CN115542279A - Weather radar clutter classification and identification method and device - Google Patents

Abstract

The invention discloses a method and a device for classifying and identifying meteorological radar clutter, wherein the method comprises the steps of obtaining tested meteorological radar data and preprocessing the meteorological radar data; inputting the preprocessed tested meteorological radar data into the trained SegNet network model to obtain a classification recognition result; the training process of the SegNet network model comprises the following steps: acquiring and preprocessing trained meteorological radar data; integrating the preprocessed meteorological radar data into a corresponding training data set based on the data type; respectively calculating the standard deviation of a training data set corresponding to the radar reflectivity and the differential phase as texture data of the radar reflectivity and the differential phase; processing training data and texture data in the training data set by adopting a fuzzy logic algorithm to obtain a particle phase state type label set; initializing a SegNet network model, and performing iterative training based on a training data set and a label set; the method can effectively classify and identify the meteorological radar clutter so as to obtain high-quality meteorological radar data.

Description

Method and device for classifying and identifying meteorological radar clutter

technical field

The invention relates to a meteorological radar clutter classification and recognition method and device, belonging to the technical field of meteorological radar.

Background technique

An important prerequisite for analyzing and processing weather radar data is to establish an intelligent weather radar quality control model. The identification and classification of radar clutter has important application value in many fields. In the field of aviation, it not only has an early warning effect on the aviation hazards brought about by complex weather, but also provides a decision-making basis for route planning; and evaluation provide an important reference basis.

In order to obtain high-quality weather radar data, one of the most important problems is to reasonably distinguish between meteorological echoes and non-meteorological echoes. The classification and recognition methods of radar clutter include: statistical decision-making methods, decision-making graph methods, etc. The statistical decision-making method integrates the polarization characteristics of precipitation particles and clutter particles and previous research experience, and realizes the classification of precipitation particles and clutter particles by setting the polarization parameter threshold values of different precipitation particles and clutter particles. The threshold value of is generally fixed, so when the environment in the research target area changes, the classification accuracy will be greatly affected. The decision diagram method classifies precipitation particles and clutter particles according to the predetermined type boundary, but because the covariance matrix of different precipitation particles and clutter particles received by the weather radar is not independent of each other, the classification of the decision diagram method will be difficult. Accuracy is somewhat affected. For the classification and recognition tasks of precipitation particles and clutter particles, it is necessary to solve the problem of low intelligence of traditional algorithms and manual detection.

Contents of the invention

The purpose of the present invention is to overcome the deficiencies in the prior art and provide a method and device for classifying and identifying weather radar clutter, which can effectively classify and identify weather radar clutter, thereby obtaining weather radar data with high quality.

In order to achieve the above object, the present invention is achieved by adopting the following technical solutions:

In a first aspect, the present invention provides a method for classifying and identifying weather radar clutter, comprising:

Obtain and preprocess the tested weather radar data;

Input the preprocessed test weather radar data into the trained SegNet network model to obtain classification recognition results;

Wherein, the training process of described SegNet network model comprises:

Obtain the weather radar data of training and carry out preprocessing; The weather radar data includes radar reflectivity, differential reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase;

Based on the data type, the preprocessed weather radar data is integrated into a corresponding training data set;

Calculate the standard deviation of the training data set corresponding to the radar reflectivity and differential phase respectively as the texture data of radar reflectivity and differential phase;

The fuzzy logic algorithm is used to process the training data and texture data in the training data set to obtain the label set of the particle phase type;

Initialize the SegNet network model, and perform iterative training based on the training data set and label set until the preset maximum number of iterations or the weight parameters of the SegNet network model converge.

Optionally, the preprocessing includes data cleaning and scale expansion;

Described data cleaning comprises finding NaN data value and the radar reflectivity less than zero in weather radar data, and set zero respectively;

The scale expansion includes expanding the meteorological radar data by aligning the upper left and filling the lower right with zeros.

