具体操作步骤如下：The specific operation steps of inputting the training set and the labels of the training set into the multi-scale spatial spectral feature and multi-dimensional feature fusion network for training are as follows:
第1步，将训练集分别输入到多尺度空谱特征和多维特征融合网络的三个不同邻域块的特征提取分支中，输出训练样本的预测标签向量。In the first step, the training set is input into the feature extraction branch of three different neighborhood blocks of the multi-scale spatial spectral feature and multi-dimensional feature fusion network, and the predicted label vector of the training sample is output.
第2步，采用梯度下降法，用多尺度空谱特征和多维特征融合网络的损失函数优化网络参数，直到网络参数收敛为止，所述损失函数的学习率设置为0.0001。In the second step, the gradient descent method is used to optimize the network parameters with the loss function of the multi-scale spatial spectral feature and the multi-dimensional feature fusion network until the network parameters converge, and the learning rate of the loss function is set to 0.0001.
第3步，利用下述的交叉熵公式，计算预测标签向量与真实标签向量之间的交叉熵。Step 3, use the following cross-entropy formula to calculate the cross-entropy between the predicted label vector and the real label vector.
其中，L表示预测标签向量与真实标签向量之间的交叉熵，Σ表示求和操作，yi表示预测标签向量中的第i个元素，ln表示以自然常数e为底的对数操作，表示预测标签向量中的第m个元素。Among them, L represents the cross entropy between the predicted label vector and the real label vector, Σ represents the summation operation, yi represents the ith element in the predicted label vector, ln represents the logarithmic operation based on the natural constant e, Represents the mth element in the predicted label vector.
下面结合仿真实验对本发明的效果做进一步的说明：Effect of the present invention is described further below in conjunction with simulation experiment:
1.仿真实验条件：1. Simulation experiment conditions:
本发明的仿真实验采用的硬件测试平台是：处理器为Inter Core i7-8750H，主频为2.20GHz，内存16GB；软件平台为：Windows 10企业版64位操作系统和Python3.6进行仿真测试。The hardware testing platform that simulation experiment of the present invention adopts is: processor is Inter Core i7-8750H, main frequency is 2.20GHz, memory 16GB; Software platform is: Windows 10 Enterprise Edition 64-bit operating system and Python3.6 carry out simulation test.
本发明的仿真实验中所使用的高光谱图像数据集，是由AVIRIS sensor在印第安纳州西北部的Indian Pines测试场地上收集的Indian pines数据集以及ROSIS高光谱遥感卫星在意大利北部帕维亚大学拍摄获得的Pavia university数据集，其中，Indian pines数据集图像的大小为145*145，具有200个光谱波段，包含16类地物，每类地物的类别与数量如表1所示。The hyperspectral image data set used in the simulation experiment of the present invention is the Indian pines data set collected by AVIRIS sensor on the Indian Pines test site in northwest Indiana and taken by the ROSIS hyperspectral remote sensing satellite at the University of Pavia in northern Italy The obtained Pavia university dataset, in which the image size of the Indian pines dataset is 145*145, has 200 spectral bands, and contains 16 types of ground objects. The categories and quantities of each type of ground objects are shown in Table 1.
表1 Indian pines样本类别与数量Table 1 Types and numbers of Indian pines samples
Pavia university数据集图像大小为610*340，具有103个光谱波段，包含9类地物，每类地物的类别与数量如表2所示。The image size of the Pavia university dataset is 610*340, with 103 spectral bands, including 9 types of ground objects. The category and quantity of each type of ground objects are shown in Table 2.
表2 Pavia university样本类别与数量Table 2 Pavia university sample category and quantity
2.仿真实验内容及结果分析：2. Simulation experiment content and result analysis:
本发明采用多尺度空谱特征和多维特征融合网络进行分类，多尺度的思想是本发明解决单一尺度分类效果不理想这一问题的关键，因此，本发明的仿真实验1中使用本发明的多尺度空谱特征和多维特征融合网络对Indian pines、Pavia university两个高光谱图像数据进行分类，并将分类结果与三个单一尺度网络分类结果进行对比，说明本发明多尺度特征融合的分类效果要优于基于单一尺度的分类效果，以此来证明多尺度特征融合的有效性。The present invention uses multi-scale spatial spectral features and multi-dimensional feature fusion network for classification. The idea of multi-scale is the key to solve the problem of unsatisfactory single-scale classification effect. Therefore, in the simulation experiment 1 of the present invention, the multi-scale The fusion network of scale-space-spectral features and multi-dimensional features classifies the two hyperspectral image data of Indian pines and Pavia university, and compares the classification results with the classification results of three single-scale networks, which shows that the classification effect of multi-scale feature fusion in the present invention should be It is better than the classification effect based on a single scale to prove the effectiveness of multi-scale feature fusion.
