CN112906300A - Polarized SAR (synthetic Aperture Radar) soil humidity inversion method based on two-channel convolutional neural network - Google Patents
Polarized SAR (synthetic Aperture Radar) soil humidity inversion method based on two-channel convolutional neural network Download PDFInfo
- Publication number
- CN112906300A CN112906300A CN202110182168.9A CN202110182168A CN112906300A CN 112906300 A CN112906300 A CN 112906300A CN 202110182168 A CN202110182168 A CN 202110182168A CN 112906300 A CN112906300 A CN 112906300A
- Authority
- CN
- China
- Prior art keywords
- polarization
- convolution
- matrix
- neural network
- soil humidity
- 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.)
- Granted
Links
- 239000002689 soil Substances 0.000 title claims abstract description 49
- 238000013527 convolutional neural network Methods 0.000 title claims abstract description 25
- 238000000034 method Methods 0.000 title claims abstract description 23
- 238000013528 artificial neural network Methods 0.000 claims abstract description 23
- 230000010287 polarization Effects 0.000 claims description 48
- 239000011159 matrix material Substances 0.000 claims description 32
- 238000012549 training Methods 0.000 claims description 20
- 210000002569 neuron Anatomy 0.000 claims description 17
- 239000013598 vector Substances 0.000 claims description 15
- 238000012795 verification Methods 0.000 claims description 11
- 230000004913 activation Effects 0.000 claims description 10
- 230000006870 function Effects 0.000 claims description 10
- 238000000605 extraction Methods 0.000 claims description 6
- 230000009977 dual effect Effects 0.000 claims description 5
- 230000004927 fusion Effects 0.000 claims description 5
- 238000004519 manufacturing process Methods 0.000 claims description 5
- 238000010586 diagram Methods 0.000 claims description 4
- 238000011049 filling Methods 0.000 claims description 4
- 230000000694 effects Effects 0.000 claims description 3
- 230000007246 mechanism Effects 0.000 claims description 3
- 150000001875 compounds Chemical class 0.000 claims description 2
- 238000001914 filtration Methods 0.000 claims description 2
- 238000005457 optimization Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 claims description 2
- 238000012545 processing Methods 0.000 claims description 2
- 230000009466 transformation Effects 0.000 claims description 2
- NAWXUBYGYWOOIX-SFHVURJKSA-N (2s)-2-[[4-[2-(2,4-diaminoquinazolin-6-yl)ethyl]benzoyl]amino]-4-methylidenepentanedioic acid Chemical compound C1=CC2=NC(N)=NC(N)=C2C=C1CCC1=CC=C(C(=O)N[C@@H](CC(=C)C(O)=O)C(O)=O)C=C1 NAWXUBYGYWOOIX-SFHVURJKSA-N 0.000 claims 1
- 238000005388 cross polarization Methods 0.000 claims 1
- 238000013507 mapping Methods 0.000 claims 1
- 238000010606 normalization Methods 0.000 claims 1
- 230000000644 propagated effect Effects 0.000 claims 1
- 238000002474 experimental method Methods 0.000 abstract description 2
- 239000000523 sample Substances 0.000 description 13
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 4
- 230000008859 change Effects 0.000 description 2
- 238000013135 deep learning Methods 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 230000008447 perception Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000003044 adaptive effect Effects 0.000 description 1
- 238000004458 analytical method Methods 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 210000004027 cell Anatomy 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 239000011362 coarse particle Substances 0.000 description 1
- 230000001054 cortical effect Effects 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001704 evaporation Methods 0.000 description 1
- 230000008020 evaporation Effects 0.000 description 1
- 239000010419 fine particle Substances 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000008595 infiltration Effects 0.000 description 1
- 238000001764 infiltration Methods 0.000 description 1
- 230000002262 irrigation Effects 0.000 description 1
- 238000003973 irrigation Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000000691 measurement method Methods 0.000 description 1
- 238000001556 precipitation Methods 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000035945 sensitivity Effects 0.000 description 1
- 238000000638 solvent extraction Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
- G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Evolutionary Computation (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Geometry (AREA)
- Medical Informatics (AREA)
- Computer Hardware Design (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Radar Systems Or Details Thereof (AREA)
Abstract
The invention discloses a polarized SAR soil humidity inversion method based on a dual-channel convolutional neural network, which aims at the problems that the soil humidity inversion accuracy is not high and sample data is limited by utilizing a polarized SAR image. The main implementation object of the invention is a fully-polarized SAR image acquired by a synthetic aperture radar, and the main work is to estimate the soil humidity. Experiments show that compared with the traditional neural network, the inversion accuracy is improved by 10.88% by using the dual-channel convolution neural network, the root mean square error is reduced by 3.2356, and the coefficient of solution is improved by 0.6633.