Optionally, said calculating the standard deviation of the training data set corresponding to the radar reflectivity and the differential phase respectively includes:

Traverse the training data set corresponding to the radar reflectivity, and calculate the standard deviation _SD (Z _H ) of the radar reflectivity within 1km:

In the formula, m(Z _H ) is the average value of radar reflectivity Z _H within 1 km, and n _Z is the number of data points of radar reflectivity within 1 km;

Traverse the training data set corresponding to the differential phase, and calculate the standard deviation of the differential phase within 2km

In the formula, m(φ _DP ) is the average value of differential phase φ _DP within 2km, and n _φ is the number of data points of differential phase within 2km.

Optionally, the step of processing the training data set and the texture data using the fuzzy logic algorithm to obtain the tag set of the particle phase state type includes:

The trapezoidal function is used as the membership function of the fuzzy logic algorithm, and the training data and texture data in the training data set are used as input parameters to perform fuzzy logic operations, and the membership degree corresponding to the particle phase state type is obtained as the fuzzy logic output; the trapezoidal function is:

In the formula, X ₁ , X ₂ , X ₃ , and X ₄ are the threshold parameters of the trapezoidal function, and P ^(j) ( _xi ) is the membership degree of the jth input parameter to the i-th particle phase state type;

The weighted average judgment method is used to defuzzify the fuzzy logic output, and the membership degree integration value corresponding to the particle phase state type is obtained:

In the formula, S _i is the integrated membership value of the jth input parameter to the i-th particle phase state type, W _ij is the weight coefficient of the j-th input parameter of the i-th particle phase state type, and J is the number of input parameters;

If the integrated value of the membership degree of the input parameter is 0, then the particle phase state type corresponding to the input parameter is no meteorological echo, and its label is set to 0;

If the integrated value of the membership degree of the input parameter is greater than the set threshold, the particle phase state type corresponding to the input parameter is clutter, and its label is set to 1;

If the integrated value of the membership degree of the input parameter is less than the set threshold and greater than 0, then the particle phase state type corresponding to the input parameter is other, and its label is set to 2;

Builds a label set for the particle phase type based on the labels of the input parameters.

Optionally, the iterative training based on the training data set and the label set includes:

The Encoder network based on the SegNet network model downsamples the training data to obtain data features;

Sampling characteristic data on Decoder network based on SegNet network model to obtain prediction results;

Calculate the loss of the prediction result and its label based on the cross-entropy loss function, and update the weight parameters of the SegNet network model according to the loss backpropagation.

Optionally, the Encoder network includes:

The first layer consists of two 64×3×3 convolutions and maximum pooling downsampling in turn;

The second layer consists of two 128×3×3 convolutions and maximum pooling downsampling in turn;

The third layer consists of three 256×3×3 convolutions and maximum pooling downsampling in turn;

The fourth layer consists of three 512×3×3 convolutions and maximum pooling downsampling in turn;

The fifth layer consists of three 5212×3×3 convolutions and maximum pooling downsampling in turn;

The Decoder network includes:

The first layer consists of inverse maximum pooling upsampling and three 521×3×3 convolutions;

The second layer consists of inverse maximum pooling upsampling and three 521×3×3 convolutions;

The third layer consists of inverse maximum pooling upsampling and two 256×3×3 convolutions;

The fourth layer consists of anti-maximum pooling upsampling and two 128×3×3 convolutions;

The fifth layer, in turn, consists of anti-maximum pooling upsampling and two 64×3×3 convolutions.

Optionally, during the iterative training process, Pixel Accuracy is used as an accuracy evaluation index to obtain the closeness between the prediction result and the label.

In a second aspect, the present invention provides a device for classifying and identifying weather radar clutter, said device comprising:

The preprocessing module is used to obtain and preprocess the weather radar data of the test;

The classification recognition module is used to input the weather radar data of the preprocessed test into the trained SegNet network model to obtain the classification recognition result;

Model training module for:

Obtain the weather radar data of training and carry out preprocessing; The weather radar data includes radar reflectivity, differential reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase;

Based on the data type, the preprocessed weather radar data is integrated into a corresponding training data set;

Calculate the standard deviation of the training data set corresponding to the radar reflectivity and differential phase respectively as the texture data of radar reflectivity and differential phase;

The fuzzy logic algorithm is used to process the training data and texture data in the training data set to obtain the label set of the particle phase type;

Initialize the SegNet network model, and perform iterative training based on the training data set and label set until the preset maximum number of iterations or the weight parameters of the SegNet network model converge.