为了说明本发明基于多尺度空谱特征和多维特征融合网络的高光谱图像分类方法具有优秀的分类能力，本发明的仿真实验2中用本发明的多尺度空谱特征和多维特征融合网络对Indian pines、Pavia university两个高光谱图像数据进行分类，并将分类结果与三种现有高光谱图像分类方法的分类结果进行对比。In order to illustrate that the hyperspectral image classification method based on the multi-scale spatial-spectral feature and multi-dimensional feature fusion network of the present invention has excellent classification ability, in the simulation experiment 2 of the present invention, the multi-scale spatial-spectral feature and multi-dimensional feature fusion network of the present invention are used for Indian Pines and Pavia university two hyperspectral image data are classified, and the classification results are compared with the classification results of three existing hyperspectral image classification methods.
仿真实验1：Simulation experiment 1:
使用本发明和基于单一尺度网络的高光谱图像分类方法分别对Indian pines、Pavia university高光谱图像数据进行分类。其中，三个单一尺度网络是本发明中多尺度空谱特征和多维特征融合网络的三个分支。本发明的仿真实验1是使用多尺度空谱特征和多维特征融合网络分别与三个单一尺度网络在分类结果上进行对比，证明多尺度特征融合的有效性。The hyperspectral image data of Indian pines and Pavia university are classified respectively by using the present invention and the hyperspectral image classification method based on a single-scale network. Among them, the three single-scale networks are the three branches of the multi-scale spatial-spectral feature and multi-dimensional feature fusion network in the present invention. In the simulation experiment 1 of the present invention, the multi-scale spatial spectral feature and multi-dimensional feature fusion network are compared with three single-scale networks on classification results to prove the effectiveness of multi-scale feature fusion.
为了对分类结果进行量化，实验一中采用了以下3个评价指标。In order to quantify the classification results, the following three evaluation indicators were used in Experiment 1.
(1)整体精度OA(overall accuracy)，将测试集上正确分类的像素点的个数除以总的像素个数，称为整体精度OA，其值在0～100％之间，此值越大说明分类效果越好。(1) Overall accuracy OA (overall accuracy), divide the number of correctly classified pixels on the test set by the total number of pixels, called the overall accuracy OA, and its value is between 0 and 100%. The larger the value, the better the classification effect.
(2)平均精度AA(average accuracy)，将测试集上每类正确分类的像素点个数除以该类所有像素的总数，得到该类的正确分类精度，将所有类别的精度的平均值称为平均精度AA，其值在0～100％之间，此值越大说明分类效果越好。(2) Average accuracy AA (average accuracy). Divide the number of correctly classified pixels of each class on the test set by the total number of all pixels of this class to obtain the correct classification accuracy of this class. The average of the accuracy of all classes is called It is the average accuracy AA, and its value is between 0 and 100%. The larger the value, the better the classification effect.
(3)Kappa(Kappa Coefficient)系数：Kappa系数是定义在混淆矩阵X上的一个评价指标，综合考虑混淆矩阵对角线上的元素和偏离对角线的元素，更客观地反映了算法的分类性能,Kappa的值在-1～1的范围，此值越大说明分类效果越好。(3) Kappa (Kappa Coefficient) coefficient: Kappa coefficient is an evaluation index defined on the confusion matrix X, which comprehensively considers the elements on the diagonal of the confusion matrix and the elements off the diagonal, which more objectively reflects the classification of the algorithm Performance, the value of Kappa is in the range of -1 to 1, the larger the value, the better the classification effect.
用上述评价指标对两种分类方法的分类性能进行评价，结果如表3。The classification performance of the two classification methods was evaluated with the above evaluation indicators, and the results are shown in Table 3.
表3三个单一尺度网络与本发明在分类精度上的对比结果Table 3 Comparison results of three single-scale networks and the present invention on classification accuracy
从表3中可以看出，在Indian pines数据的分类结果上，7*7的小邻域块在分类结果上，AA要明显大于OA，说明小尺度邻域块对地物类别数量较少的样本分类性能较好。而15*15的大尺度邻域块在分类结果上，OA明显大于AA，说明大尺度邻域块对地物类别数量较大的样本分类性能较好。而本发明融合多个尺度提取到的特征，结合了各个尺度的优点，并得到了更加优秀的分类结果。It can be seen from Table 3 that in the classification results of the Indian pines data, AA is significantly larger than OA in the classification results of the 7*7 small neighborhood blocks, indicating that the small-scale neighborhood blocks have a small number of feature categories. The sample classification performance is better. In terms of the classification results of the large-scale neighborhood blocks of 15*15, OA is significantly larger than AA, which shows that the large-scale neighborhood blocks have better classification performance for samples with a large number of object categories. However, the present invention fuses features extracted from multiple scales, combines the advantages of each scale, and obtains better classification results.
同时，在Pavia university数据的分类结果上，本发明提出的多尺度空谱特征和多维特征融合网络在三个分类精度指标上也明显优于三个单一尺度网络，说明本发明具有较强的泛化能力和鲁棒性。At the same time, in the classification results of Pavia university data, the multi-scale spatial spectral feature and multi-dimensional feature fusion network proposed by the present invention are also significantly better than the three single-scale networks in the three classification accuracy indicators, indicating that the present invention has a strong general capability and robustness.