Description
Technical Field
The invention relates to a polarized SAR soil humidity inversion method based on a two-channel convolutional neural network, and belongs to the field of polarized synthetic aperture radar quantitative inversion.
Background
The high-resolution polarimetric synthetic aperture radar (PolSAR) data has an important role in the inversion of soil moisture, the surface soil moisture is a basic condition for the growth and development of crops, and the soil moisture is an important index for drought monitoring before the crops are irrigated; after crops are irrigated, the change condition of soil moisture is an important basis for evaluating the irrigation effect. Soil moisture is also a very important variable in studying land water circulation and energy circulation. It not only can influence net radiant energy conversion and become latent heat and sensible heat's distribution proportion, can also influence precipitation and change into the proportion of infiltration, runoff and evaporation, consequently accurate acquisition soil water content can carry out reasonable utilization to the soil, can improve production level and production quality. The existing method for accurately measuring the water content of soil adopts traditional methods such as a probe and the like. The traditional methods can accurately measure the local soil water content, but consume a large amount of manpower and material resources, and are not suitable for extracting the soil humidity in a large scale. By virtue of high sensitivity of PolSAR to soil humidity, the PolSAR can make up the defects of the traditional measurement method, becomes a new method and means for monitoring the soil humidity, and has been used for estimating the soil humidity by utilizing SAR data generation theory, experience and semi-experience backscattering models. The theoretical model includes a Physical Optics (PO) model, a metrology optics (GO) model, an Integral Equation Model (IEM), and the like.
With the rise of the neural network, modeling between the polarization parameters and the soil humidity by using the neural network becomes an important means for soil humidity inversion, but the traditional neural network only uses the polarization parameters of the sample points. In recent years, deep learning is the most fierce technology in the current era, and a Convolutional Neural Network (CNN) is an important branch of deep learning, and is widely applied in the image field due to its excellent feature extraction capability. The method applies the convolutional neural network to the polarized SAR soil humidity inversion, and adopts the dual-channel convolutional neural network to fit the nonlinear relation between the polarization parameters and the soil humidity, so that the inversion accuracy is improved.
Disclosure of Invention
The invention mainly aims to solve the problems of low accuracy and limited samples when soil humidity inversion is carried out by utilizing a polarized SAR image, provides a polarized SAR soil humidity inversion method based on a dual-channel convolutional neural network, and designs a coarse and fine particle size inversion system applied to various scenes based on the neural network. The method uses a dual-channel convolutional neural network, fully utilizes the spatial characteristics of the polarized SAR image, improves the inversion accuracy while reducing training samples, introduces Dropout into a convolutional layer and a full-link layer, improves the robustness of the method, and designs a coarse-grained classification network and a fine-grained regression network through the network. The main implementation object of the invention is a fully-polarized SAR image acquired by a synthetic aperture radar, and the main work is to estimate the soil humidity. The experimental result shows that compared with the traditional neural network, the average inversion precision is improved by 10.88%, the root mean square error is reduced by 3.2356, and the coefficient of solution is improved by 0.6633.
The technical scheme of the invention specifically comprises the following contents:
1. extracting polarization parameters: carrying out refined Lee filtering processing with the window size of 7 x 7 on the PolSAR image, extracting 6 polarization parameters, and applying the polarization parameters to soil humidity inversion
2. Manufacturing a training sample and a verification sample: partitioning of samples using 7-by-7 sliding windows for polarization parameters
3. Establishing a two-channel convolutional neural network and training: the method constructs a dual-channel convolutional neural network, divides 6 polarization parameters in each training sample into two groups, respectively sends the two groups of polarization parameters into the convolutional neural networks of the two channels for training, and generates a prediction model after fusing the characteristics of the two channels after extracting the characteristics.
4. And predicting the verification sample: and predicting the verification sample through the trained two-channel convolutional neural network, and calculating the average accuracy of the classification network and the root mean square error and the coefficient of decision of the regression network.
Drawings
FIG. 1 illustrates a schematic diagram of training and validation sample preparation.
FIG. 2 is a diagram of a two-channel convolutional neural network architecture.
FIG. 3 is a general flow chart of SAR soil humidity inversion based on dual-channel convolution neural network polarization.
Detailed Description
The present invention will be described in detail below with reference to the accompanying drawings and examples.
A polarized SAR soil humidity inversion method based on a dual-channel convolution neural network comprises the following steps,
step 1, extracting polarization parameters for the PolSAR image:
for fully polarized data, the scattering matrix for the target is:
wherein SHH、SVVIs the echo power of the co-polarized channel, SHV、SVHIs the echo power of the cross-polarized channel. When the reciprocity theorem is satisfied, SHV＝SVH。
By [ S ]]The matrix can obtain the polarization covariance C of the target point3And (4) matrix.
Selecting C3Elements in the matrix:<|SHH|2>、<|SVV|2>as the first two polarization parameters, wherein let δHH＝<|SHH|2>、δVV＝<|SVV|2>And calculating the homopolarity ratio deltaHH/δVVAs a third polarization parameter.
Matrix transformation of the polarization covariance matrix to a polarization coherence matrix T3：
T3＝U3C3U3 -1 (3)
For T3And (3) decomposing the eigenvalue and the eigenvector by the matrix:
in the formula (I), the compound is shown in the specification,
is T3Eigenvectors of the matrix, λiIs T3Eigenvalues of the matrix. By normalizing the absolute scattering amplitude of the characteristic value, the scattering probability p can be obtainedi：
Using probability piAnd a characteristic value lambdaiThe fourth and fifth polarization parameters used can be obtained: the entropy of scattering (H) and the degree of anisotropy (a), which are defined as:
the scattering entropy (H) represents the degree of scattering polarization of the object point, and the degree of anisotropy (a) defines a relationship between the second and third eigenvalues.
The sixth polarization parameter can be obtained through the eigenvector through the target scattering mechanism alphaiAn average scattering angle α can be obtained, which is defined as:
alpha may represent the type of target point mean scattering process.
The extracted 6 polarization parameters have a certain nonlinear relation with the soil humidity, so that the soil humidity is inverted by utilizing the polarization parameters.
Step 2, manufacturing of training samples and verification samples:
after extracting characteristic parameters of experimental data, a matrix of 6 characteristic parameters can be obtained and is expressed according to H, A, alpha and deltaHH、δVV、δHH/δVVM x n x 6, where m, n are the width and height of the matrix and 6 is the depth. Then, as shown in fig. 1, a 7 × 7 × 6 sliding window is used to cross the large matrix pixel by pixel, the matrix extracted by each window is a sample, and the soil humidity corresponding to the center point of the sample is the label of the point. 6000 samples are randomly selected from each type of humidity samples, 10% of the samples are randomly selected as training samples, and 90% of the samples are selected as verification samples.
Step 3, establishing a double-channel convolution neural network and training:
the convolutional neural network is inspired by the visual imaging principle, different convolutional kernels are used for replacing the perception of visual cortical cells on a target object, and then the obtained perception result is processed by a nonlinear activation function so as to obtain the complex characteristics of the target object. The two-channel convolutional neural network designed by the invention is shown in figure 2. The 6 polarization parameters of the sample are divided into two groups and respectively placed into two channels, wherein for the three polarization parameters of H, A and alpha input into the first channel, 4 convolutional layers and a full-connection layer are selected. Second channel input deltaHH、δVV、δHH/δVVTwo convolutional layers and one full-link layer are selected as the three parameters. The essence of a convolutional layer (conv layer) is a filter that can perform local feature extraction on input data, where the input data I is convolved with a set of preset filters W and then added with an offset b. And finally, the convolution addition result is subjected to a nonlinear activation function, as shown in a formula (23), so that the local feature F can be obtained.
Where f (-) is a nonlinear activation function to a modified linear unit (ReLU) that avoids gradient dissipation well and reduces trainingExercise time[11]It is defined as:
f(xij)＝max{0,xij} (10)
the Full Connection Layer (FCL) maps the feature map after several convolution operations to a sample mark space, so that the influence of feature positions on classification can be greatly reduced. The Dropout layer is mixed between every two layers, and the Dropout layer has the function of stopping the activation value of a certain neuron with a certain probability p when the activation value propagates forwards, so that the overfitting of the model can be reduced, and the generalization is stronger. A corrected linear activation function (ReLU) is inserted behind each layer. And performing feature fusion on the two channels after feature extraction, accessing the fused features into a full connection layer, and finally classifying and regressing.
Specific setup details of the present invention are shown in table 1 and table 2, and the present invention generates 8 feature maps with size 7 × 7 by filling convolutional layers with 1 for H, A, α channel through a first set of convolution kernels with 3, step size 1, and filling 1 is to expand 0 around original data, which is done for the purpose of fully utilizing edge information of data and for keeping the size of data unchanged after convolution operation. And then passing through a second set of convolution kernels with a convolution kernel of 3 and a step size of 1 to generate 16 feature maps with a size of 5 x 5. Then, the convolution layers with step size of 1 are filled by a third group of convolution kernels with step size of 3 to generate 24 characteristic graphs with size of 5 x 5. And generating 32 characteristic graphs with the size of 5 x 5 through the convolution layers with the fourth set of convolution kernels of 3 and the step size of 1. Finally, 32 feature maps of size 3 x 3 are transformed into a one-dimensional vector with 120 neurons via a fully connected layer. For deltaHH、δVV、δHH/δVVThe channel first passes through a first set of convolution kernels with 3 and step size 1 convolution layers to generate 8 feature maps with size 5 x 5. And then passing through a second set of convolution kernels with the convolution kernel size of 3 and the step size of 1 to generate 16 characteristic graphs with the size of 3 x 3. Finally, the 24 3 × 3 feature maps are transformed into one-dimensional vectors with 120 neurons through a fully connected layer. After each operation described above, a Dropout layer is switched in and the ratio is set to 0.2. After two branches pass through the full connecting layerAnd obtaining a one-dimensional feature vector with 120 neurons extracted from respective data, and connecting and fusing the two one-dimensional feature vectors to obtain a one-dimensional fusion feature vector with 240 neurons. And rearranging the connected features of the 240 neurons through a full connection layer to obtain a new feature vector which is fused and has 84 neurons more in line with the requirement. When performing the classification task, the output passes through a fully connected layer and outputs a vector with C neurons, where C is the number of classes. When performing the regression task, the output goes through two fully connected layers, the first connecting layer outputs a vector with 32 neurons, and the second connecting layer outputs the regression value. Training 100 epochs in total during training, storing the model with the highest accuracy, and adopting an Adma optimization algorithm during training[26]Each batch size has 256 samples, the initial learning rate is set to be 0.001, the algorithm optimizes a random objective function based on first-order gradient through estimation of adaptive moment of low order, the calculation is efficient, the memory requirement is low, and the automatic adjustment of the learning rate can be realized without adjusting hyper-parameters.
TABLE 1 two-channel classification convolutional neural network structure parameter table
TABLE 2 two-channel regression convolution neural network structure parameter table
Step 4, predicting the verification sample:
in the present invention, three indexes are used to evaluate the effect
1) Inversion accuracy
2) Root mean square error
3) Coefficient of determinability
Where N is the total number of verification samples, yp tam e i tFor inverting the output value, yTam ealIs the true value of the soil humidity,
the soil moisture is the mean value of the true value of soil moisture.
The invention uses E-SAR complete polarization data in DEMM area in North Germany to train and verify, and calculates the three indexes mentioned above. The results of the experiments are shown in tables 2 and 3. From the analysis of experimental results, the inversion accuracy of the dual-channel convolutional neural network is 99.42%, the inversion accuracy of the traditional neural network is 88.54%, and the dual-channel convolutional neural network is improved by 10.88% in comparison with the traditional neural network. It can be shown that the inversion accuracy of the method provided by the invention is obviously improved. The root mean square error is reduced by 3.2356 compared with the traditional neural network, and the coefficient of block is improved by 0.6633 compared with the traditional neural network.
TABLE 2 inversion accuracy for different networks
| Network type | Inversion accuracy |
| Dual channel convolutional neural network | 99.42％ |
| Traditional neural networks | 88.54％ |
TABLE 3 different net RMS errors and coefficients
| Network type | Root mean square error | Coefficient of determinability |
| Dual channel convolutional neural network | 0.4194 | 0.9912 |
| Traditional neural networks | 3.6550 | 0.3279 |
Claims (5)
1. A SAR soil humidity inversion method based on dual-channel convolution neural network polarization is characterized by comprising the following steps: the method comprises the following steps:
extracting polarization parameters: lee filtering processing with the window size of 7 × 7 is carried out on the PolSAR image, 6 polarization parameters are extracted, and the polarization parameters are applied to soil humidity inversion;
and (2) manufacturing a training sample and a verification sample: dividing samples by extracting 6 polarization parameters in the step (1) by adopting a 7-by-7 sliding window;
step (3), establishing a dual-channel convolution neural network and training: constructing a dual-channel convolutional neural network, dividing 6 polarization parameters in each training sample in the step (2) into two groups, respectively sending the two groups of polarization parameters into the convolutional neural networks of the two channels for training, fusing the characteristics of the two channels after characteristic extraction, and generating a prediction model;
predicting the verification sample: and predicting the verification sample through the trained two-channel convolutional neural network, and calculating the average accuracy, the root mean square error and the coefficient.
2. The SAR soil humidity inversion method based on dual convolution neural network polarization is characterized in that: in step (1), for the fully polarized data, the scattering matrix of the target is:
wherein SHH、SVVIs the echo power of the co-polarized channel of the polarized SAR data used for soil moisture inversion, SHV、SVHThe echo power of a polarized SAR data cross polarization channel used for soil humidity inversion; when the reciprocity theorem is satisfied, SHV＝SVH；
By [ S ]]Obtaining the polarization covariance C of the target point by matrix3A matrix;
at C3In the matrix, represents a conjugate operation, wherein
Is SHHThe complex number of the conjugate of (a),
is SVVThe conjugate complex number of (a); selecting C3Elements in the matrix:<|SHH|2>、<|SVV|2>as the first two polarization parameters, let δHH＝<|SHH|2>、δVV＝<|SVV|2>And calculating the homopolarity ratio deltaHH/δVVAs a third polarization parameter;
matrix transformation of the polarization covariance matrix to a polarization coherence matrix T3：
T3＝U3C3U3 -1 (3)
Wherein C is3As a polarization covariance matrix, U3Is a unitary matrix, which is defined as:
for T3And (3) decomposing the eigenvalue and the eigenvector by the matrix:
in the formula (I), the compound is shown in the specification,
is T3Eigenvectors of the matrix, whereiDenotes the scattering mechanism of the target, betaiIs the target azimuth angle phii、δiAnd gammaiIs a target phase angle; i, taking 1, 2 and 3 to represent parameters corresponding to the characteristic values; lambda [ alpha ]iIs T3Eigenvalues of the matrix and satisfy λ1≥λ2≥λp；
Obtaining the scattering probability p by carrying out absolute scattering amplitude normalization on the characteristic valuei：
Using probability piAnd a characteristic value lambdaiThe fourth and fifth polarization parameters used were obtained: the scattering entropy H and the degree of anisotropy a, defined as:
the scattering entropy H represents the scattering polarization degree of the target point, and the anisotropy A defines the relation between the second characteristic value and the third characteristic value;
the sixth polarization parameter is obtained through the eigenvector and is obtained through the target scattering mechanism alphaiThe average scattering angle α, defined as:
α represents the type of target point mean scattering process; the extracted 6 polarization parameters have a certain nonlinear relation with the soil humidity, so that the soil humidity is inverted by utilizing the polarization parameters.
3. The base of claim 1The SAR soil humidity inversion method based on double convolution neural network polarization is characterized by comprising the following steps: in the step (2), after extracting characteristic parameters of the experimental data, obtaining a matrix of 6 characteristic parameters according to H, A, alpha and deltaHH、δVV、δHH/δVVThe order m × n × 6 large matrix of (a), where m, n are the width and height of the matrix, and 6 is the depth; and then, a 7 multiplied by 6 sliding window is adopted to draw a large matrix pixel by pixel, the matrix extracted by each window is a sample, and the soil humidity corresponding to the center point of the sample is the label of the point.
4. The SAR soil humidity inversion method based on dual convolution neural network polarization is characterized in that: in the step (3), the 6 polarization parameters of the sample are divided into two groups and respectively placed into two channels, wherein for the three polarization parameters of H, A and alpha input into the first channel, 4 convolutional layers and one full-connection layer are selected; second channel input deltaHH、δVV、δHH/δVVSelecting two convolution layers and a full connection layer according to the three parameters; the essence of a convolutional layer convlayer is a filter that performs local feature extraction on the input data, i.e., in the convolutional layer, the input data IiAnd a set of preset filters WijConvolution is performed and then with an offset bjAdding; finally, the convolution addition result is processed through a nonlinear activation function, as shown in a formula (10), and then local characteristics are obtained; wherein FjIs the jth characteristic diagram output after convolution, and M is the channel number of input data;
where f (-) is a nonlinear activation function to modify the linear unit ReLU, which avoids gradient dissipation and reduces training time, defined as:
f(x)＝max{0,x} (11)
wherein x represents a feature map obtained after convolution;
the full connection layer FCL is used for mapping the characteristic diagram after a plurality of convolution operations to a sample mark space, and is mixed with one Dropout layer between every two layers, and the Dropout layer has the function of stopping the activation value of a certain neuron from working with a certain probability p when the Dropout layer is propagated forwards, so that overfitting of a model is reduced, and the generalization is stronger; a corrected linear activation function ReLU is connected behind each layer; performing feature extraction on the two channels, performing feature fusion on the channels, accessing the fused features into a full connection layer, and finally selecting a softmax classifier for classification on an output layer;
for H, A, an alpha channel firstly passes through a first group of convolution kernels with the step size of 3 and the step size of 1 and is filled with convolution layers with the step size of 1, 8 characteristic graphs with the size of 7 × 7 are generated, wherein the filling with the step size of 1 is to expand 0 around original data for one circle, and the purpose of doing so is to fully utilize edge information of the data and keep the size of the data unchanged after the convolution operation; generating 16 characteristic graphs with the size of 5 x 5 through a convolution layer with a second group of convolution kernels of 3 and the step length of 1; then, filling the convolution layers with 1 by a third group of convolution kernels with the step length of 3 and the step length of 1 to generate 24 characteristic graphs with the size of 5 x 5; generating 32 characteristic graphs with the size of 5 x 5 through a convolution layer with a fourth group of convolution kernels of 3 and the step length of 1; finally, changing 32 characteristic graphs with the size of 3 x 3 into one-dimensional vectors with 120 neurons through a full connection layer; for deltaHH、δVV、δHH/δVVThe channel firstly passes through a first group of convolution kernels with the convolution kernel size of 3 and the step size of 1 to generate 8 characteristic graphs with the size of 5 x 5; generating 16 characteristic graphs with the size of 3 x 3 through a second group of convolution kernels with the step size of 1; finally, changing 24 characteristic maps of 3 x 3 into one-dimensional vectors with 120 neurons through a full connection layer; after each step of operation, a Dropout layer is accessed, and the ratio is set to be 0.2; the two branches pass through a full connection layer to obtain a one-dimensional feature vector with 120 neurons extracted from respective data, and the two one-dimensional vectors are connected and fused to obtain a one-dimensional fusion feature vector with 240 neurons; the obtained 240 neurons pass through a full connection layer, and the characteristics of the connections are rearrangedObtaining a new feature vector which meets the requirement after fusion and has 84 neurons, and finally outputting a vector with C neurons by an output layer through a full connection layer, wherein C is the number of categories; and training 100 epochs in total during training, storing the model with the highest accuracy, and adopting an Adma optimization algorithm during training.
5. The SAR soil humidity inversion method based on dual convolution neural network polarization is characterized in that: in the step (4), three indexes are adopted to evaluate the effect; wherein the inversion accuracy IA is used for describing the accuracy of classification and is defined as an equation (11), and the deviation between the predicted value and the true value is expressed as a root mean square error RMSE and is defined as an equation (12); coefficient of coefficient r2The fitting degree of the model is judged by the formulas (13) to (15);
1) inversion accuracy
2) Root mean square error
3) Coefficient of determinability
Where N is the total number of verification samples,
in order to invert the output values,
is the true value of the soil humidity,
the soil moisture is the mean value of the true value of soil moisture.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182168.9A CN112906300B (en) | 2021-02-09 | 2021-02-09 | Polarization SAR soil humidity inversion method based on double-channel convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110182168.9A CN112906300B (en) | 2021-02-09 | 2021-02-09 | Polarization SAR soil humidity inversion method based on double-channel convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112906300A true CN112906300A (en) | 2021-06-04 |
| CN112906300B CN112906300B (en) | 2023-11-21 |
Family
ID=76123348
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110182168.9A Active CN112906300B (en) | 2021-02-09 | 2021-02-09 | Polarization SAR soil humidity inversion method based on double-channel convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112906300B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113406584A (en) * | 2021-06-22 | 2021-09-17 | 北京化工大学 | Cross calibration method for synthetic aperture radar |
| CN113744237A (en) * | 2021-08-31 | 2021-12-03 | 华中科技大学 | Deep learning-based automatic detection method and system for muck fluidity |
| CN114154886A (en) * | 2021-12-08 | 2022-03-08 | 重庆大学 | Land transition simulation method based on delay communication CA model |
| CN116821694A (en) * | 2023-08-30 | 2023-09-29 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN117114627A (en) * | 2023-10-18 | 2023-11-24 | 日照市自然资源和规划局 | land resource management system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006086566A2 (en) * | 2005-02-09 | 2006-08-17 | Inlight Solutions, Inc. | Methods and apparatuses for noninvasive determinations of analytes |
| CN105930664A (en) * | 2016-04-26 | 2016-09-07 | 中国科学院地理科学与资源研究所 | Method for estimating instantaneous surface emissivity from passive microwave data |
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN108564006A (en) * | 2018-03-26 | 2018-09-21 | 西安电子科技大学 | Based on the polarization SAR terrain classification method from step study convolutional neural networks |
-
2021
- 2021-02-09 CN CN202110182168.9A patent/CN112906300B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2006086566A2 (en) * | 2005-02-09 | 2006-08-17 | Inlight Solutions, Inc. | Methods and apparatuses for noninvasive determinations of analytes |
| CN105930664A (en) * | 2016-04-26 | 2016-09-07 | 中国科学院地理科学与资源研究所 | Method for estimating instantaneous surface emissivity from passive microwave data |
| WO2018028255A1 (en) * | 2016-08-11 | 2018-02-15 | 深圳市未来媒体技术研究院 | Image saliency detection method based on adversarial network |
| CN108564006A (en) * | 2018-03-26 | 2018-09-21 | 西安电子科技大学 | Based on the polarization SAR terrain classification method from step study convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| 杨磊;吴秋兰;张波;梁勇;洪学宝;邹文博;: "SVRM辅助的北斗GEO卫星反射信号土壤湿度反演方法", 北京航空航天大学学报, no. 06 * |
| 谢凯浪;赵泉华;李玉;: "结合FCN和多特征的全极化SAR土地覆盖分类", 测绘科学, no. 01 * |
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113406584A (en) * | 2021-06-22 | 2021-09-17 | 北京化工大学 | Cross calibration method for synthetic aperture radar |
| CN113744237A (en) * | 2021-08-31 | 2021-12-03 | 华中科技大学 | Deep learning-based automatic detection method and system for muck fluidity |
| CN114154886A (en) * | 2021-12-08 | 2022-03-08 | 重庆大学 | Land transition simulation method based on delay communication CA model |
| CN114154886B (en) * | 2021-12-08 | 2024-05-10 | 重庆大学 | Land transition simulation method based on delay communication CA model |
| CN116821694A (en) * | 2023-08-30 | 2023-09-29 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN116821694B (en) * | 2023-08-30 | 2023-12-01 | 中国石油大学(华东) | Soil humidity inversion method based on multi-branch neural network and segmented model |
| CN117114627A (en) * | 2023-10-18 | 2023-11-24 | 日照市自然资源和规划局 | land resource management system |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112906300B (en) | 2023-11-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112906300A (en) | Polarized SAR (synthetic Aperture Radar) soil humidity inversion method based on two-channel convolutional neural network | |
| CN110659722B (en) | Electric vehicle lithium ion battery health state estimation method based on AdaBoost-CBP neural network | |
| Zeng et al. | Comparison of partial least square regression, support vector machine, and deep-learning techniques for estimating soil salinity from hyperspectral data | |
| CN114254561A (en) | Waterlogging prediction method, waterlogging prediction system and storage medium | |
| CN107423537B (en) | Surface temperature downscaling method based on self-adaptive threshold | |
| CN111597756A (en) | Water quality parameter inversion method based on multispectral data of unmanned aerial vehicle | |
| CN113255972B (en) | Short-term rainfall prediction method based on Attention mechanism | |
| CN114117886A (en) | Water depth inversion method for multispectral remote sensing | |
| Xu et al. | Soil moisture estimation with SVR and data augmentation based on alpha approximation method | |
| CN113159439A (en) | Crop yield prediction method and system, storage medium and electronic equipment | |
| Qiao et al. | Estimating maize LAI by exploring deep features of vegetation index map from UAV multispectral images | |
| Yao et al. | Enhanced supervised descent learning technique for electromagnetic inverse scattering problems by the deep convolutional neural networks | |
| CN110954860B (en) | DOA and polarization parameter estimation method | |
| Zhou et al. | Developing a PCA–ANN model for predicting chlorophyll a concentration from field hyperspectral measurements in Dianshan Lake, China | |
| CN116343046A (en) | Multi-source remote sensing urban river water quality monitoring method based on self-optimizing machine learning | |
| Ren et al. | Study on a fast EC measurement method of soda saline-alkali soil based on wavelet decomposition texture feature | |
| CN116299247B (en) | InSAR atmospheric correction method based on sparse convolutional neural network | |
| Sihvonen et al. | Spectral profile partial least-squares (SP-PLS): Local multivariate pansharpening on spectral profiles | |
| CN110231625B (en) | Synthetic aperture imaging method based on multi-scale fusion | |
| CN116994144A (en) | High-resolution winter wheat yield estimation method, system, storage medium and electronic equipment based on synthetic aperture radar image | |
| CN114065931B (en) | Land water vapor inversion method and system based on physical model and neural network fusion | |
| CN114692971A (en) | Crop yield prediction method and device based on yield difference | |
| CN114239274A (en) | Method for calculating root layer soil water by multi-source remote sensing data driven index filtering model | |
| CN114047563A (en) | All-weather assimilation method of infrared hyperspectrum | |
| CN116542132B (en) | Water vapor data calibration and optimization method combining spherical crown harmonic model and neural network |
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 |