In a third aspect, the present invention provides a weather radar clutter classification and identification device, including a processor and a storage medium;

The storage medium is used to store instructions;

The processor is configured to operate in accordance with the instructions to perform the steps according to the method described above.

In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

Compared with the prior art, the beneficial effects achieved by the present invention are as follows:

A method and device for classifying and identifying meteorological radar clutter provided by the present invention constructs a five-channel training data set by obtaining radar reflectivity, differential reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase, and obtains it through a fuzzy logic algorithm. The label set of the training data, the SegNet network model is trained through the training data set and the label set, and the weather radar clutter is classified and identified through the trained SegNet network model, which can effectively classify and identify the weather radar clutter, thereby obtaining quality Higher weather radar data.

Description of drawings

FIG. 1 is a flow chart of a method for classifying and identifying weather radar clutter provided by Embodiment 1 of the present invention.

detailed description

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention.

Embodiment one:

As shown in Figure 1, the present invention provides a kind of meteorological radar clutter classification identification method, comprising:

1. Obtain and preprocess the tested weather radar data;

2. Input the preprocessed test weather radar data into the trained SegNet network model to obtain classification and recognition results;

Among them, the training process of the SegNet network model includes:

S101. Acquiring weather radar data for training and performing preprocessing;

S102. Based on the data type, integrate the preprocessed weather radar data into a corresponding training data set;

S103, respectively calculating the standard deviation of the training data set corresponding to the radar reflectivity and the differential phase as the texture data of the radar reflectivity and the differential phase;

S104, using a fuzzy logic algorithm to process the training data and texture data in the training data set to obtain a label set of the particle phase type;

S105. Initialize the SegNet network model, and perform iterative training based on the training data set and the label set, until reaching the preset maximum number of iterations or convergence of weight parameters of the SegNet network model.

Among them, iterative training based on the training data set and label set includes:

S201. The Encoder network based on the SegNet network model downsamples the training data to obtain data features;

S202. Sampling feature data on the Decoder network based on the SegNet network model to obtain prediction results;

S203. Calculate the loss of the predicted result and its label based on the cross-entropy loss function, and update the weight parameters of the SegNet network model according to the loss backpropagation.

(1) The meteorological radar data provided in this embodiment include radar reflectivity, differential reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase, and are saved through numpy files.

(2) Preprocessing includes data cleaning and scale expansion;

Data cleaning includes finding NaN (null data) data values and radar reflectivity less than zero in weather radar data, and setting them to zero respectively;

Scale expansion includes expanding the meteorological radar data by aligning the upper left and filling the lower right with zeros. Generally, the size of the collected data is 336×920. In order to meet the maximum pooling operation of the SegNet network model, it needs to be expanded to 384×1088.

(3) Calculate the standard deviation of the training data set corresponding to the radar reflectivity and differential phase respectively:

Traverse the training data set corresponding to the radar reflectivity, and calculate the standard deviation _SD (Z _H ) of the radar reflectivity within 1km:

In the formula, m(Z _H ) is the average value of radar reflectivity Z _H within 1 km, and n _Z is the number of data points of radar reflectivity within 1 km;

Traverse the training data set corresponding to the differential phase, and calculate the standard deviation of the differential phase within 2km

In the formula, m(φ _DP ) is the average value of differential phase φ _DP within 2km, and n _φ is the number of data points of differential phase within 2km.

(4) Use the fuzzy logic algorithm to process the training data set and texture data to obtain the label set of the particle phase type, including:

The trapezoidal function is used as the membership function of the fuzzy logic algorithm, and the training data and texture data in the training data set are used as input parameters to perform fuzzy logic operations, and the membership degree corresponding to the particle phase state type is obtained as the fuzzy logic output; the trapezoidal function is:

In the formula, X ₁ , X ₂ , X ₃ , and X ₄ are the threshold parameters of the trapezoidal function, and P ^(j) ( _xi ) is the membership degree of the jth input parameter to the i-th particle phase state type;

The weighted average judgment method is used to defuzzify the fuzzy logic output, and the membership degree integration value corresponding to the particle phase state type is obtained:

In the formula, S _i is the integrated membership value of the jth input parameter to the i-th particle phase state type, W _ij is the weight coefficient of the j-th input parameter of the i-th particle phase state type, and J is the number of input parameters;

If the integrated value of the membership degree of the input parameter is 0, then the particle phase state type corresponding to the input parameter is no meteorological echo, and its label is set to 0;

If the integrated value of the membership degree of the input parameter is greater than the set threshold, the particle phase state type corresponding to the input parameter is clutter, and its label is set to 1;

If the integrated value of the membership degree of the input parameter is less than the set threshold and greater than 0, then the particle phase state type corresponding to the input parameter is other, and its label is set to 2;

Builds a label set for the particle phase type based on the labels of the input parameters.

(5) Convert the numpy-type training data set into (5×384×1088) tensor-type data, and combine the training data and its corresponding labels into a dataset. Set the batch of the training data set to 1, and use the DataLoader method in pytorch to load a tensor with a size of (1×channel×384×1088), where the channel of the training data is 5 (5 types of training data are used as 5 inputs channel data), the channel of the label is 1.

Design the structure of the Encoder network as follows:

The first layer consists of two 64×3×3 convolutions and maximum pooling downsampling in turn;

The second layer consists of two 128×3×3 convolutions and maximum pooling downsampling in turn;

The third layer consists of three 256×3×3 convolutions and maximum pooling downsampling in turn;

The fourth layer consists of three 512×3×3 convolutions and maximum pooling downsampling in turn;

The fifth layer consists of three 5212×3×3 convolutions and maximum pooling downsampling in turn;

According to the network structure, process (1×5×384×1088) training data to obtain (1×512×6×17) feature data, where 512 represents the number of feature channels.

Design the Decoder network structure as:

The first layer consists of inverse maximum pooling upsampling and three 521×3×3 convolutions;

The second layer consists of inverse maximum pooling upsampling and three 521×3×3 convolutions;

The third layer consists of inverse maximum pooling upsampling and two 256×3×3 convolutions;

The fourth layer consists of anti-maximum pooling upsampling and two 128×3×3 convolutions;

The fifth layer, in turn, consists of anti-maximum pooling upsampling and two 64×3×3 convolutions.

According to the network structure, the feature data of (1×512×6×17) is processed to obtain the prediction data of (1×3×384×1088), where 3 represents the three categories of labels corresponding to the number of channels.

Among them, the Encoder network also includes:

Set the parameter size of the convolution kernel of the Encoder network in the SegNet network model to 3, the stride to 1, and the padding parameter to 1, indicating that a square convolution kernel with a 3×3 moving step of 1 is used. When the convolution operation exceeds the edge of the data , pad a circle of zeros around the data to prevent out-of-bounds exceptions. Different convolutional layers use the same convolution to extract training data features; the expression of the convolution operation is as follows:

In the formula, H _in represents the length of the input data, _Win represents the width of the input data, K _H represents the length of the convolution kernel, K _W represents the width of the convolution kernel, P represents the number of zero padding circles, and S represents the convolution kernel movement stride.

Due to the inconsistency of the data distribution characteristics after each convolution, the latter convolutional layer needs to continuously adapt to the output changes of the previous convolutional layer, which reduces the decoupling between layers in the network. Batch normalization is used to normalize the data before convolution and fix its distribution characteristics, making network learning more stable. The expression for batch normalization is as follows:

In the formula, ( _xi ) ^(b) represents the value of the i _th input node of the layer when the b _th sample of the current batch is input, μ( _xi ) is the mean value of the input data, and σ( _xi ) is the standard deviation of the input data , ∈ is to prevent the extremely small value introduced by zero division, γ and β are the scale and shift parameters to be learned.