仿真实验2：Simulation experiment 2:
为了验证本发明所提方法的有效性，相同条件下，将本发明与高光谱领域现有的三种分类方法在高光谱数据集Indian pines上的分类结果进行对比，结果如表4。In order to verify the effectiveness of the method proposed in the present invention, under the same conditions, the classification results of the present invention and the three existing classification methods in the hyperspectral field on the hyperspectral data set Indian pins were compared, and the results are shown in Table 4.
这三种现有方法分别是：The three existing methods are:
1)经典的支持向量机SVM用于高光谱图像分类的方法，该方法通过支持向量机直接对光谱信息进行分类。1) The classic support vector machine (SVM) is used for the hyperspectral image classification method, which directly classifies the spectral information through the support vector machine.
2)Zilong Zhong等人在其发表的论文“Spectral-Spatial Residual Networkfor Hyperspectral Image Classification:A 3-D Deep Learning Framework”(IEEETransactions on Geoscience and Remote Sensing,2017:1-12)中提出一种利用端到端的光谱-空间残差网络(SSRN)对高光谱图像进行分类的方法。该方法以原始三维立方体作为输入数据，且无需进行特征工程。在端到端的光谱-空间残差网络中，光谱和空间残差块连续地从高光谱图像中丰富的光谱特征和空间背景中学习识别特征，三维卷积神经网络得到的光谱特征和二维卷积神经网络得到的空间特征以级联的方式融合，最后将融合的特征输入分类层进行高光谱图像分类。2) In their paper "Spectral-Spatial Residual Network for Hyperspectral Image Classification: A 3-D Deep Learning Framework" (IEEE Transactions on Geoscience and Remote Sensing, 2017:1-12), Zilong Zhong et al. A method for classifying hyperspectral images using a spectral-spatial residual network (SSRN) on the end. The method takes raw 3D cubes as input data and does not require feature engineering. In the end-to-end spectral-spatial residual network, the spectral and spatial residual blocks continuously learn recognition features from the rich spectral features and spatial background in hyperspectral images, spectral features obtained by 3D convolutional neural networks and 2D volumetric The spatial features obtained by the product neural network are fused in a cascaded manner, and finally the fused features are input into the classification layer for hyperspectral image classification.
3)西安电子科技大学在其申请的专利文献“基于空间坐标与空谱特征融合的高光谱分类方法”(专利申请号：201710644479.6，申请公布号：CN 107451614 A)提出了一种基于空间坐标与空谱特征融合的高光谱图像分类方法SPE-SPA-SVM。3) The patent document "Hyperspectral Classification Method Based on Fusion of Spatial Coordinates and Spatial Spectral Features" applied by Xidian University (patent application number: 201710644479.6, application publication number: CN 107451614 A) proposes a method based on spatial coordinates and Hyperspectral Image Classification Method SPE-SPA-SVM with Spatial Spectral Feature Fusion.
本发明与三种现有技术在两个高光谱数据集上的整体分类精度OA、平均分类精度AA和Kappa系数的对比，如表4所示。The comparison of the overall classification accuracy OA, average classification accuracy AA and Kappa coefficient between the present invention and three existing technologies on two hyperspectral datasets is shown in Table 4.
表4现有技术与本发明在分类精度上的对比结果Table 4 prior art and the comparison result of the present invention on classification accuracy
从表3可以看出，在Indian pines数据集上，本发明的分类结果在关于分类精度的3个指标上均明显优于这三种现有技术。It can be seen from Table 3 that on the Indian pines data set, the classification results of the present invention are significantly better than the three existing technologies on the three indicators of classification accuracy.
本发明将三维卷积神经网络提取到的特征以自身映射的方式与二维卷积神经网络提取到特征相融合，实现了光谱特征和空间特征、低维特征和高维特征相融合，大大提升了网络的分类能力。同时，本发明提出的多尺度融合的思想，大大改善了单一尺度网络分类性能不佳的问题，所以本发明的分类性能要明显优于其他三个现有分类方法。The present invention fuses the features extracted by the three-dimensional convolutional neural network with the features extracted by the two-dimensional convolutional neural network in the form of self-mapping, realizes the fusion of spectral features and spatial features, low-dimensional features and high-dimensional features, and greatly improves classification ability of the network. At the same time, the idea of multi-scale fusion proposed by the present invention greatly improves the problem of poor classification performance of a single-scale network, so the classification performance of the present invention is significantly better than the other three existing classification methods.
综合上述仿真实验1和仿真实验2的结果分析，本发明提出的方法能有效解决卷积神经网络在训练时特征太过单一和尺度太过单一的问题，并且能够解决进行高光谱分类时平均分类精度AA低的问题。Based on the analysis of the results of the above-mentioned simulation experiment 1 and simulation experiment 2, the method proposed by the present invention can effectively solve the problem of too single feature and too single scale of the convolutional neural network during training, and can solve the problem of average classification during hyperspectral classification. The problem of low precision AA.
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616371.5ACN110321963B (en) | 2019-07-09 | 2019-07-09 | Hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral features |
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910616371.5ACN110321963B (en) | 2019-07-09 | 2019-07-09 | Hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral features |
| Publication Number | Publication Date |
|---|---|
| CN110321963Atrue CN110321963A (en) | 2019-10-11 |
| CN110321963B CN110321963B (en) | 2022-03-04 |
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910616371.5AActiveCN110321963B (en) | 2019-07-09 | 2019-07-09 | Hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral features |
| Country | Link |
|---|---|
| CN (1) | CN110321963B (en) |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991236A (en)* | 2019-10-29 | 2020-04-10 | 成都华为技术有限公司 | Image classification method and related device |
| CN111008664A (en)* | 2019-12-05 | 2020-04-14 | 上海海洋大学 | A hyperspectral sea ice detection method based on combined spatial and spectral features |
| CN111126256A (en)* | 2019-12-23 | 2020-05-08 | 武汉大学 | Hyperspectral image classification method based on self-adaptive space-spectrum multi-scale network |
| CN111160273A (en)* | 2019-12-31 | 2020-05-15 | 北京云智空间科技有限公司 | Hyperspectral image space spectrum combined classification method and device |
| CN111191736A (en)* | 2020-01-05 | 2020-05-22 | 西安电子科技大学 | Hyperspectral image classification method based on depth feature cross fusion |
| CN111368940A (en)* | 2020-03-25 | 2020-07-03 | 西北工业大学 | A Hyperspectral Image Classification Method Based on Guided Filtering and Kernel Extreme Learning Machine |
| CN111368896A (en)* | 2020-02-28 | 2020-07-03 | 南京信息工程大学 | A classification method of hyperspectral remote sensing images based on dense residual 3D convolutional neural network |
| CN111539447A (en)* | 2020-03-17 | 2020-08-14 | 广东省智能制造研究所 | Hyperspectrum and terahertz data depth fusion-based classification method |
| CN111563528A (en)* | 2020-03-31 | 2020-08-21 | 西北工业大学 | SAR image classification method based on multi-scale feature learning network and bilateral filtering |
| CN111639587A (en)* | 2020-05-27 | 2020-09-08 | 西安电子科技大学 | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network |
| CN111652038A (en)* | 2020-04-13 | 2020-09-11 | 上海海洋大学 | Remote sensing sea ice image classification method based on convolutional neural network |
| CN111695467A (en)* | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Spatial spectrum full convolution hyperspectral image classification method based on superpixel sample expansion |
| CN111723838A (en)* | 2020-04-21 | 2020-09-29 | 中国人民解放军火箭军工程大学 | A hyperspectral feature classification method based on multi-core machine learning and fusion of multi-scale features |
| CN111723731A (en)* | 2020-06-18 | 2020-09-29 | 西安电子科技大学 | Hyperspectral image classification method, storage medium and equipment based on space-spectral convolution kernel |
| CN112052758A (en)* | 2020-08-25 | 2020-12-08 | 西安电子科技大学 | A hyperspectral image classification method based on attention mechanism and recurrent neural network |
| CN112052755A (en)* | 2020-08-24 | 2020-12-08 | 西安电子科技大学 | Semantic convolutional hyperspectral image classification method based on multi-channel attention mechanism |
| CN112101271A (en)* | 2020-09-23 | 2020-12-18 | 台州学院 | Hyperspectral remote sensing image classification method and device |
| CN112215267A (en)* | 2020-09-25 | 2021-01-12 | 天津大学 | Hyperspectral image-oriented depth space spectrum subspace clustering method |
| CN112308152A (en)* | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection |
| CN112541892A (en)* | 2020-12-07 | 2021-03-23 | 山东众阳健康科技集团有限公司 | Cerebral aneurysm image segmentation method based on 3D activation function |
| CN112577744A (en)* | 2020-11-02 | 2021-03-30 | 西南交通大学 | Rolling bearing fault mode identification method based on combination of SPA-map and ResBet |
| CN112733659A (en)* | 2020-12-30 | 2021-04-30 | 华东师范大学 | Hyperspectral image classification method based on self-learning double-flow multi-scale dense connection network |
| CN112733736A (en)* | 2021-01-13 | 2021-04-30 | 西安电子科技大学 | Class imbalance hyperspectral image classification method based on enhanced oversampling |
| CN112733769A (en)* | 2021-01-18 | 2021-04-30 | 西安电子科技大学 | Hyperspectral image classification method based on multiband entropy rate superpixel segmentation |
| CN112766199A (en)* | 2021-01-26 | 2021-05-07 | 武汉大学 | Hyperspectral image classification method based on self-adaptive multi-scale feature extraction model |
| CN112949416A (en)* | 2021-02-04 | 2021-06-11 | 东华理工大学 | Supervised hyperspectral multi-scale graph volume integral classification method |
| CN113095409A (en)* | 2021-04-13 | 2021-07-09 | 西安电子科技大学 | Hyperspectral image classification method based on attention mechanism and weight sharing |
| CN113361485A (en)* | 2021-07-08 | 2021-09-07 | 齐齐哈尔大学 | Hyperspectral image classification method based on spectral space attention fusion and deformable convolution residual error network |
| CN113392931A (en)* | 2021-07-02 | 2021-09-14 | 西安电子科技大学 | Hyperspectral open set classification method based on self-supervision learning and multitask learning |
| CN113743429A (en)* | 2020-05-28 | 2021-12-03 | 中国人民解放军战略支援部队信息工程大学 | Hyperspectral image classification method and device |
| CN113837314A (en)* | 2021-10-09 | 2021-12-24 | 陕西科技大学 | A Hyperspectral Image Classification Method Based on Hybrid Convolutional Neural Networks |
| CN113920442A (en)* | 2021-09-29 | 2022-01-11 | 中国人民解放军火箭军工程大学 | Hyperspectral classification method combining graph structure and convolutional neural network |
| CN113963207A (en)* | 2021-10-21 | 2022-01-21 | 江南大学 | Hyperspectral image classification method based on spatial spectral information feature-guided fusion network |
| CN114022960A (en)* | 2022-01-05 | 2022-02-08 | 阿里巴巴达摩院(杭州)科技有限公司 | Model training and behavior recognition method, device, electronic device, and storage medium |
| CN114330591A (en)* | 2022-01-05 | 2022-04-12 | 安徽大学 | A hyperspectral classification method based on spatial-spectral dimension transformation |
| CN114494846A (en)* | 2021-12-18 | 2022-05-13 | 南京理工大学 | Hyperspectral image classification method based on multi-scale internal volume network |
| CN114694140A (en)* | 2022-03-22 | 2022-07-01 | 南京林业大学 | Directional training platycodon grandiflorum impurity identification method based on spectral imaging |
| CN114708457A (en)* | 2022-03-31 | 2022-07-05 | 南京林业大学 | A hyperspectral deep learning recognition method for anti-purple fringing recognition |
| CN114842264A (en)* | 2022-05-17 | 2022-08-02 | 北京理工大学 | Hyperspectral image classification method based on multi-scale spatial spectral feature joint learning |
| CN115240074A (en)* | 2022-09-22 | 2022-10-25 | 山东锋士信息技术有限公司 | Hyperspectral image classification method and device based on covariance expression |
| CN115937565A (en)* | 2022-02-14 | 2023-04-07 | 苏州大学 | Hyperspectral Image Classification Method Based on Adaptive L-BFGS Algorithm |
| CN116310572A (en)* | 2023-03-23 | 2023-06-23 | 齐齐哈尔大学 | Pyramid multi-scale convolution and self-attention combined hyperspectral image classification method |
| CN116580252A (en)* | 2023-06-16 | 2023-08-11 | 西安电子科技大学 | Hyperspectral Image Classification Method Based on Multiscale Dense Connection and Feature Aggregation Network |
| CN118196033A (en)* | 2024-03-18 | 2024-06-14 | 江苏商贸职业学院 | Non-contact high-precision measurement method and system based on machine vision fusion |
| CN118317110A (en)* | 2024-06-11 | 2024-07-09 | 国网安徽省电力有限公司超高压分公司 | A method for enhancing image transmission by joint coding of substation source and channel |
| CN118429800A (en)* | 2024-04-17 | 2024-08-02 | 湖南农业大学 | Hyperspectral image classification method based on tensor width learning network |
| CN119672440A (en)* | 2024-12-23 | 2025-03-21 | 西安电子科技大学 | Few-shot HSI classification method based on prototype mask comparison and cross-stage feature refinement |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105740773A (en)* | 2016-01-25 | 2016-07-06 | 重庆理工大学 | Deep learning and multi-scale information based behavior identification method |
| US20170076179A1 (en)* | 2015-09-15 | 2017-03-16 | Samsung Electronics Co., Ltd. | Learning combinations of homogenous feature arrangements |
| CN107316013A (en)* | 2017-06-14 | 2017-11-03 | 西安电子科技大学 | Hyperspectral image classification method with DCNN is converted based on NSCT |
| WO2018052586A1 (en)* | 2016-09-14 | 2018-03-22 | Konica Minolta Laboratory U.S.A., Inc. | Method and system for multi-scale cell image segmentation using multiple parallel convolutional neural networks |
| CN109726757A (en)* | 2018-12-26 | 2019-05-07 | 山东女子学院 | A DCNN hyperspectral image classification method fused with multi-scale spatial features |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20170076179A1 (en)* | 2015-09-15 | 2017-03-16 | Samsung Electronics Co., Ltd. | Learning combinations of homogenous feature arrangements |
| CN105740773A (en)* | 2016-01-25 | 2016-07-06 | 重庆理工大学 | Deep learning and multi-scale information based behavior identification method |
| WO2018052586A1 (en)* | 2016-09-14 | 2018-03-22 | Konica Minolta Laboratory U.S.A., Inc. | Method and system for multi-scale cell image segmentation using multiple parallel convolutional neural networks |
| CN107316013A (en)* | 2017-06-14 | 2017-11-03 | 西安电子科技大学 | Hyperspectral image classification method with DCNN is converted based on NSCT |
