Remote sensing hyperspectral image classification method based on digital elevation model feature extraction

Abstract

The invention discloses a remote sensing hyperspectral image classification method based on digital elevation model feature extraction, and belongs to the technical field of hyperspectral image classification. The problem that an existing hyperspectral remote sensing image classification method is low in classification accuracy is solved. According to the method, the digital elevation model DEM, the digital elevation model INSAR-DEM and the Landsat8 satellite remote sensing hyperspectral image of the to-be-classified area are obtained, the finally corrected digital elevation model DEM can be obtained based on the digital elevation model DEM and the digital elevation model INSAR-DEM, and then classification is performed by combining the elevation characteristics of the final digital elevation model DEM and the spectral characteristics of the image. The elevation information is considered in the classification process, so that the classification accuracy is effectively improved. The method can be applied to remote sensing hyperspectral image classification.

Description

Technical Field

[0001] The invention belongs to the technical field of hyperspectral image classification, and in particular relates to a remote sensing hyperspectral image classification method based on digital elevation model feature extraction. Background Art

[0002] Hyperspectral remote sensing is a new remote sensing technology characterized by the integration of remote sensing images and spectra. Its spectral resolution can reach nanometers and is widely used in mineral exploration, precision agriculture, and environmental monitoring. Therefore, the study of efficient processing and interpretation of hyperspectral data has important theoretical significance and practical application value. Hyperspectral classification and target detection are the main contents of hyperspectral data processing. Hyperspectral data provides rich spectral information, which creates conditions for people to study the characteristics of surface objects and identify objects. Many classification and feature extraction methods, with the development of machine learning and deep learning technology, hyperspectral image classification based on machine learning and deep learning has made breakthrough progress in methods and performance, such as the nearest neighbor classifier and support vector machine classifier in machine learning, the autoencoder neural network in the deep learning model can achieve nonlinear dimensionality reduction of high-dimensional data, and the convolutional neural network (CNN) can automatically extract spatial features from images and then classify based on the extracted spatial features.

[0003] However, the existing hyperspectral remote sensing image classification methods only consider the image features themselves, but do not consider the elevation features of the image area. Therefore, the classification accuracy of the existing hyperspectral remote sensing image classification methods is still low. It is very necessary to propose a hyperspectral remote sensing image classification method that combines elevation features. Summary of the invention

[0004] The purpose of the present invention is to solve the problem of low classification accuracy of existing hyperspectral remote sensing image classification methods, and propose a remote sensing hyperspectral image classification method based on digital elevation model feature extraction.

[0005] The technical solution adopted by the present invention to solve the above technical problems is: a remote sensing hyperspectral image classification method based on digital elevation model feature extraction, the method specifically comprises the following steps:

[0006] Step 1: Obtain the digital elevation model DEM, digital elevation model INSAR-DEM and Landsat8 satellite remote sensing hyperspectral image of the area to be classified;

[0007] Step 2: According to the digital elevation model DEM and the digital elevation model INSAR-DEM, the abnormal pixel values ​​in the digital elevation model DEM are determined, and the abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixels to obtain the corrected digital elevation model DEM;

[0008] Step 3: using the digital elevation model DEM corrected in step 2 as the input of the first neural network model, and outputting the final digital elevation model DEM through the first neural network model;

[0009] Step 4: Use the Landsat 8 satellite remote sensing hyperspectral image as the input of the second neural network model to extract the spectral features of the image;

[0010] Step 5: Extract elevation features from the final digital elevation model DEM, and then combine the elevation features with the spectral features to obtain combined features;

[0011] The combined features are input into the support vector machine, and the classification results are output through the support vector machine.

[0012] Furthermore, the method of determining the abnormal pixel values ​​in the digital elevation model DEM according to the digital elevation model DEM and the digital elevation model INSAR-DEM is as follows:

[0013] Step 1: Obtain the corresponding pixel value of each pixel value in the digital elevation model DEM in the digital elevation model INSAR-DEM, and convert the pixel value a in the digital elevation model DEM into i The corresponding pixel value in the digital elevation model INSAR-DEM is denoted as b i , i = 1, 2, ..., n, n is the total number of pixel values ​​in the digital elevation model DEM, that is, the total number of pixel values ​​in the digital elevation model INSAR-DEM;

[0014] Step 2: Calculate a i and b i The average value c i , i=1,2,…,n;

[0015] Step 3: Calculate the deviation value |a i -c i |,|·| represent taking absolute values;

[0016] If the deviation value |a i -c i | is greater than the set threshold, then a i is the abnormal pixel value in the digital elevation model DEM;

[0017] If the deviation value |a i -c i| is less than or equal to the set threshold, then a i is the normal pixel value in the digital elevation model DEM;

[0018] Step 4: c i As the abnormal pixel value a i Correction value.

[0019] Furthermore, the abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixel values ​​to obtain a corrected digital elevation model DEM; specifically:

[0020] The abnormal pixel value a in the digital elevation model DEM i Replace with c i .

[0021] Furthermore, the first neural network model includes a first convolution module, a second convolution module, a third convolution module, a flattening layer, a first BP layer, a first Dropout layer, a second BP layer, a second Dropout layer and a regression layer in sequence;

[0022] The pixel values ​​in the corrected digital elevation model DEM are used as the input of the first convolution module;

[0023] Using the output of the first convolution module as the input of the second convolution module;

[0024] The output of the second convolution module is used as the input of the third convolution module;

[0025] Use the output of the third convolutional module as the input of the flattening layer;

[0026] Use the output of the flattening layer as the input of the first BP layer;

[0027] The output of the first BP layer is used as the input of the first Dropout layer;

[0028] The output of the first Dropout layer is used as the input of the second BP layer;

[0029] The output of the second BP layer is used as the input of the second Dropout layer;

[0030] The output of the second Dropout layer is used as the input of the regression layer, and the elevation value of each pixel in the final digital elevation model DEM is output through the regression layer.

[0031] Furthermore, the first convolution module includes a convolution layer, a pooling layer, a batch normalization layer and a Relu activation function layer in sequence, the input of the first convolution module is used as the input of the convolution layer, the output of the convolution layer is used as the input of the pooling layer, the output of the pooling layer is used as the input of the batch normalization layer, the output of the batch normalization layer is used as the input of the Relu activation function layer, and the output of the Relu activation function layer is used as the output of the first convolution module;

[0032] The structures and working processes of the second convolution module and the third convolution module are the same as those of the first convolution module.

[0033] Furthermore, the second neural network model includes a multi-scale feature extraction module and a long short-term memory module based on an attention mechanism, wherein:

[0034] The multi-scale feature extraction module includes a first convolution unit, a second convolution unit and a third convolution unit;

[0035] Using the input of the second neural network model as the input of the first convolution unit;

[0036] The output of the first convolution unit is used as the input of the second convolution unit;

[0037] The output of the second convolution unit is used as the input of the third convolution unit;

[0038] The long short-term memory module based on the attention mechanism includes a first long short-term memory unit, a second long short-term memory unit and a third long short-term memory unit;

[0039] Using the output of the first convolutional unit as the input of the first long short-term memory unit;

[0040] Using the output of the second convolutional unit as the input of the second long short-term memory unit;

[0041] Using the output of the third convolutional unit as the input of the third long short-term memory unit;

[0042] The features output by the first long short-term memory unit, the second long short-term memory unit and the third long short-term memory unit are then fused to obtain the spectral features.

[0043] Furthermore, the working process of the first convolution unit is:

[0044] In the first convolution unit, the input image passes through the convolution layer and the maximum pooling layer in sequence, and the output of the maximum pooling layer is used as the output of the first convolution unit;

[0045] The working processes of the first convolution unit and the second convolution unit are the same as the working process of the first convolution unit.

[0046] Furthermore, the first long short-term memory unit includes a long short-term memory network, a first fully connected layer, a tanh activation function layer, a second fully connected layer and a softmax activation function layer.

[0047] Furthermore, the working process of the first long short-term memory unit is:

[0048] First, the feature vector of each time step is output through the long short-term memory network, and then the feature vectors of different time steps are spliced ​​into a feature matrix;

[0049] Then the feature matrix passes through the first fully connected layer, the tanh activation function layer and the second fully connected layer in sequence;

[0050] The output of the second fully connected layer is used as the input of the softmax activation function layer, and the weight vector is output through the softmax activation function layer;

[0051] Perform weighted summation of the feature vector of each time step and the weight vector to obtain the feature output by the first long short-term memory unit;

[0052] The working process of the second long short-term memory unit and the third long short-term memory unit is the same as the working process of the first long short-term memory unit.

[0053] Furthermore, the final digital elevation model DEM is visualized.

[0054] The beneficial effects of the present invention are:

[0055] The present invention obtains a digital elevation model DEM, a digital elevation model INSAR-DEM and a Landsat 8 satellite remote sensing hyperspectral image of a region to be classified, and a final corrected digital elevation model DEM can be obtained based on the digital elevation model DEM and the digital elevation model INSAR-DEM, and then classification is performed in combination with the elevation features of the final digital elevation model DEM and the spectral features of the image. Since the classification process of the present invention takes elevation information into consideration, the accuracy of classification is effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0056] Figure 1 The invention discloses a flow chart of a remote sensing hyperspectral image classification method based on digital elevation model feature extraction. DETAILED DESCRIPTION

[0057] Specific implementation method 1: Combination Figure 1 The present embodiment is described as follows. The present embodiment describes a remote sensing hyperspectral image classification method based on digital elevation model feature extraction, and the method specifically comprises the following steps:

[0058] Step 1: Obtain the digital elevation model DEM, digital elevation model INSAR-DEM and Landsat8 satellite remote sensing hyperspectral image of the area to be classified;

[0059] Step 2: According to the digital elevation model DEM and the digital elevation model INSAR-DEM, the abnormal pixel values ​​in the digital elevation model DEM are determined, and the abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixels to obtain the corrected digital elevation model DEM;

[0060] Step 3: using the digital elevation model DEM corrected in step 2 as the input of the first neural network model, and outputting the final digital elevation model DEM through the first neural network model;

[0061] Step 4: Use the Landsat 8 satellite remote sensing hyperspectral image as the input of the second neural network model to extract the spectral features of the image;

[0062] Step 5: Extract elevation features from the final digital elevation model DEM, and then combine the elevation features with the spectral features to obtain combined features;

[0063] The combined features are input into the support vector machine, and the classification results are output through the support vector machine.

[0064] The present invention first obtains a digital elevation model DEM, a digital elevation model INSAR-DEM and a Landsat 8 satellite remote sensing hyperspectral image of a to-be-classified area, then obtains a final corrected digital elevation model DEM based on the digital elevation model DEM and the digital elevation model INSAR-DEM, and finally, classification is performed in combination with the elevation features of the final digital elevation model DEM and the spectral features of the image. Since the classification process of the present invention takes elevation information into consideration, the accuracy of classification is effectively improved.

[0065] Specific implementation method 2: This implementation method is different from the specific implementation method 1 in that the abnormal pixel values ​​in the digital elevation model DEM are determined according to the digital elevation model DEM and the digital elevation model INSAR-DEM; specifically:

[0066] Step 1: Obtain the corresponding pixel value of each pixel value in the digital elevation model DEM in the digital elevation model INSAR-DEM, and convert the pixel value a in the digital elevation model DEM into i The corresponding pixel value in the digital elevation model INSAR-DEM is denoted as b i , i = 1, 2, ..., n, n is the total number of pixel values ​​in the digital elevation model DEM, that is, the total number of pixel values ​​in the digital elevation model INSAR-DEM;

[0067] Step 2: Calculate a i and b i The average value c i , i=1,2,…,n;

[0068] Step 3: Calculate the deviation value |a i -c i |,|·| represent taking absolute values;

[0069] If the deviation value |a i -c i | is greater than the set threshold, then a i is the abnormal pixel value in the digital elevation model DEM;

[0070] If the deviation value |a i -c i | is less than or equal to the set threshold, then a i is the normal pixel value in the digital elevation model DEM;

[0071] Step 4: c i As the abnormal pixel value a i Correction value.

[0072] The other steps and parameters are the same as those in the first embodiment.

[0073] Specific implementation method three: This implementation method is different from specific implementation method one or two in that the abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixel values ​​to obtain the corrected digital elevation model DEM; specifically:

[0074] The abnormal pixel value a in the digital elevation model DEM i Replace with c i .

[0075] The other steps and parameters are the same as those in the first or second embodiment.

[0076] Specific implementation method 4: This implementation method is different from any one of specific implementation methods 1 to 3 in that the first neural network model sequentially includes a first convolution module, a second convolution module, a third convolution module, a flattening layer, a first BP layer, a first Dropout layer, a second BP layer, a second Dropout layer and a regression layer;

[0077] The pixel values ​​in the corrected digital elevation model DEM are used as the input of the first convolution module;

[0078] Using the output of the first convolution module as the input of the second convolution module;

[0079] The output of the second convolution module is used as the input of the third convolution module;

[0080] Use the output of the third convolutional module as the input of the flattening layer;

[0081] Use the output of the flattening layer as the input of the first BP layer;

[0082] The output of the first BP layer is used as the input of the first Dropout layer;

[0083] The output of the first Dropout layer is used as the input of the second BP layer;

[0084] The output of the second BP layer is used as the input of the second Dropout layer;

[0085] The output of the second Dropout layer is used as the input of the regression layer, and the elevation value of each pixel in the final digital elevation model DEM is output through the regression layer.

[0086] The other steps and parameters are the same as those in Specific Embodiments 1 to 3.

[0087] Specific implementation mode five: This implementation mode is different from any one of specific implementation modes one to four in that the first convolution module includes a convolution layer, a pooling layer, a batch normalization layer and a Relu activation function layer in sequence, the input of the first convolution module is used as the input of the convolution layer, the output of the convolution layer is used as the input of the pooling layer, the output of the pooling layer is used as the input of the batch normalization layer, the output of the batch normalization layer is used as the input of the Relu activation function layer, and the output of the Relu activation function layer is used as the output of the first convolution module;

[0088] The structures and working processes of the second convolution module and the third convolution module are the same as those of the first convolution module.

[0089] The other steps and parameters are the same as those in Specific Embodiments 1 to 4.

[0090] Specific implementation example 6: This implementation example is different from any one of specific implementation examples 1 to 5 in that the second neural network model includes a multi-scale feature extraction module and a long short-term memory module based on an attention mechanism, wherein:

[0091] The multi-scale feature extraction module includes a first convolution unit, a second convolution unit and a third convolution unit;

[0092] Using the input of the second neural network model as the input of the first convolution unit;

[0093] The output of the first convolution unit is used as the input of the second convolution unit;

[0094] The output of the second convolution unit is used as the input of the third convolution unit;

[0095] The long short-term memory module based on the attention mechanism includes a first long short-term memory unit, a second long short-term memory unit and a third long short-term memory unit;

[0096] Using the output of the first convolutional unit as the input of the first long short-term memory unit;

[0097] Using the output of the second convolutional unit as the input of the second long short-term memory unit;

[0098] Using the output of the third convolutional unit as the input of the third long short-term memory unit;

[0099] The features output by the first long short-term memory unit, the second long short-term memory unit and the third long short-term memory unit are then fused to obtain the spectral features.

[0100] The other steps and parameters are the same as those in Specific Implementation Methods 1 to 5.

[0101] After extracting spectral features of different scales, the present invention fuses the extracted features, so that the classification result comprehensively considers the spectral features of different scales, thereby improving the classification accuracy.

[0102] Specific implementation method 7: This implementation method is different from any one of specific implementation methods 1 to 6 in that the working process of the first convolution unit is:

[0103] In the first convolution unit, the input image passes through the convolution layer and the maximum pooling layer in sequence, and the output of the maximum pooling layer is used as the output of the first convolution unit;

[0104] The working processes of the first convolution unit and the second convolution unit are the same as the working process of the first convolution unit.

[0105] The other steps and parameters are the same as those in Specific Embodiments 1 to 6.

[0106] Specific implementation eight: This implementation differs from any one of specific implementations one to seven in that the first long short-term memory unit includes a long short-term memory network, a first fully connected layer, a tanh activation function layer, a second fully connected layer and a softmax activation function layer.

[0107] The other steps and parameters are the same as those in Specific Embodiments 1 to 7.

[0108] Specific implementation method 9: This implementation method is different from any one of specific implementation methods 1 to 8 in that the working process of the first long short-term memory unit is:

[0109] First, the feature vector of each time step is output through the long short-term memory network, and then the feature vectors of different time steps are spliced ​​into a feature matrix;

[0110] Then the feature matrix passes through the first fully connected layer, the tanh activation function layer and the second fully connected layer in sequence;

[0111] The output of the second fully connected layer is used as the input of the softmax activation function layer, and the weight vector is output through the softmax activation function layer;

[0112] Perform weighted summation of the feature vector of each time step and the weight vector to obtain the feature output by the first long short-term memory unit;

[0113] The working process of the second long short-term memory unit and the third long short-term memory unit is the same as the working process of the first long short-term memory unit.

[0114] The other steps and parameters are the same as those in Specific Embodiments 1 to 8.

[0115] Specific implementation method ten: This implementation method is different from any one of specific implementation methods one to nine in that the final digital elevation model DEM is visualized.

[0116] The other steps and parameters are the same as those in Specific Embodiments 1 to 9.

[0117] The above calculation examples of the present invention are only used to explain the calculation model and calculation process of the present invention in detail, and are not intended to limit the implementation methods of the present invention. For ordinary technicians in the relevant field, other different forms of changes or modifications can be made based on the above description. It is impossible to list all the implementation methods here. All obvious changes or modifications derived from the technical solution of the present invention are still within the scope of protection of the present invention.

Claims

1. A remote sensing hyperspectral image classification method based on digital elevation model feature extraction, characterized in that: The method specifically comprises the following steps: Step 1: Obtain the digital elevation model DEM, digital elevation model INSAR-DEM and Landsat8 satellite remote sensing hyperspectral image of the area to be classified; Step 2: According to the digital elevation model DEM and the digital elevation model INSAR-DEM, the abnormal pixel values ​​in the digital elevation model DEM are determined, and the abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixels to obtain the corrected digital elevation model DEM; Step 3: using the digital elevation model DEM corrected in step 2 as the input of the first neural network model, and outputting the final digital elevation model DEM through the first neural network model; Step 4: Use the Landsat 8 satellite remote sensing hyperspectral image as the input of the second neural network model to extract the spectral features of the image; Step 5: Extract elevation features from the final digital elevation model DEM, and then combine the elevation features with the spectral features to obtain combined features; The combined features are input into the support vector machine, and the classification results are output through the support vector machine.

2. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 1 is characterized in that: The method of determining the abnormal pixel value in the digital elevation model DEM according to the digital elevation model DEM and the digital elevation model INSAR-DEM is as follows: Step 1: Obtain the corresponding pixel value of each pixel value in the digital elevation model DEM in the digital elevation model INSAR-DEM, and convert the pixel value a in the digital elevation model DEM into i The corresponding pixel value in the digital elevation model INSAR-DEM is denoted as b i , i = 1, 2, ..., n, n is the total number of pixel values ​​in the digital elevation model DEM, that is, the total number of pixel values ​​in the digital elevation model INSAR-DEM; Step 2: Calculate a i and b i The average value c i , i=1,2,…,n; Step 3: Calculate the deviation value |a i -c i |,|·| represent taking absolute values; If the deviation value |a i -c i | is greater than the set threshold, then a i is the abnormal pixel value in the digital elevation model DEM; If the deviation value |a i -c i | is less than or equal to the set threshold, then a i is the normal pixel value in the digital elevation model DEM; Step 4: c i As the abnormal pixel value a i Correction value.

3. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 2 is characterized in that: The abnormal pixel values ​​in the digital elevation model DEM are corrected according to the correction values ​​of the abnormal pixel values ​​to obtain the corrected digital elevation model DEM; specifically: The abnormal pixel value a in the digital elevation model DEM i Replace with c i .

4. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 3 is characterized in that: The first neural network model includes a first convolution module, a second convolution module, a third convolution module, a flattening layer, a first BP layer, a first Dropout layer, a second BP layer, a second Dropout layer and a regression layer in sequence; The pixel values ​​in the corrected digital elevation model DEM are used as the input of the first convolution module; Using the output of the first convolution module as the input of the second convolution module; The output of the second convolution module is used as the input of the third convolution module; Use the output of the third convolutional module as the input of the flattening layer; Use the output of the flattening layer as the input of the first BP layer; The output of the first BP layer is used as the input of the first Dropout layer; The output of the first Dropout layer is used as the input of the second BP layer; The output of the second BP layer is used as the input of the second Dropout layer; The output of the second Dropout layer is used as the input of the regression layer, and the elevation value of each pixel in the final digital elevation model DEM is output through the regression layer.

5. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 4 is characterized in that: The first convolution module includes a convolution layer, a pooling layer, a batch normalization layer and a Relu activation function layer in sequence. The input of the first convolution module is used as the input of the convolution layer, the output of the convolution layer is used as the input of the pooling layer, the output of the pooling layer is used as the input of the batch normalization layer, and the output of the batch normalization layer is used as the input of the Relu activation function layer. The output of the Relu activation function layer is used as the output of the first convolution module; The structures and working processes of the second convolution module and the third convolution module are the same as those of the first convolution module.

6. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 5 is characterized in that: The second neural network model includes a multi-scale feature extraction module and a long short-term memory module based on an attention mechanism, wherein: The multi-scale feature extraction module includes a first convolution unit, a second convolution unit and a third convolution unit; Using the input of the second neural network model as the input of the first convolution unit; The output of the first convolution unit is used as the input of the second convolution unit; The output of the second convolution unit is used as the input of the third convolution unit; The long short-term memory module based on the attention mechanism includes a first long short-term memory unit, a second long short-term memory unit and a third long short-term memory unit; Using the output of the first convolutional unit as the input of the first long short-term memory unit; Using the output of the second convolutional unit as the input of the second long short-term memory unit; Using the output of the third convolutional unit as the input of the third long short-term memory unit; The features output by the first long short-term memory unit, the second long short-term memory unit and the third long short-term memory unit are then fused to obtain the spectral features.

7. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 6 is characterized in that: The working process of the first convolution unit is: In the first convolution unit, the input image passes through the convolution layer and the maximum pooling layer in sequence, and the output of the maximum pooling layer is used as the output of the first convolution unit; The working processes of the first convolution unit and the second convolution unit are the same as the working process of the first convolution unit.

8. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 7 is characterized in that: The first long short-term memory unit includes a long short-term memory network, a first fully connected layer, a tanh activation function layer, a second fully connected layer and a softmax activation function layer.

9. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 8 is characterized in that: The working process of the first long short-term memory unit is: First, the feature vector of each time step is output through the long short-term memory network, and then the feature vectors of different time steps are spliced ​​into a feature matrix; Then the feature matrix passes through the first fully connected layer, the tanh activation function layer and the second fully connected layer in sequence; The output of the second fully connected layer is used as the input of the softmax activation function layer, and the weight vector is output through the softmax activation function layer; Perform weighted summation of the feature vector of each time step and the weight vector to obtain the feature output by the first long short-term memory unit; The working process of the second long short-term memory unit and the third long short-term memory unit is the same as the working process of the first long short-term memory unit.

10. The remote sensing hyperspectral image classification method based on digital elevation model feature extraction according to claim 9, characterized in that: The final digital elevation model DEM is visualized.

Images