Use the Relu activation function to increase the nonlinear relationship between the convolutional layers to avoid the disappearance of the gradient. Relu will make a part of the convolution output 0, which causes the sparsity of the network, reduces the interdependence of parameters, and alleviates the occurrence of overfitting problems. The expression of Relu is as follows:

f(x)=max(0,x)

For the data after Relu, perform maximum pooling down-sampling, extract these data points with more information, effectively remove redundant information, and realize feature extraction. While downsampling, the index information of the maximum value is saved. The expression of the maximum pooling is as follows:

In the formula, f represents the size of the pooling kernel, and s represents the step size of the pooling kernel movement.

The Decoder network also includes:

Unpool the feature data after the convolution operation, and gradually restore the size of the data. Since the maximum pooling will record the index with the maximum value, the data at the index position is directly restored during unpooling, and the other positions are filled with 0, thereby gradually restoring the original size of the data. The expression of anti-pooling is as follows:

X _out ＝(X _in -1)×s+f

The Decoder network uses the same convolution to reduce the data feature channels after unpooling. The parameter size of the convolution kernel is 3, the stride is 1, and the padding parameter is 1. The length and width of the data before and after convolution of different convolution layers do not occur, only Change the number of channels of the feature. For the convolutional data, batch normalization is used to stabilize the distribution characteristics of the data and improve the learning ability of the network. Secondly, the Relu activation function is used to increase the nonlinearity between the convolutional layers to avoid the disappearance of the gradient.

(6) Calculating the loss of the prediction result and its label based on the cross-entropy loss function includes:

Use the softmax function to calculate the probability of each category of the predicted data. The sum of the corresponding position probabilities on each channel is 1, and the channel index of the maximum probability is the predicted label of the data point. The expression of Softmax is as follows:

In the formula, Xi is the data point on the _{i th} channel.

Calculate the cross-entropy loss between the predicted data and the label as the loss between the predicted value and the real value, and use the backpropagation mechanism in pytorch to update the gradient and parameter values of the weights in the SegNet network model to improve the prediction accuracy of SegNet. The expression of the cross-entropy loss function is as follows:

In the formula, p _i is the real label value, and q _i is the predicted data value after softmax calculation.

(7) In the process of iterative training, use Pixel Accuracy as an accuracy evaluation index to obtain the closeness of the prediction result and the label;

According to the value of the hyperparameter learning rate, select the appropriate number of training times to continuously train the SegNet network model. During the training iterations, save the current model parameters with the smallest loss. After the network training is completed, load the optimal network model, input the test data set, and obtain the predicted data. Use Pixel Accuracy as the accuracy evaluation index to check the closeness of the predicted data to the true value of the label. The expression for Pixel Accuracy calculation is as follows:

In the formula, k is the number of target categories, P _ii is the number of pixels of the prediction pair, and P _ij is the number of pixels belonging to class i but classified into class j.

Embodiment two:

An embodiment of the present invention provides a weather radar clutter classification and recognition device, the device comprising:

The preprocessing module is used to obtain and preprocess the weather radar data of the test;

The classification recognition module is used to input the weather radar data of the preprocessed test into the trained SegNet network model to obtain the classification recognition result;

Model training module for:

Obtain and preprocess the weather radar data for training; the weather radar data includes radar reflectivity, differential reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase;

Based on the data type, the preprocessed weather radar data is integrated into a corresponding training data set;

Calculate the standard deviation of the training data set corresponding to the radar reflectivity and differential phase respectively as the texture data of radar reflectivity and differential phase;

The fuzzy logic algorithm is used to process the training data and texture data in the training data set to obtain the label set of the particle phase type;

Initialize the SegNet network model, and perform iterative training based on the training data set and label set until the preset maximum number of iterations or the weight parameters of the SegNet network model converge.

Embodiment three:

Based on Embodiment 1, the embodiment of the present invention provides a weather radar clutter classification and identification device, including a processor and a storage medium;

The storage medium is used to store instructions;

The processor is configured to operate in accordance with the instructions to perform the steps according to the methods described above.

Embodiment four:

Based on the first embodiment, the embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the program is executed by a processor, the steps of the above method are implemented.

Those skilled in the art should understand that the embodiments of the present application may be provided as methods, systems, or computer program products. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present application is described with reference to flowcharts and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the present application. It should be understood that each procedure and/or block in the flowchart and/or block diagram, and a combination of procedures and/or blocks in the flowchart and/or block diagram can be realized by computer program instructions. These computer program instructions may be provided to a general purpose computer, special purpose computer, embedded processor, or processor of other programmable data processing equipment to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing equipment produce a An apparatus for realizing the functions specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to operate in a specific manner, such that the instructions stored in the computer-readable memory produce an article of manufacture comprising instruction means, the instructions The device realizes the function specified in one or more procedures of the flowchart and/or one or more blocks of the block diagram.

These computer program instructions can also be loaded onto a computer or other programmable data processing device, causing a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process, thereby The instructions provide steps for implementing the functions specified in the flow chart or blocks of the flowchart and/or the block or blocks of the block diagrams.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (10)

1. A meteorological radar clutter classification and identification method is characterized by comprising the following steps:

acquiring and preprocessing tested meteorological radar data;

inputting the preprocessed tested meteorological radar data into a trained SegNet network model to obtain a classification recognition result;

wherein the training process of the SegNet network model comprises the following steps:

acquiring and preprocessing trained meteorological radar data; the meteorological radar data comprise radar reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase;

integrating the preprocessed meteorological radar data into a corresponding training data set based on the data type;

respectively calculating the standard deviation of a training data set corresponding to the radar reflectivity and the differential phase as texture data of the radar reflectivity and the differential phase;

processing training data and texture data in the training data set by adopting a fuzzy logic algorithm to obtain a particle phase state type label set;

initializing a SegNet network model, and performing iterative training based on a training data set and a label set until a preset maximum iteration number is reached or a weight parameter of the SegNet network model is converged.

2. The weather radar clutter classification and identification method according to claim 1, wherein the preprocessing comprises data cleaning and scale expansion;

the data cleaning comprises searching for NaN data values and radar reflectivity smaller than zero in meteorological radar data, and respectively setting the NaN data values and the radar reflectivity smaller than zero;

and the scale expansion comprises the step of expanding the meteorological radar data in a mode of upper left alignment and lower right zero padding.

3. The method for classifying and identifying weather radar clutter according to claim 1, wherein the calculating the standard deviation of the training data set corresponding to radar reflectivity and differential phase respectively comprises:

traversing a training data set corresponding to the radar reflectivity, and calculating the standard deviation S of the radar reflectivity within 1km range _D (Z _H )：

In the formula, m (Z) _H ) Is radar reflectivity Z in the range of 1km _H Average value of (1), n _Z The number of data points of radar reflectivity within a range of 1 km;

traversing a training data set corresponding to the differential phase, and calculating the standard deviation of the differential phase within the range of 2km

In the formula, m (phi) _DP ) Is differential phase phi in the range of 2km _DP Average value of (3), n _φ The number of data points of differential phase in the range of 2 km.

4. The weather radar clutter classification and identification method according to claim 1, wherein the processing the training data set and the texture data by the fuzzy logic algorithm to obtain the tag set of the particle phase type comprises:

adopting a trapezoidal function as a membership function of a fuzzy logic algorithm, carrying out fuzzy logic operation by taking training data and texture data in a training data set as input parameters, and obtaining membership corresponding to a particle phase type as fuzzy logic output; the trapezoidal function is:

in the formula, X ₁ 、X ₂ 、X ₃ 、X ₄ Is a threshold parameter of a trapezoidal function, P ^(j) (x _i ) The membership degree of the jth input parameter to the ith particle phase state type;

and (3) performing defuzzification on the fuzzy logic output by adopting a weighted average judgment method to obtain a membership integrated value corresponding to the particle phase type:

in the formula, S _i Membership level integration value, W, of ith particle phase type for jth input parameter _ij The weight coefficient of the jth input parameter of the ith particle phase type is obtained, and J is the number of the input parameters;

if the membership integrated value of the input parameter is 0, the particle phase type corresponding to the input parameter is no meteorological echo, and the label of the particle phase type is set to be 0;

if the membership integrated value of the input parameter is greater than the set threshold, the particle phase type corresponding to the input parameter is a clutter, and the label of the particle phase type is set to be 1;

if the membership integration value of the input parameter is smaller than the set threshold and larger than 0, the particle phase type corresponding to the input parameter is other, and the label of the particle phase type is set to be 2;

and constructing a tag set of the particle phase type according to the tags of the input parameters.

5. The weather radar clutter classification and identification method according to claim 1, wherein the iterative training based on the training data set and the tag set comprises:

an Encoder network downsampling training data based on a SegNet network model to obtain data characteristics;

sampling characteristic data on a Decoder network based on a SegNet network model to obtain a prediction result;

and calculating the loss of the prediction result and the label thereof based on the cross entropy loss function, and updating the weight parameter of the SegNet network model according to the loss back propagation.

6. The weather radar clutter classification and identification method according to claim 5, wherein the Encoder network comprises:

a first layer consisting of two 64 x 3 convolutions and maximum pooled downsampling in sequence;

a second layer consisting of, in order, two 128 x 3 convolutions and maximum pooled downsampling;

the third layer consists of three convolutions of 256 multiplied by 3 and the maximum pooled downsampling in sequence;

a fourth layer consisting of three 512 x 3 convolutions and maximum pooled downsampling in sequence;

a fifth layer consisting of three 5212 × 3 × 3 convolutions and maximum pooled downsampling in sequence;

the Decoder network includes:

a first layer consisting of, in order, inverse max-pooling upsampling and three 521 × 3 × 3 convolutions;

a second layer consisting of, in order, inverse maximal pooled upsampling and three 521 × 3 × 3 convolutions;

the third layer consists of inverse maximum pooling upsampling and two convolutions of 256 multiplied by 3 in turn;

a fourth layer consisting of inverse maximal pooled upsampling and two 128 x 3 convolutions in sequence;

the fifth layer, in turn, consists of inverse max pooling upsampling and two 64 x 3 convolutions.

7. The weather radar clutter classification and identification method according to claim 5, wherein in the iterative training process, the Pixel Accuracy is used as an Accuracy evaluation index to obtain the proximity of the prediction result and the tag.

8. A weather radar clutter classification and identification device is characterized in that the device comprises:

the preprocessing module is used for acquiring and preprocessing the tested meteorological radar data;

the classification identification module is used for inputting the preprocessed tested meteorological radar data into the trained SegNet network model to obtain a classification identification result;

a model training module to:

acquiring and preprocessing trained meteorological radar data; the meteorological radar data comprise radar reflectivity, differential propagation phase shift rate, correlation coefficient and differential phase;

integrating the preprocessed meteorological radar data into a corresponding training data set based on the data type;

respectively calculating the standard deviation of a training data set corresponding to the radar reflectivity and the differential phase as texture data of the radar reflectivity and the differential phase;

processing training data and texture data in the training data set by adopting a fuzzy logic algorithm to obtain a particle phase state type label set;

initializing a SegNet network model, and performing iterative training based on a training data set and a label set until a preset maximum iteration number is reached or a weight parameter of the SegNet network model is converged.

9. A meteorological radar clutter classification and identification device is characterized by comprising a processor and a storage medium;

the storage medium is used for storing instructions;

the processor is configured to operate in accordance with the instructions to perform the steps of the method according to any one of claims 1 to 7.

10. Computer-readable storage medium, on which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 7.

Images