| CN109726757A (en)* | 2018-12-26 | 2019-05-07 | 山东女子学院 | A DCNN hyperspectral image classification method fused with multi-scale spatial features |
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110991236A (en)* | 2019-10-29 | 2020-04-10 | 成都华为技术有限公司 | Image classification method and related device |
| CN111008664A (en)* | 2019-12-05 | 2020-04-14 | 上海海洋大学 | A hyperspectral sea ice detection method based on combined spatial and spectral features |
| CN111008664B (en)* | 2019-12-05 | 2023-04-07 | 上海海洋大学 | Hyperspectral sea ice detection method based on space-spectrum combined characteristics |
| CN111126256B (en)* | 2019-12-23 | 2022-02-15 | 武汉大学 | Hyperspectral image classification method based on self-adaptive space-spectrum multi-scale network |
| CN111126256A (en)* | 2019-12-23 | 2020-05-08 | 武汉大学 | Hyperspectral image classification method based on self-adaptive space-spectrum multi-scale network |
| CN111160273A (en)* | 2019-12-31 | 2020-05-15 | 北京云智空间科技有限公司 | Hyperspectral image space spectrum combined classification method and device |
| CN111160273B (en)* | 2019-12-31 | 2023-05-09 | 北京云智空间科技有限公司 | Hyperspectral image spatial spectrum joint classification method and device |
| CN111191736A (en)* | 2020-01-05 | 2020-05-22 | 西安电子科技大学 | Hyperspectral image classification method based on depth feature cross fusion |
| CN111368896A (en)* | 2020-02-28 | 2020-07-03 | 南京信息工程大学 | A classification method of hyperspectral remote sensing images based on dense residual 3D convolutional neural network |
| CN111539447A (en)* | 2020-03-17 | 2020-08-14 | 广东省智能制造研究所 | Hyperspectrum and terahertz data depth fusion-based classification method |
| CN111539447B (en)* | 2020-03-17 | 2022-04-22 | 广东省科学院智能制造研究所 | A classification method based on deep fusion of hyperspectral and terahertz data |
| CN111368940A (en)* | 2020-03-25 | 2020-07-03 | 西北工业大学 | A Hyperspectral Image Classification Method Based on Guided Filtering and Kernel Extreme Learning Machine |
| CN111563528A (en)* | 2020-03-31 | 2020-08-21 | 西北工业大学 | SAR image classification method based on multi-scale feature learning network and bilateral filtering |
| CN111563528B (en)* | 2020-03-31 | 2022-03-11 | 西北工业大学 | SAR image classification method based on multi-scale feature learning network and bilateral filtering |
| CN111652038A (en)* | 2020-04-13 | 2020-09-11 | 上海海洋大学 | Remote sensing sea ice image classification method based on convolutional neural network |
| CN111652038B (en)* | 2020-04-13 | 2023-03-03 | 上海海洋大学 | Sea ice image classification method based on convolutional neural network remote sensing |
| CN111723838A (en)* | 2020-04-21 | 2020-09-29 | 中国人民解放军火箭军工程大学 | A hyperspectral feature classification method based on multi-core machine learning and fusion of multi-scale features |
| CN111639587A (en)* | 2020-05-27 | 2020-09-08 | 西安电子科技大学 | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network |
| CN111639587B (en)* | 2020-05-27 | 2023-03-24 | 西安电子科技大学 | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network |
| CN113743429A (en)* | 2020-05-28 | 2021-12-03 | 中国人民解放军战略支援部队信息工程大学 | Hyperspectral image classification method and device |
| CN111695467A (en)* | 2020-06-01 | 2020-09-22 | 西安电子科技大学 | Spatial spectrum full convolution hyperspectral image classification method based on superpixel sample expansion |
| CN111723731A (en)* | 2020-06-18 | 2020-09-29 | 西安电子科技大学 | Hyperspectral image classification method, storage medium and equipment based on space-spectral convolution kernel |
| CN111723731B (en)* | 2020-06-18 | 2023-09-29 | 西安电子科技大学 | Hyperspectral image classification method, storage medium and equipment based on spatial spectrum convolution kernel |
| CN112052755B (en)* | 2020-08-24 | 2023-06-02 | 西安电子科技大学 | Semantic Convolutional Hyperspectral Image Classification Method Based on Multi-way Attention Mechanism |
| CN112052755A (en)* | 2020-08-24 | 2020-12-08 | 西安电子科技大学 | Semantic convolutional hyperspectral image classification method based on multi-channel attention mechanism |
| CN112052758B (en)* | 2020-08-25 | 2023-05-23 | 西安电子科技大学 | Hyperspectral image classification method based on attention mechanism and cyclic neural network |
| CN112052758A (en)* | 2020-08-25 | 2020-12-08 | 西安电子科技大学 | A hyperspectral image classification method based on attention mechanism and recurrent neural network |
| CN112101271A (en)* | 2020-09-23 | 2020-12-18 | 台州学院 | Hyperspectral remote sensing image classification method and device |
| CN112215267B (en)* | 2020-09-25 | 2022-11-01 | 天津大学 | A deep spatial spectral subspace clustering method for hyperspectral images |
| CN112215267A (en)* | 2020-09-25 | 2021-01-12 | 天津大学 | Hyperspectral image-oriented depth space spectrum subspace clustering method |
| CN112577744A (en)* | 2020-11-02 | 2021-03-30 | 西南交通大学 | Rolling bearing fault mode identification method based on combination of SPA-map and ResBet |
| CN112577744B (en)* | 2020-11-02 | 2021-12-03 | 西南交通大学 | Rolling bearing fault identification method based on combination of SPA-map and ResNet |
| CN112308152A (en)* | 2020-11-03 | 2021-02-02 | 西安电子科技大学 | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection |
| CN112308152B (en)* | 2020-11-03 | 2023-04-28 | 西安电子科技大学 | Hyperspectral image ground object classification method based on spectrum segmentation and homogeneous region detection |
| CN112541892A (en)* | 2020-12-07 | 2021-03-23 | 山东众阳健康科技集团有限公司 | Cerebral aneurysm image segmentation method based on 3D activation function |
| CN112733659A (en)* | 2020-12-30 | 2021-04-30 | 华东师范大学 | Hyperspectral image classification method based on self-learning double-flow multi-scale dense connection network |
| CN112733659B (en)* | 2020-12-30 | 2022-09-20 | 华东师范大学 | A method for hyperspectral image classification based on self-learning dual-stream multi-scale densely connected network |
| CN112733736A (en)* | 2021-01-13 | 2021-04-30 | 西安电子科技大学 | Class imbalance hyperspectral image classification method based on enhanced oversampling |
| CN112733769B (en)* | 2021-01-18 | 2023-04-07 | 西安电子科技大学 | Hyperspectral image classification method based on multiband entropy rate superpixel segmentation |
| CN112733769A (en)* | 2021-01-18 | 2021-04-30 | 西安电子科技大学 | Hyperspectral image classification method based on multiband entropy rate superpixel segmentation |
| US11783569B2 (en) | 2021-01-26 | 2023-10-10 | Wuhan University | Method for classifying hyperspectral images on basis of adaptive multi-scale feature extraction model |
| CN112766199A (en)* | 2021-01-26 | 2021-05-07 | 武汉大学 | Hyperspectral image classification method based on self-adaptive multi-scale feature extraction model |
| WO2022160771A1 (en)* | 2021-01-26 | 2022-08-04 | 武汉大学 | Method for classifying hyperspectral images on basis of adaptive multi-scale feature extraction model |
| CN112766199B (en)* | 2021-01-26 | 2022-04-29 | 武汉大学 | Hyperspectral image classification method based on self-adaptive multi-scale feature extraction model |
| CN112949416A (en)* | 2021-02-04 | 2021-06-11 | 东华理工大学 | Supervised hyperspectral multi-scale graph volume integral classification method |
| CN113095409A (en)* | 2021-04-13 | 2021-07-09 | 西安电子科技大学 | Hyperspectral image classification method based on attention mechanism and weight sharing |
| CN113392931A (en)* | 2021-07-02 | 2021-09-14 | 西安电子科技大学 | Hyperspectral open set classification method based on self-supervision learning and multitask learning |
| CN113361485A (en)* | 2021-07-08 | 2021-09-07 | 齐齐哈尔大学 | Hyperspectral image classification method based on spectral space attention fusion and deformable convolution residual error network |
| CN113920442A (en)* | 2021-09-29 | 2022-01-11 | 中国人民解放军火箭军工程大学 | Hyperspectral classification method combining graph structure and convolutional neural network |
| CN113920442B (en)* | 2021-09-29 | 2024-06-18 | 中国人民解放军火箭军工程大学 | Hyperspectral classification method combining graph structure and convolutional neural network |
| CN113837314A (en)* | 2021-10-09 | 2021-12-24 | 陕西科技大学 | A Hyperspectral Image Classification Method Based on Hybrid Convolutional Neural Networks |
| CN113963207A (en)* | 2021-10-21 | 2022-01-21 | 江南大学 | Hyperspectral image classification method based on spatial spectral information feature-guided fusion network |
| CN113963207B (en)* | 2021-10-21 | 2024-03-29 | 江南大学 | Hyperspectral image classification method based on spatial spectrum information feature guided fusion network |
| CN114494846A (en)* | 2021-12-18 | 2022-05-13 | 南京理工大学 | Hyperspectral image classification method based on multi-scale internal volume network |
| CN114022960A (en)* | 2022-01-05 | 2022-02-08 | 阿里巴巴达摩院(杭州)科技有限公司 | Model training and behavior recognition method, device, electronic device, and storage medium |
| CN114330591A (en)* | 2022-01-05 | 2022-04-12 | 安徽大学 | A hyperspectral classification method based on spatial-spectral dimension transformation |
| CN115937565A (en)* | 2022-02-14 | 2023-04-07 | 苏州大学 | Hyperspectral Image Classification Method Based on Adaptive L-BFGS Algorithm |
| CN115937565B (en)* | 2022-02-14 | 2025-08-22 | 苏州大学 | Hyperspectral image classification method based on adaptive L-BFGS algorithm |
| CN114694140A (en)* | 2022-03-22 | 2022-07-01 | 南京林业大学 | Directional training platycodon grandiflorum impurity identification method based on spectral imaging |
| CN114708457A (en)* | 2022-03-31 | 2022-07-05 | 南京林业大学 | A hyperspectral deep learning recognition method for anti-purple fringing recognition |
| CN114842264A (en)* | 2022-05-17 | 2022-08-02 | 北京理工大学 | Hyperspectral image classification method based on multi-scale spatial spectral feature joint learning |
| CN115240074A (en)* | 2022-09-22 | 2022-10-25 | 山东锋士信息技术有限公司 | Hyperspectral image classification method and device based on covariance expression |
| CN116310572A (en)* | 2023-03-23 | 2023-06-23 | 齐齐哈尔大学 | Pyramid multi-scale convolution and self-attention combined hyperspectral image classification method |
| CN116310572B (en)* | 2023-03-23 | 2024-01-23 | 齐齐哈尔大学 | Hyperspectral image classification method combining pyramid multi-scale convolution and self-attention |
| CN116580252A (en)* | 2023-06-16 | 2023-08-11 | 西安电子科技大学 | Hyperspectral Image Classification Method Based on Multiscale Dense Connection and Feature Aggregation Network |
| CN116580252B (en)* | 2023-06-16 | 2025-07-01 | 西安电子科技大学 | Hyperspectral image classification method based on multiscale dense connection and feature aggregation network |
| CN118196033A (en)* | 2024-03-18 | 2024-06-14 | 江苏商贸职业学院 | Non-contact high-precision measurement method and system based on machine vision fusion |
| CN118429800A (en)* | 2024-04-17 | 2024-08-02 | 湖南农业大学 | Hyperspectral image classification method based on tensor width learning network |
| CN118317110A (en)* | 2024-06-11 | 2024-07-09 | 国网安徽省电力有限公司超高压分公司 | A method for enhancing image transmission by joint coding of substation source and channel |
| CN119672440A (en)* | 2024-12-23 | 2025-03-21 | 西安电子科技大学 | Few-shot HSI classification method based on prototype mask comparison and cross-stage feature refinement |
| Publication number | Publication date |
|---|---|
| CN110321963B (en) | 2022-03-04 |
| Publication | Publication Date | Title |
|---|---|---|
| CN110321963B (en) | Hyperspectral image classification method based on fusion of multi-scale and multi-dimensional spatial spectral features | |
| Lu et al. | Bidirectional adaptive feature fusion for remote sensing scene classification | |
| CN102646200B (en) | Image classifying method and system for self-adaption weight fusion of multiple classifiers | |
| Liu et al. | Class-specific sparse multiple kernel learning for spectral–spatial hyperspectral image classification | |
| CN108491849A (en) | Hyperspectral image classification method based on three-dimensional dense connection convolutional neural networks | |
| CN111639587B (en) | Hyperspectral image classification method based on multi-scale spectrum space convolution neural network | |
| CN109766858A (en) | Three-dimensional convolution neural network hyperspectral image classification method combined with bilateral filtering | |
| CN107358260B (en) | A Multispectral Image Classification Method Based on Surface Wave CNN | |
| CN107480620B (en) | Remote sensing image automatic target identification method based on heterogeneous feature fusion | |
| CN113095409A (en) | Hyperspectral image classification method based on attention mechanism and weight sharing | |
| CN110084159A (en) | Hyperspectral image classification method based on the multistage empty spectrum information CNN of joint | |
| Tu et al. | Hyperspectral image classification using a superpixel–pixel–subpixel multilevel network | |
| CN103208011B (en) | Based on average drifting and the hyperspectral image space-spectral domain classification method organizing sparse coding | |
| CN104778476B (en) | A kind of image classification method | |
| CN102096825A (en) | Graph-based semi-supervised high-spectral remote sensing image classification method | |
| CN110084311A (en) | Hyperspectral image band selection method based on ternary weight convolutional neural networks | |
| CN108985360A (en) | Hyperspectral classification method based on expanding morphology and Active Learning | |
| CN108564111A (en) | A kind of image classification method based on neighborhood rough set feature selecting | |
| CN103390170A (en) | Surface feature type texture classification method based on multispectral remote sensing image texture elements | |
| CN109034213B (en) | Method and system for hyperspectral image classification based on correlation entropy principle | |
| CN115240072A (en) | Hyperspectral multi-class change detection method based on multidirectional multi-scale spectrum-space residual convolution neural network | |
| CN108647719B (en) | Non-surveillance clustering method for big data quantity spectral remote sensing image classification | |
| CN113205143A (en) | Multi-scale superpixel hyperspectral remote sensing image classification method based on space-spectrum coupling characteristics | |
| CN111199251B (en) | A Weighted Neighborhood-Based Multiscale Hyperspectral Image Classification Method | |
| CN112966629A (en) | Remote sensing image scene classification method based on image transformation and BoF model |
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |