CN107239751B - High-resolution SAR image classification method based on non-subsampled contourlet full convolution network - Google Patents
High-resolution SAR image classification method based on non-subsampled contourlet full convolution network Download PDFInfo
- Publication number
- CN107239751B CN107239751B CN201710364900.8A CN201710364900A CN107239751B CN 107239751 B CN107239751 B CN 107239751B CN 201710364900 A CN201710364900 A CN 201710364900A CN 107239751 B CN107239751 B CN 107239751B
- Authority
- CN
- China
- Prior art keywords
- data set
- layer
- image
- matrix
- size
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/13—Satellite images
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
- G06V10/443—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components by matching or filtering
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Astronomy & Astrophysics (AREA)
- Remote Sensing (AREA)
- Image Analysis (AREA)
Abstract
A high-resolution SAR image classification method based on a non-downsampling contourlet full convolution network comprises the steps of inputting a high-resolution SAR image to be classified, carrying out multilayer non-downsampling contourlet transformation on each pixel point in the image, and obtaining a low-frequency coefficient and a high-frequency coefficient of each pixel point; selecting and fusing the low-frequency coefficient and the high-frequency coefficient to form a characteristic matrix F based on pixel points; normalizing the element values in the feature matrix F to obtain a normalized feature matrix F1; dicing the normalized feature matrix F1 to obtain a feature block matrix F2 as sample data; constructing a training data set characteristic matrix W1 and a testing data set characteristic matrix W2; constructing a classification model based on a full convolution neural network; training a classification model; the trained model is used for classifying the test data set T to obtain the category of each pixel point in the test data set T, the obtained category of each pixel point is compared with the category label chart, the classification accuracy is calculated, and the classification precision and speed are improved.
Description
Technical Field
The invention belongs to the field of image processing, and particularly relates to a high-resolution SAR image classification method based on a non-downsampling contourlet full convolution network, which can be applied to a high-resolution SAR image and effectively improve the identification precision of a target.
Background
Synthetic Aperture Radar (SAR) is a remote sensing sensor widely researched and applied in recent years, and compared with other sensors such as optical sensors and infrared sensors, SAR imaging is not limited by conditions such as weather and illumination, and can perform all-weather and all-day investigation on an interested target. And the SAR also has certain penetrating power, and can realize the detection of the target under the adverse conditions of cloud interference, tree bundle shielding, shallow ground surface burying of the target and the like. In addition, due to the special imaging mechanism of the SAR, the high-resolution SAR image contains different contents from other sensors, and richer and comprehensive information is provided for target detection. The SAR has a great number of remarkable advantages and great application potential. In recent years, research on the SAR technology has attracted much attention, and many research results have been successfully applied to the aspects of environmental monitoring, topographic survey, target detection, and the like.
The key of the high-resolution SAR image classification is the extraction of target features of the high-resolution SAR image, and the existing SAR image classification technologies comprise a statistical-based classification method, an image texture-based classification method and a deep learning-based classification method.
The statistical-based classification method classifies images according to the difference of statistical characteristics of image regions with different properties, but the method ignores the spatial distribution characteristics of the images, so the classification result is not ideal. In recent years, some classification methods based on texture features, such as a method based on a gray level co-occurrence matrix (GLCM), a method based on a Markov Random Field (MRF), a Gabor wavelet method, and the like, have appeared, but due to the mechanism of coherent imaging of the SAR image, the texture in the SAR image is not obvious and robust, and in addition, the computer texture features need to scan the image point by point, and the calculation amount is huge and cannot meet the real-time requirement.
The conventional SAR image classification method can only depend on manual extraction of some shallow features representing target characteristics, the shallow features are obtained only by converting original input signals into specific problem spaces, and neighborhood correlation among target pixel points cannot be completely represented. In 2006, Hinton et al proposed an unsupervised, layer-by-layer greedy training method that solved the "gradient dissipation" problem caused by depth increase. Subsequently, many scholars propose various DL models according to different application backgrounds, such as Deep Belief Network (DBN), Stacked denoising self-coders (SDA), Convolutional Neural Network (CNN), and the like. However, the above feature extraction methods do not consider the multi-scale, multi-directional, and multi-resolution characteristics of the high-resolution SAR image, and therefore it is difficult to obtain high classification accuracy for a high-resolution SAR image with a complex background.
Disclosure of Invention
The invention aims to provide a high-resolution SAR image classification method based on a non-downsampling contourlet full convolution network, which combines the characteristics of multi-scale, multi-direction and multi-resolution of a high-resolution SAR image, improves the accuracy and the classification speed of image classification, and further effectively improves the identification precision of a target.
In order to achieve the purpose, the technical scheme adopted by the invention comprises the following steps:
1) inputting a high-resolution SAR image to be classified, and carrying out multilayer non-downsampling contourlet transformation on each pixel point in the image to obtain a low-frequency coefficient and a high-frequency coefficient of each pixel point;
2) selecting and fusing the low-frequency coefficient and the high-frequency coefficient to form a characteristic matrix F based on pixel points;
3) normalizing the element values in the feature matrix F to be between [0 and 1] to obtain a normalized feature matrix F1;
4) dicing the normalized feature matrix F1 to obtain a feature block matrix F2 as sample data;
5) constructing a training data set characteristic matrix W1 through a training data set D, and constructing a test data set characteristic matrix W2 through a test data set T;
6) constructing a classification model based on a full convolution neural network;
7) training the classification model through a training data set D to obtain a trained model;
8) classifying the test data set T by using the trained model to obtain the category of each pixel point in the test data set T, comparing the obtained category of each pixel point with the category label chart, and calculating the classification accuracy.
The step 1) carries out three-layer non-down sampling contourlet transformation on each pixel point in the image; the non-downsampling contourlet transformation comprises non-downsampling pyramid decomposition and non-downsampling direction filter decomposition, the non-downsampling pyramid decomposition decomposes a time-frequency plane into a low-frequency filial generation and a plurality of annular high-frequency filial generations through a non-downsampling filter bank, and a band-pass image formed by the non-downsampling pyramid decomposition is decomposed through the non-downsampling direction filter to obtain coefficients of the band-pass sub-images.
And 2) sequencing the high-frequency coefficients from large to small, selecting the high-frequency coefficients of the first 50% of the high-frequency coefficients, fusing the high-frequency coefficients with the low-frequency coefficients after the third layer of transformation, defining the size of a characteristic matrix F based on the pixel points as M1 multiplied by M2 multiplied by 1, defining the size of the characteristic matrix F based on the pixel points as M1 as the length of the SAR image to be classified, defining the size of the characteristic matrix F based on the pixel points as M2 as the width of the SAR image to be classified, and assigning the fusion result to.
The normalization in the step 3) is realized by a characteristic linear scaling method, a characteristic standardization method or a characteristic whitening method; firstly, solving the maximum value max (F) of a characteristic matrix F based on pixel points by a characteristic linear scaling method; and dividing each element in the characteristic matrix F based on the pixel points by the maximum value max (F) to obtain a normalized characteristic matrix F1.
The step 4) is to slice the normalized feature matrix F1 according to the size of 128 × 128 and the interval of 50.
The specific operation of the step 5) is as follows:
5a) dividing the SAR image surface features into 3 classes, recording the positions of pixel points corresponding to each class in the image to be classified, and generating three positions A1, A2 and A3 respectively representing the positions of the pixel points of the three classes of surface features in the image to be classified;
5b) randomly selecting 5% of elements from the A1, A2 and A3 to generate three pixel positions B1, B2 and B3 which correspond to different types of ground objects and are selected as training data sets, wherein B1 corresponds to the position of a pixel selected as a training data set in a1 st type of ground object in an image to be classified, B2 corresponds to the position of a pixel selected as a training data set in a2 nd type of ground object in the image to be classified, B3 corresponds to the position of a pixel selected as a training data set in A3 rd type of ground object in the image to be classified, and combining the elements from the B1, the B2 and the B3 to form a position L1 of all pixels of the training data set in the image to be classified;
5c) generating 3 pixel point positions C1, C2 and C3 corresponding to different types of ground objects and selected as test data sets by using the rest 95% of the elements in the A1, A2 and A3, wherein C1 is the position of a pixel point selected as a test data set in a1 st type of ground object in an image to be classified, C2 is the position of a pixel point selected as a test data set in a2 nd type of ground object in the image to be classified, C3 is the position of a pixel point selected as a test data set in A3 rd type of ground object in the image to be classified, and combining the elements in the C1, C2 and C3 to form the position L2 of all the pixel points of the test data set in the image to be classified;
5d) defining a training data set characteristic matrix W1 of a training data set D, taking values at corresponding positions in a characteristic block matrix F2 according to L1, and assigning the values to a training data set characteristic matrix W1 of the training data set D;
5e) the test data set feature matrix W2 for the test data set T is defined, the values at the corresponding positions are taken according to L2 in the feature block matrix F2 and assigned to the test data set feature matrix W2 for the test data set T.
The step 6) of constructing the classification model based on the full convolution neural network comprises the following steps:
6a) selecting a 17-layer deep neural network consisting of an input layer, a convolutional layer, a pooling layer, a convolutional layer, a Dropout layer, a convolutional layer, a deconvolution up-sampling layer, a Crop layer and a softmax classifier in sequence, wherein the parameters of each layer are as follows:
setting the number of feature maps to be 3 for the input layer of the layer 1;
for the 2 nd convolutional layer, setting the number of feature maps to be 32 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 3 rd pooling layer, the down-sampling size is set to 2;
for the 4 th convolutional layer, setting the number of feature maps to be 64 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 5 th pooling layer, the down-sampling size is set to 2;
for the convolution layer of the layer 6, setting the number of feature maps to be 96 and the size of a convolution kernel to be 3 multiplied by 3;
for the 7 th pooling layer, the downsampling size is set to 2;
for the 8 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the downsampling size to be 2 for the 9 th pooling layer;
for the 10 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the sparsity factor to 0.5 for the 11 th Dropout layer;
for the 12 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 13 th Dropout layer, setting the sparsity factor to 0.5;
for the 14 th convolutional layer, setting the number of feature maps to be 2 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 15 th deconvolution up-sampling layer, setting the number of feature maps to be 2 and the convolution kernel size to be 32 multiplied by 32;
setting the final cutting specification to be 128 multiplied by 128 for the 16 th Crop layer;
setting the number of feature maps to be 2 for the 17 th layer Softmax classifier;
6b) the convolution kernel size of the second layer of convolutional layers is set to 5 x 5, reducing the receptive field.
And 7) taking the training data set characteristic matrix W1 as the input of the classification model, taking the category of each pixel point in the training data set D as the output of the classification model, solving the error between the category and the correct category marked by the manual marker, performing back propagation on the error, and optimizing the network parameters of the classification model to obtain the trained classification model.
And 8) inputting the test data set feature matrix W2 as a trained classification model, wherein the output result of the trained classification model is the classification category obtained by classifying each pixel point in the test data set T.
Compared with the prior art, the invention has the following beneficial effects: by expanding the image block features into pixel-level features, repeated storage and computational convolution due to the use of pixel blocks are avoided, and the speed and efficiency of classification are improved. Because multilayer non-downsampling contourlet transformation is introduced in front of the full convolution neural network, a low-frequency coefficient and a high-frequency coefficient are obtained, the low-frequency coefficient represents rough approximation to a target, namely basic information such as an area where the target is located, and the high-frequency coefficient can accurately acquire detailed information of the target, so that the low-frequency coefficient has classification discrimination capability compared with the high-frequency coefficient. The invention selects and fuses the low-frequency coefficient and the high-frequency coefficient, improves the classification accuracy, can accept input images of any size because the full-connection layer in the convolutional neural network is replaced by the deconvolution layer, and does not require that all training images and test images have the same size. In conclusion, the high-resolution SAR image classification method can improve the classification accuracy and the classification speed.
Drawings
FIG. 1 is a flow chart of the classification method of the present invention;
FIG. 2 is a diagram of the present invention of artificial labeling of images to be classified;
FIG. 3 is a diagram of the classification result of an image to be classified according to the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings.
Referring to fig. 1, the image classification method of the present invention is implemented as follows:
step 1, inputting a high-resolution SAR image to be classified, and performing 3-layer non-downsampling contourlet transformation on each pixel point to obtain high and low frequency coefficients of the high-resolution SAR image; the high-resolution SAR image to be classified is an X-band horizontal polarization map obtained by a German space navigation bureau (DLR) F-SAR aviation system in 2007, the resolution is 1m, and the image size is 6187 × 4278.
1a) Transforming the classification characteristics of each pixel point to obtain a transformation coefficient, wherein the transformation method comprises wavelet transformation, non-downsampling stationary wavelet transformation, curvelet transformation, non-downsampling contourlet transformation and the like;
1b) in the embodiment, non-downsampling contourlet transformation is adopted to carry out 3-layer transformation on each pixel point, wherein the non-downsampling contourlet transformation comprises non-downsampling pyramid (NSP) decomposition and a non-downsampling direction filter (NSDFB);
1c) non-downsampling pyramid (NSP) transform decomposes the time-frequency plane into one low frequency offspring and many ring-shaped high frequency offspring using a non-downsampling filter bank (NSFs);
1d) the non-downsampling direction filter (NSDFB) is a two-channel non-downsampling filter bank;
in the example, the image is subjected to 3-level NSP filtering to obtain coefficients of 1 low-pass image and 3 band-pass images;
after the image of the example is subjected to multi-scale decomposition by NSP, the band-pass image further completes 0, 1 and 3-level multi-directional decomposition of the image by NSDFB, so that coefficients of 1, 2 and 8 band-pass sub-images are obtained respectively.
And 2, selecting and fusing high and low frequency coefficients to form a characteristic matrix F based on the pixel points. In the embodiment, the high-frequency coefficients obtained by decomposition are sorted from large to small, and the first 50% of the high-frequency coefficients are selected to be fused with the low-frequency coefficients after the layer 3 transformation to be used as the transformation domain classification features. Defining a matrix with the size of M1 xM 2 x1, and assigning a fusion result to the matrix to obtain a characteristic F based on a pixel point, wherein M1 is the length of the SAR image to be classified, and M2 is the width of the SAR image to be classified.
And 3, normalizing the characteristic matrix F based on the pixel points.
Common normalization methods are: feature linear scaling, feature normalization and feature whitening.
In the embodiment, a characteristic linear scaling method is adopted, namely, the maximum value max (F) of a characteristic matrix F based on pixel points is firstly solved; and dividing each element in the characteristic matrix F based on the pixel points by the maximum value max (F) to obtain a normalized characteristic matrix F1.
And 4, performing blocking processing on the normalized feature matrix F1 according to the size of 128 multiplied by 128 and the interval of 50 to form a small feature block matrix F2 as sample data.
Step 5, constructing a training data set characteristic matrix W1 through a training data set D, and constructing a test data set characteristic matrix W2 through a test data set T; the method specifically comprises the following steps:
5a) dividing the SAR image surface features into 3 types, recording the position of a pixel point corresponding to each type in an image to be classified, and generating 3 positions A1, A2 and A3 corresponding to the pixel points of different types of surface features;
a1 corresponds to the position of the type 1 surface feature pixel point in the image to be classified, A2 corresponds to the position of the type 2 surface feature pixel point in the image to be classified, and A3 corresponds to the position of the type 3 surface feature pixel point in the image to be classified;
5b) randomly selecting 5% elements from the positions A1, A2 and A3 of the pixel points of the heterogeneous ground objects, and generating 3 positions B1, B2 and B3 which correspond to the positions A1, B2 and B3 of the pixel points of the heterogeneous ground objects selected as training data sets;
b1 is the position of the pixel point selected as the training data set in the image to be classified in the corresponding class 1 feature, B2 is the position of the pixel point selected as the training data set in the corresponding class 2 feature in the image to be classified, B3 is the position of the pixel point selected as the training data set in the corresponding class 3 feature in the image to be classified, and the elements in B1, B2 and B3 are combined to form the position L1 of all the pixel points of the training data set in the image to be classified;
5c) generating 3 positions C1, C2 and C3 corresponding to the pixel points of different types of ground objects selected as the test data set by using the remaining 95% of the elements in the A1, A2 and A3, wherein C1 is the position of the pixel point selected as the test data set in the 1 st type of ground object in the image to be classified, C2 is the position of the pixel point selected as the test data set in the 2 nd type of ground object in the image to be classified, C3 is the position of the pixel point selected as the test data set in the 3 rd type of ground object in the image to be classified, and combining the elements in the C1, C2 and C3 to form the position L2 of all the pixel points of the test data set in the image to be classified;
5d) defining a training data set characteristic matrix W1 of a training data set D, taking values at corresponding positions in an image block-based characteristic matrix F2 according to L1, and assigning the values to a training data set characteristic matrix W1;
5e) the test data set feature matrix W2 defining the test data set T takes values at corresponding positions in the feature block matrix F2 according to L2 and assigns values to the test data set feature matrix W2.
And 6, constructing a classification model based on the full convolution neural network.
6a) Selecting a 17-layer deep neural network consisting of an input layer → a convolutional layer → a pooling layer → a convolutional layer → a Dropout layer → a convolutional layer → an upsampling layer (deconvolution) → Crop layer → softmax classifier, and the parameters of each layer are as follows:
setting the number of feature maps to be 3 for the input layer of the layer 1;
for the 2 nd convolutional layer, setting the number of feature maps to be 32 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 3 rd pooling layer, the down-sampling size is set to 2;
for the 4 th convolutional layer, setting the number of feature maps to be 64 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 5 th pooling layer, the down-sampling size is set to 2;
for the convolution layer of the layer 6, setting the number of feature maps to be 96 and the size of a convolution kernel to be 3 multiplied by 3;
for the 7 th pooling layer, the downsampling size is set to 2;
for the 8 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the downsampling size to be 2 for the 9 th pooling layer;
for the 10 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the sparsity factor to 0.5 for the 11 th Dropout layer;
for the 12 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 13 th Dropout layer, setting the sparsity factor to 0.5;
for the 14 th convolutional layer, setting the number of feature maps to be 2 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 15 th upsampling layer, setting the number of feature maps to be 2 and the size of a convolution kernel to be 32 multiplied by 32;
setting the final cutting specification to be 128 multiplied by 128 for the 16 th Crop layer;
for the layer 17 Softmax classifier, the number of feature maps is set to 2.
6b) Setting the convolution kernel size of the second layer of convolution layer to be 5 multiplied by 5, and reducing the receptive field;
and 7, training the classification model by using the training data set to obtain the trained classification model.
The training data set characteristic matrix W1 is used as the input of a classification model, the category of each pixel point in the training data set D is used as the output of the classification model, the network parameters of the classification model are optimized by solving the error between the category and the correct category of the artificial marker and performing back propagation on the error, and the trained classification model is obtained, wherein the correct category of the artificial marker is shown in figure 2.
And 8, classifying the test data set by using the trained classification model.
And taking the test data set characteristic matrix W2 of the test data set T as the input of the trained classification model, and the output of the trained classification model is the classification category obtained by classifying each pixel point in the test data set.
The effect of the invention is further illustrated by the following simulation experiment:
1. simulation conditions are as follows:
the hardware platform is as follows: HPZ 840.
The software platform is as follows: caffe.
2. Simulation content and results:
the method of the invention is used for carrying out experiments under the simulation condition, namely, 5 percent of marked pixel points are randomly selected from SAR data as training samples, and the rest marked pixel points are used as test samples, so that the classification result shown in figure 3 is obtained.
As can be seen from fig. 3: the region consistency of the classification result is better, the edges of farmland, forest and town are clearer, and the detail information is kept.
Then, the training samples are sequentially reduced to make the training samples account for 4%, 3% and 2% of the total number of the samples, the classification precision of the test data set of the full convolution neural network is compared with the classification precision of the test data set of the full convolution neural network, and the result is shown in table 1:
TABLE 1
| Proportion of training sample | FCN-8 | The invention |
| 5％ | 94.0039％ | 94.3360％ |
| 4％ | 93.3933％ | 94.1524％ |
| 3％ | 92.6727％ | 93.3117％ |
| 2％ | 91.4413％ | 92.4162％ |
As can be seen from Table 1, when the training samples account for 5%, 4%, 3% and 2% of the total number of the samples, the classification accuracy of the test data set of the invention is higher than that of a pure full convolution neural network. In conclusion, the invention introduces non-subsampled contourlet transformation in the full convolution neural network, considers the direction information and the spatial information of the high-resolution SAR image, effectively improves the expression capability of the image characteristics, enhances the generalization capability of the model, and can still achieve high classification accuracy under the condition of less training samples.
Claims (8)
1. A high-resolution SAR image classification method based on a non-downsampling contourlet full convolution network is characterized by comprising the following steps:
1) inputting a high-resolution SAR image to be classified, and carrying out multilayer non-downsampling contourlet transformation on each pixel point in the image to obtain a low-frequency coefficient and a high-frequency coefficient of each pixel point;
2) selecting and fusing the low-frequency coefficient and the high-frequency coefficient to form a characteristic matrix F based on pixel points;
3) normalizing the element values in the feature matrix F to be between [0 and 1] to obtain a normalized feature matrix F1;
4) dicing the normalized feature matrix F1 to obtain a feature block matrix F2 as sample data;
5) constructing a training data set characteristic matrix W1 through a training data set D, and constructing a test data set characteristic matrix W2 through a test data set T; the specific operation is as follows:
5a) dividing the high-resolution SAR image surface features into 3 classes, recording the positions of pixel points corresponding to each class in the image to be classified, and generating three positions A1, A2 and A3 respectively representing the positions of the pixel points of the three classes of surface features in the image to be classified;
5b) randomly selecting 5% of elements from the A1, A2 and A3 to generate three pixel positions B1, B2 and B3 which correspond to different types of ground objects and are selected as training data sets, wherein B1 corresponds to the position of a pixel selected as a training data set in a1 st type of ground object in an image to be classified, B2 corresponds to the position of a pixel selected as a training data set in a2 nd type of ground object in the image to be classified, B3 corresponds to the position of a pixel selected as a training data set in A3 rd type of ground object in the image to be classified, and combining the elements from the B1, the B2 and the B3 to form a position L1 of all pixels of the training data set in the image to be classified;
5c) generating 3 pixel point positions C1, C2 and C3 corresponding to different types of ground objects and selected as test data sets by using the rest 95% of the elements in the A1, A2 and A3, wherein C1 is the position of a pixel point selected as a test data set in a1 st type of ground object in an image to be classified, C2 is the position of a pixel point selected as a test data set in a2 nd type of ground object in the image to be classified, C3 is the position of a pixel point selected as a test data set in A3 rd type of ground object in the image to be classified, and combining the elements in the C1, C2 and C3 to form the position L2 of all the pixel points of the test data set in the image to be classified;
5d) defining a training data set characteristic matrix W1 of a training data set D, taking values at corresponding positions in a characteristic block matrix F2 according to L1, and assigning the values to a training data set characteristic matrix W1 of the training data set D;
5e) defining a test data set characteristic matrix W2 of the test data set T, taking values at corresponding positions in a characteristic block matrix F2 according to L2, and assigning the values to a test data set characteristic matrix W2 of the test data set T;
6) constructing a classification model based on a full convolution neural network;
7) training the classification model through a training data set D to obtain a trained model;
8) classifying the test data set T by using the trained model to obtain the category of each pixel point in the test data set T, comparing the obtained category of each pixel point with the category label chart, and calculating the classification accuracy.
2. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network according to claim 1, wherein the method comprises the following steps: step 1) carrying out three-layer non-downsampling contourlet transformation on each pixel point in an image; the non-downsampling contourlet transformation comprises non-downsampling pyramid decomposition and non-downsampling direction filter decomposition, the non-downsampling pyramid decomposition decomposes a time-frequency plane into a low-frequency filial generation and a plurality of annular high-frequency filial generations through a non-downsampling filter bank, and a band-pass image formed by the non-downsampling pyramid decomposition is decomposed through the non-downsampling direction filter to obtain coefficients of the band-pass sub-images.
3. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network according to claim 2, wherein the method comprises the following steps: and 2) sequencing the high-frequency coefficients from large to small, selecting the high-frequency coefficients of the first 50% of the high-frequency coefficients, fusing the high-frequency coefficients with the low-frequency coefficients after the third layer of transformation, defining the size of a characteristic matrix F based on the pixel points as M1 multiplied by M2 multiplied by 1, defining the size of the characteristic matrix F based on the pixel points as M1 as the length of the SAR image to be classified and defining the size of the characteristic matrix F based on the pixel points as M2 as the width of the SAR image to be classified, and assigning the fusion.
4. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network according to claim 1, wherein the method comprises the following steps: the normalization in the step 3) is realized by a characteristic linear scaling method, a characteristic standardization method or a characteristic whitening method; firstly, solving the maximum value max (F) of a characteristic matrix F based on pixel points by a characteristic linear scaling method; and dividing each element in the characteristic matrix F based on the pixel points by the maximum value max (F) to obtain a normalized characteristic matrix F1.
5. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network according to claim 1, wherein the method comprises the following steps: step 4) the normalized feature matrix F1 is diced at intervals of 50 and size 128 x 128.
6. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network as claimed in claim 1, wherein the step 6) of constructing the classification model based on the full convolution neural network comprises the following steps:
6a) selecting a 17-layer deep neural network consisting of an input layer, a convolutional layer, a pooling layer, a convolutional layer, a Dropout layer, a convolutional layer, a deconvolution up-sampling layer, a Crop layer and a softmax classifier in sequence, wherein the parameters of each layer are as follows:
setting the number of feature maps to be 3 for the input layer of the layer 1;
for the 2 nd convolutional layer, setting the number of feature maps to be 32 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 3 rd pooling layer, the down-sampling size is set to 2;
for the 4 th convolutional layer, setting the number of feature maps to be 64 and the size of a convolutional kernel to be 5 multiplied by 5;
for the 5 th pooling layer, the down-sampling size is set to 2;
for the convolution layer of the layer 6, setting the number of feature maps to be 96 and the size of a convolution kernel to be 3 multiplied by 3;
for the 7 th pooling layer, the downsampling size is set to 2;
for the 8 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the downsampling size to be 2 for the 9 th pooling layer;
for the 10 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 3 multiplied by 3;
setting the sparsity factor to 0.5 for the 11 th Dropout layer;
for the 12 th convolutional layer, setting the number of feature maps to be 128 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 13 th Dropout layer, setting the sparsity factor to 0.5;
for the 14 th convolutional layer, setting the number of feature maps to be 2 and the size of a convolutional kernel to be 1 multiplied by 1;
for the 15 th deconvolution up-sampling layer, setting the number of feature maps to be 2 and the convolution kernel size to be 32 multiplied by 32;
setting the final cutting specification to be 128 multiplied by 128 for the 16 th Crop layer;
setting the number of feature maps to be 2 for the 17 th layer Softmax classifier;
6b) the convolution kernel size of the second layer of convolutional layers is set to 5 x 5, reducing the receptive field.
7. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network as claimed in claim 1, wherein the step 7) is to use a training data set feature matrix W1 as an input of a classification model, use the category of each pixel point in a training data set D as an output of the classification model, solve an error between the category and a correct category of an artificial marker, perform back propagation on the error, and optimize the network parameters of the classification model to obtain the trained classification model.
8. The method for classifying the high-resolution SAR image based on the non-subsampled contourlet full convolution network as claimed in claim 7, wherein the step 8) is to input a test data set feature matrix W2 as a trained classification model, and an output result of the trained classification model is a classification category obtained by classifying each pixel point in the test data set T.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710364900.8A CN107239751B (en) | 2017-05-22 | 2017-05-22 | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710364900.8A CN107239751B (en) | 2017-05-22 | 2017-05-22 | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107239751A CN107239751A (en) | 2017-10-10 |
| CN107239751B true CN107239751B (en) | 2020-11-03 |
Family
ID=59984361
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710364900.8A Active CN107239751B (en) | 2017-05-22 | 2017-05-22 | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107239751B (en) |
Families Citing this family (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107944470A (en) * | 2017-11-03 | 2018-04-20 | 西安电子科技大学 | SAR image sorting technique based on profile ripple FCN CRF |
| CN107944353B (en) * | 2017-11-10 | 2019-12-24 | 西安电子科技大学 | SAR image change detection method based on contour wave BSPP network |
| CN107832798B (en) * | 2017-11-20 | 2020-04-14 | 西安电子科技大学 | Polarized SAR image target detection method based on NSCT ladder network model |
| CN109886992A (en) * | 2017-12-06 | 2019-06-14 | 深圳博脑医疗科技有限公司 | For dividing the full convolutional network model training method in abnormal signal area in MRI image |
| CN108062575A (en) * | 2018-01-03 | 2018-05-22 | 广东电子工业研究院有限公司 | A kind of high similarity graph picture identification and sorting technique |
| CN108492319B (en) * | 2018-03-09 | 2021-09-03 | 西安电子科技大学 | Moving target detection method based on deep full convolution neural network |
| CN109447124B (en) * | 2018-09-28 | 2019-11-19 | 北京达佳互联信息技术有限公司 | Image classification method, device, electronic equipment and storage medium |
| CN109344898A (en) * | 2018-09-30 | 2019-02-15 | 北京工业大学 | Convolutional neural networks image classification method based on sparse coding pre-training |
| CN109444667B (en) * | 2018-12-17 | 2021-02-19 | 国网山东省电力公司电力科学研究院 | Power distribution network early fault classification method and device based on convolutional neural network |
| CN109903301B (en) * | 2019-01-28 | 2021-04-13 | 杭州电子科技大学 | Image contour detection method based on multistage characteristic channel optimization coding |
| CN110097129B (en) * | 2019-05-05 | 2023-04-28 | 西安电子科技大学 | Remote sensing target detection method based on profile wave grouping characteristic pyramid convolution |
| CN110188774B (en) * | 2019-05-27 | 2022-12-02 | 昆明理工大学 | Eddy current scanning image classification and identification method based on deep learning |
| CN110702648B (en) * | 2019-09-09 | 2020-11-13 | 浙江大学 | Fluorescent spectrum pollutant classification method based on non-subsampled contourlet transformation |
| CN111899232B (en) * | 2020-07-20 | 2023-07-04 | 广西大学 | Method for nondestructive detection of bamboo-wood composite container bottom plate by image processing |
| CN113139579B (en) * | 2021-03-23 | 2024-02-02 | 广东省科学院智能制造研究所 | Image classification method and system based on image feature self-adaptive convolution network |
| CN114037747B (en) * | 2021-11-25 | 2024-06-21 | 佛山技研智联科技有限公司 | Image feature extraction method, device, computer equipment and storage medium |
| CN115310482A (en) * | 2022-07-31 | 2022-11-08 | 西南交通大学 | Radar intelligent identification method for bridge reinforcing steel bar |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915676A (en) * | 2015-05-19 | 2015-09-16 | 西安电子科技大学 | Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method |
| CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
| CN105718957A (en) * | 2016-01-26 | 2016-06-29 | 西安电子科技大学 | Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US7817082B2 (en) * | 2007-03-11 | 2010-10-19 | Vawd Applied Science And Technology Corporation | Multi frequency spectral imaging radar system and method of target classification |
-
2017
- 2017-05-22 CN CN201710364900.8A patent/CN107239751B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104915676A (en) * | 2015-05-19 | 2015-09-16 | 西安电子科技大学 | Deep-level feature learning and watershed-based synthetic aperture radar (SAR) image classification method |
| CN105512680A (en) * | 2015-12-02 | 2016-04-20 | 北京航空航天大学 | Multi-view SAR image target recognition method based on depth neural network |
| CN105718957A (en) * | 2016-01-26 | 2016-06-29 | 西安电子科技大学 | Polarized SAR image classification method based on nonsubsampled contourlet convolutional neural network |
Non-Patent Citations (3)
| Title |
|---|
| SAR and Infrared Image Fusion Using Nonsubsampled Contourlet Transform;Ying Zhang等;《IEEE》;20090707;第398-401页 * |
| 基于深度脊波神经网络的极化SAR影像地物分类;张亚楠;《中国优秀硕士学位论文全文数据库信息科技辑》;20170315;第2017年卷(第3期);第7-62页 * |
| 基于非下采样轮廓波的图像检索;陈丽燕等;《福州大学学报（自然科学版）》;20120430;第40卷(第2期);第172-216页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107239751A (en) | 2017-10-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107239751B (en) | High-resolution SAR image classification method based on non-subsampled contourlet full convolution network | |
| CN108230329B (en) | Semantic segmentation method based on multi-scale convolution neural network | |
| CN109584248B (en) | Infrared target instance segmentation method based on feature fusion and dense connection network | |
| CN108427920B (en) | Edge-sea defense target detection method based on deep learning | |
| CN108573276B (en) | Change detection method based on high-resolution remote sensing image | |
| CN108154192B (en) | High-resolution SAR terrain classification method based on multi-scale convolution and feature fusion | |
| CN112488210A (en) | Three-dimensional point cloud automatic classification method based on graph convolution neural network | |
| CN111310666B (en) | High-resolution image ground feature identification and segmentation method based on texture features | |
| CN107358258B (en) | SAR image target classification based on NSCT double CNN channels and selective attention mechanism | |
| CN107016677A (en) | A kind of cloud atlas dividing method based on FCN and CNN | |
| CN107944353B (en) | SAR image change detection method based on contour wave BSPP network | |
| CN107358203B (en) | A kind of High Resolution SAR image classification method based on depth convolution ladder network | |
| CN104299232B (en) | SAR image segmentation method based on self-adaptive window directionlet domain and improved FCM | |
| CN111027497B (en) | Weak and small target rapid detection method based on high-resolution optical remote sensing image | |
| CN113344110B (en) | Fuzzy image classification method based on super-resolution reconstruction | |
| CN107133653B (en) | High-resolution SAR image classification method based on deep ladder network | |
| CN110070545B (en) | Method for automatically extracting urban built-up area by urban texture feature density | |
| CN110598564A (en) | OpenStreetMap-based high-spatial-resolution remote sensing image transfer learning classification method | |
| CN113239736A (en) | Land cover classification annotation graph obtaining method, storage medium and system based on multi-source remote sensing data | |
| CN114119621A (en) | SAR remote sensing image water area segmentation method based on depth coding and decoding fusion network | |
| Al-Ghrairi et al. | Classification of satellite images based on color features using remote sensing | |
| CN113239829B (en) | Cross-dimension remote sensing data target identification method based on space occupation probability characteristics | |
| CN113361407B (en) | PCANet-based spatial spectrum feature combined hyperspectral sea ice image classification method | |
| CN114612315A (en) | High-resolution image missing region reconstruction method based on multi-task learning | |
| Kazimi et al. | Semantic segmentation of manmade landscape structures in digital terrain models |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |