CN116310470A - Method for classifying migratory flies and birds based on dual-polarized weather radar data - Google Patents

Abstract

The invention discloses a method for classifying migratory birds based on dual-polarized weather radar data. The invention uses the reflectivity factor Z, the speed spectrum width W and the differential reflectivity Z in the dual-polarized weather radar data _DR Differential phase shift ψ _DP Correlation coefficient ρ _HV As the characteristic input, the migratory birds are classified based on the classifier, and the result shows that the characteristic consisting of the 5 parameters can effectively improve the accuracy of quantifying the biomass of the migratory birds and the biomass of the migratory insects, thereby being beneficial to monitoring the biological migration condition, analyzing the influence of meteorological factors on biological activities and researching biological interaction.

Description

Method for classifying migratory flies and birds based on dual-polarized weather radar data

Technical Field

The invention relates to the technical field of weather radar, in particular to a migratory insect and bird classification method based on dual-polarized weather radar data.

Background

Hundreds of millions of aerial animals such as insects, birds and the like fly away from the breeding ground and the overwintering ground in a long distance every year, and the flying range covers the whole world. Animal flight behavior involves many aspects of ecosystem function, processes, and bio-geochemistry. Animal flying is also closely related to human life, insect flying can cause crop diseases and insect pests, birds and bat flying can cause epidemic diseases of human and livestock.

In order to conduct intensive studies on the migration of organisms, it is first necessary to effectively monitor the migration organisms. The detection distance of the weather radar can reach hundreds of kilometers, and the weather radar can work continuously day and night, so that the weather radar is an effective observation means for researching large-scale biological migration. Biological echoes monitored by the weather radar are mainly divided into birds and insects, and accurate identification of the bird echoes and the insect echoes is beneficial to improvement of accuracy of quantification of the migratory biomass. The existing dual-polarized weather radar worm and bird classification method is to utilize the differential reflectivity Z of organisms _DR Differential phase shift ψ _DP Correlation coefficient ρ _HV The classifier is trained to distinguish insects from birds, however the classification accuracy of this approach is only around 85%.

Disclosure of Invention

In view of the above, the invention provides a method for classifying migratory birds based on dual-polarized weather radar data, which provides an effective means for monitoring large-scale migratory birds and migratory insects. Compared with the existing insect and bird classification method, the method has higher classification accuracy.

The invention relates to a method for classifying migratory flies and birds based on dual-polarized weather radar data, which uses a reflectivity factor Z, a speed spectrum width W and a differential reflectivity Z in PPI data of a dual-polarized weather radar plane position display _DR Differential phase shift ψ _DP Coefficient of correlation ρ _HV And (5) classifying the flying insects and birds based on the classifier for inputting the characteristics.

Preferably, the method also comprises the steps of measuring the reflectivity factor Z, the speed spectrum width W and the differential reflectivity Z _DR Differential phase shift ψ _DP Coefficient of correlation ρ _HV Performing data processing, the data processing comprising: the reflectivity factor Z, the velocity spectrum width W and the differential reflectivity Z _DR Differential phase ψ _DP Coefficient of correlation ρ _HV According to radar azimuthThe angular ascending order is reordered and interpolated to 0:360 °, regular space evenly spaced.

Preferably, clutter removal of PPI data of the dual polarized weather radar planar position display is further included, specifically:

firstly, changing the PPI data of a dual-polarized weather radar planar position display from polar coordinates to rectangular coordinates; then binarization processing is carried out; then, the communication domain at the center of the PPI is a main communication domain, other parts which are not connected with the main communication domain are secondary communication domains, and other secondary communication domains which are not connected with the main communication domain except the main communication domain at the center of the PPI are removed to complete clutter removal.

Preferably, the binarization processing is specifically: setting the range gate containing the echo to 1 and the rest to 0;

the connected domain is removed, specifically: with binarized image as mask M ₀ Mask M for determining the main echo region ₁ The method comprises the following steps:

wherein (1)>

For an expansion operator, B is an 8 connected domain window, and max represents a connected domain with the largest area;

image I after removal of other communication areas ₂ The method comprises the following steps: i ₂ ＝I ₁ ×M ₁ Wherein I ₁ The PPI data of the dual-polarized weather radar planar position display is an image obtained by changing polar coordinates into rectangular coordinates.

Preferably, the method further comprises noise removal of PPI data of the dual-polarized weather radar planar position display, specifically: and averaging the data of the weather radar PPI at different distance gates to finish noise removal.

Preferably, the method also comprises a reflectivity factor Z, a speed spectrum width W and a differential reflectivity Z _DR And differential phase ψ _DP Is performed in the normalization process.

Preferably, the normalization process specifically includes:

wherein Z is ^* 、W ^* 、Z ^* _DR And psi is ^* _DP Respectively normalized reflectivity factor Z, velocity spectrum width W and differential reflectivity Z _DR And differential phase ψ _DP 。

Preferably, the classifier is a support vector machine, a K nearest neighbor, naive Bayes, a random forest, a decision tree, a neural network and the like.

Preferably, the first step is to use [ Z, W, Z _DR ,Ψ _DP ,ρ _HV ]Training a classifier for input, and testing to obtain a classification accuracy A;

then, Z, W, Z are removed from the input parameters _DR ,Ψ _DP And ρ _HV Training the classifier, and testing to obtain respective classification accuracy B, C, D, E and F;

the classification accuracy A is respectively subtracted by the classification accuracy B, C, D, E and F, and normalization is carried out to obtain corresponding input parameters Z, W and Z _DR ,Ψ _DP And ρ _HV Is a weight of (2).

Preferably, the working wave band of the dual-polarized weather radar data is an S wave band, a C wave band or an X wave band.

The beneficial effects are that:

the invention breaks the cognition of the prior art and proposes a method based on dual-polarized weather radar data Z, W and Z _DR ,Ψ _DP ,ρ _HV The classification of the migratory birds shows that the characteristics consisting of the 5 parameters can improve the classification accuracy of the migratory birds so as to improve the accuracy of quantifying the biomass of the migratory birds and the biomass of the migratory insects. This helps to monitor the conditions of biological flight, analyze biological activity for weather factors, and study biological interactions.

The invention carries out data normalization, clutter removal, noise removal, normalization and other treatments on the radar data, so that the classifier can realize classification more quickly and accurately. A bird detection radar is used for verifying the classification result of the method, and the result shows that the accuracy of the classifier reaches 95.8%.

Drawings

FIG. 1 is a flow chart of the classification method of the present invention.

Fig. 2 is a statistical result of dual polarized parameter distribution of insect and bird in the specific embodiment.

Fig. 3 is a schematic diagram of small communication area removal in an embodiment.

Fig. 4 is a typical migratory insect classification result in an embodiment.

FIG. 5 is a typical migratory bird classification result in an embodiment.

FIG. 6 is a classification of mixed cases of migratory birds in an embodiment.

Fig. 7 is a weight evaluation result of different input parameters in the embodiment.

Fig. 8 is a classification result of a bird detection radar output tag versus a worm-bird classifier in the specific embodiment.

Detailed Description

The invention will now be described in detail by way of example with reference to the accompanying drawings.

The invention provides a method for classifying migratory flies and birds based on dual-polarized weather radar data, which uses a reflectivity factor Z, a speed spectrum width W and a differential reflectivity Z in the dual-polarized radar data _DR Differential phase shift ψ _DP Coefficient of correlation ρ _HV The method is characterized in that flying insects and flying birds are classified, and the accuracy rate reaches 95.8%.

The prior method only uses Z when training the insect-bird classifier _DR ，Ψ _DP ，ρ _HV The reflectance factor Z and the velocity spectrum width W are not used. Z is determined by the Radar Cross Section (RCS) of all targets and the number of targets in the weather radar beam range, although birds have RCS much larger than insects, the number of birds and insects is not certain, and existing methods believe Z may be ambiguous between birds and insects and therefore not used. W is a measure of the degree of dispersion of all target speeds within a range gate of the weather radar, the flying speed of birds is faster than insects, the degree of dispersion is higher than insects, butThe existing method considers that the W signal-to-noise ratio of the biological target is not high, so the method is not used. The present invention has been found that although there may be ambiguity in Z of insects and birds, there is only a certain probability and not absolute. As with W, a low signal-to-noise ratio may lead to blurred insect-bird characteristics, but is not absolute. In fact, insects and birds Z _DR ，Ψ _DP ，ρ _HV And the two parts are overlapped, so long as Z and W of the insects and birds are not identical, the two parts can provide assistance for classifying the insects and birds. Therefore, the invention breaks the convention and proposes a method based on dual-polarized weather radar data Z, W and Z _DR ,Ψ _DP ,ρ _HV The classification of the migratory birds shows that the characteristics consisting of the 5 parameters can improve the classification accuracy of the migratory birds, thereby improving the accuracy of quantifying the biomass of the migratory birds and the biomass of the migratory insects. This helps to monitor the conditions of biological flight, analyze biological activity for weather factors, and study biological interactions.

The specific embodiments of the present invention are as follows:

step one, detecting a monitoring area by adopting a dual-polarized weather radar; screening biological echoes from weather radar echoes for the dual polarized weather radar flying insect bird dataset for training and evaluation of the present invention:

step two, screening a data characteristic set suitable for classifying insects and birds from biological echo data in weather radar echo:

the flying insects and birds have different sizes and numbers, so that the weather radar can observe different parameter characteristics of the insect-bird planar position display (Plan Position Indicator, PPI). Z is Z _DR ，Ψ _DP There is a strong azimuthal dependence. The insect is characterized by Z _DR ，Ψ _DP There is no apparent azimuthal dependence. Based on the characteristics of the PPI of the insect bird, the reflectivity factor Z, the velocity spectrum width W and the correlation coefficient rho in the PPI data are added _HV And (5) forming a data set to classify insects and birds.

Next, a series of data processing operations are performed on the selected PPI data so that it is suitable for input to a classifier, which in turn may be based on the classifier. The classifier of the present invention may employ, but is not limited to, the following: support vector machines, K-nearest neighbors, naive bayes, random forests, decision trees, neural networks, and the like. The embodiment specifically comprises the following data processing operations:

step three, the initial azimuth angle and the interval angle of each azimuth scanning of the weather radar are not completely the same, and in order to facilitate subsequent data processing, the reflectivity factor Z, the speed spectrum width W and the differential reflectivity Z are adopted _DR Differential phase ψ _DP And a correlation coefficient ρ _HV Reordered in azimuth ascending order, and interpolated to 0:360 deg., regular spaces evenly spaced (e.g., 1 deg. apart).

In the fourth step, the PPI image of the migrated echo is mostly an approximately circular area centered on the radar, but some clutter far from the radar may exist besides this area. To eliminate such clutter, PPI image I ₀ (r, θ) from polar coordinates to rectangular coordinates to obtain an image I ₁ (x,y),

Where r is the detection distance of the weather radar and θ is the azimuth of the antenna.

Then for I ₁ Binarizing, setting distance gate containing echo to 1 and the rest to 0 to obtain mask M ₀ . Mask M in which the main echo region ₁ It can be expressed as that,

wherein, the liquid crystal display device comprises a liquid crystal display device,

for the dilation operator, B is an 8-connected-domain window, and max represents a connected-domain where the area is the largest, which may also be referred to as a main connected-domain.

Finally, removing the image I after the secondary communication area ₂ It may be expressed as that,

I ₂ ＝I ₁ ×M ₁ (3)

fifthly, the polarization parameters of birds are obviously changed along with azimuth angles, but are not obviously changed along with distances, and the polarization parameters of insects are not obviously changed along with azimuth angles and distances, so that the data of the weather radar PPI at different distance gates can be averaged to reduce noise influence. Taking the reflectance factor Z as an example, the average

It can be expressed as that,

where n is the number of all valid range gates at azimuth angle θ.

Step six, reflectivity factor Z, velocity spectrum width W and differential reflectivity Z _DR Differential phase ψ _DP And a correlation coefficient ρ _HV The 5 parameters are different in size range, and the embodiment normalizes the parameters, namely maps the parameters to the range of 0-1, so that the training speed of the classifier is increased, and the classification accuracy is improved. Wherein ρ is _HV The method is within the range of 0 to 1, normalization is not needed, the specific calculation modes of other 4 parameters needing normalization are as follows,

seventh, each pixel point on the weather radar PPI has a reflectivity factor Z, a speed spectrum width W and a differential reflectivity Z _DR Differential phase ψ _DP And a correlation coefficient ρ _HV These features are combined into a feature vector X, which can be expressed as: x= [ Z, W, Z _DR ,φ _DP ,ρ _HV ]. Inputting the feature vector X into a classifier, adjusting super-parameter training to obtain a migratory insect-bird classifier based on dual-polarized weather radar data, and using a personThe test set selected by the worker verifies the classifier accuracy.

Step eight, in order to further evaluate the contribution of the 5 input parameters to the classifier, an ablation experiment method is also adopted to evaluate the five input parameters, and the weight of the five input parameters is calculated; the method comprises the following steps: will Z, W, Z _DR ,Ψ _DP ,ρ _HV And respectively removing the input parameters, repeating the step seven for training, and testing to obtain the classification accuracy. And (3) subtracting the classification accuracy obtained by the step seven from the classification accuracy obtained by removing all the input parameters in the step, and then normalizing to obtain the weight of the corresponding input parameter.

And step nine, verifying the classification accuracy of the trained flying insect bird classifier by using the test set. In addition, a bird detection radar is used for verifying the prediction result of the migratory fly bird classifier. The method comprises the following steps: and (3) selecting an airspace which can be observed jointly by the bird detection radar and the weather radar, and converting the bird vertical cumulative density measured by the bird detection radar into a worm-bird tag according to a threshold of 11 dB. And taking the tag output by the bird detection radar as a true value, and verifying the classification result of the insect-bird classifier. Wherein, the bird detection radar can select Ku wave band phased array bird detection radar.

In order to verify the method, based on the measured data of the S-band dual-polarized weather radar in the south of Shandong province, the classification of the migratory winged insect birds is completed by adopting the method for classifying the migratory winged insect birds based on the dual-polarized weather radar data, and the classification result is verified by using a bird detection radar.

Step one, 637 migratory bird echoes PPI and 537 migratory insect echoes PPI are manually selected from measurement data of the dual-polarized weather radar 2021 in 5,6 and 10 months to establish 1 migratory organism echo data set, and all data are selected in clear weather.

And step two, counting the parameter distribution of the worm and bird data set, as shown in figure 1. It can be seen that although the worm and bird are at the reflectance factor Z, the velocity spectrum width W, the differential reflectance Z _DR Differential phase ψ _DP The probability distributions on the upper are different, but there is still a partial overlap region.

And thirdly, sequencing and interpolating the worm and bird echoes PPIs in the data set to obtain regular PPIs.

And step four, small area removal is carried out on the regular PPI, as shown in figure 2.

And fifthly, averaging the data of different distance gates of the same azimuth angle of the PPI.

And step six, normalizing the averaged data.

And seventh, forming the normalized data into feature vectors, inputting the feature vectors into a support vector machine, and adjusting the super-parameter training to obtain the insect-bird classifier. The classifier is tested by using a manually selected test set, the classification accuracy reaches 92.6%, the typical cases of flying insects and birds are shown in the classification results of the mixed cases of flying insects and birds as shown in figures 3, 4 and 5.

Step eight, Z, W and Z _DR ,Ψ _DP ,ρ _HV And respectively removing the input parameters, repeating the step seven for training, and testing to obtain the classification accuracy. The classification accuracy obtained in the step seven is subtracted from the classification accuracy obtained in the step seven, and then normalization is carried out, so that weights of different input parameters are obtained, as shown in fig. 6.

And step nine, using a bird penetrating radar to verify the accuracy of the classifier. And (3) selecting an airspace which can be observed jointly by the bird detection radar and the weather radar, and converting the bird vertical cumulative density measured by the bird detection radar into a worm-bird tag according to a threshold of 11 dB. The data labels smaller than 11dB are insects, the data labels larger than or equal to 11dB are birds, the labels output by the bird detection radar are used as true values, and the classification results of the classifier are compared, so that the accuracy reaches 95.8%, as shown in figure 7.

In summary, the above embodiments are only preferred embodiments of the present invention, and are not intended to limit the scope of the present invention. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims (10)

1. Migratory insect and bird classification method based on dual-polarized weather radar dataCharacterized in that the reflectivity factor Z, the speed spectrum width W and the differential reflectivity Z in the PPI data of the dual-polarized weather radar planar position display _DR Differential phase shift ψ _DP Coefficient of correlation ρ _HV And (5) classifying the flying insects and birds based on the classifier for inputting the characteristics.

2. The method for classifying the migratory birds based on the dual-polarized weather radar data according to claim 1, further comprising the steps of classifying the reflectivity factor Z, the speed spectrum width W and the differential reflectivity Z _DR Differential phase shift ψ _DP Coefficient of correlation ρ _HV Performing data processing, the data processing comprising: the reflectivity factor Z, the velocity spectrum width W and the differential reflectivity Z _DR Differential phase ψ _DP Coefficient of correlation ρ _HV Reordered in radar azimuth ascending order, and interpolated to 0:360 °, regular space evenly spaced.

3. The method for classifying migratory birds based on dual polarized weather radar data according to claim 1 or 2, further comprising clutter removal for PPI data of dual polarized weather radar flat panel display, in particular:

and (3) converting the PPI data of the dual-polarized weather radar planar position display from polar coordinates to rectangular coordinates, performing binarization processing, and then removing other connected domains outside the main connected domain at the center of the PPI to finish clutter removal.

4. The method for classifying the migratory birds based on the dual-polarized weather radar data according to claim 3, wherein the binarization process is specifically as follows: setting the range gate containing the echo to 1 and the rest to 0;

the connected domain is removed, specifically: with binarized image as mask M ₀ Mask M for determining the main echo region ₁ The method comprises the following steps:

wherein (1)>

For an expansion operator, B is an 8 connected domain window, and max represents a connected domain with the largest area;

image I after removal of other communication areas ₂ The method comprises the following steps: i ₂ ＝I ₁ ×M ₁ Wherein I ₁ The PPI data of the dual-polarized weather radar planar position display is an image obtained by changing polar coordinates into rectangular coordinates.

5. The method for classifying migratory birds based on dual polarized weather radar data according to any one of claims 1, 2, 4, further comprising noise removal of PPI data of the dual polarized weather radar flat panel display, in particular: and averaging the data of the weather radar PPI at different distance gates to finish noise removal.

6. The method for classifying the migratory insect birds based on the dual polarized weather radar data according to any one of claims 1, 2 and 4, further comprising a reflectivity factor Z, a speed spectrum width W and a differential reflectivity Z _DR And differential phase ψ _DP Is performed in the normalization process.

7. The method for classifying the migratory birds based on the dual-polarized weather radar data according to claim 6, wherein the normalization process is specifically as follows:

wherein Z is ^* 、W ^* 、Z ^* _DR And psi is ^* _DP Respectively normalized reflectivity factor Z, velocity spectrum width W and differential reflectivity Z _DR And differential phase ψ _DP 。

8. The method for classifying the migratory fly birds based on the dual-polarized weather radar data according to claim 1, wherein the classifier is a support vector machine, a K-nearest neighbor, a naive bayes, a random forest, a decision tree or a neural network.

9. The method for classifying migratory birds based on dual polarized weather radar data according to claim 1 or 8,

first by [ Z, W, Z _DR ,Ψ _DP ,ρ _HV ]Training a classifier for input, and testing to obtain a classification accuracy A;

then, Z, W, Z are removed from the input parameters _DR ,Ψ _DP And ρ _HV Training the classifier, and testing to obtain respective classification accuracy B, C, D, E and F;

the classification accuracy A is respectively subtracted by the classification accuracy B, C, D, E and F, and normalization is carried out to obtain corresponding input parameters Z, W and Z _DR ,Ψ _DP And ρ _HV Is a weight of (2).

10. The method for classifying the migratory birds based on the dual-polarized weather radar data according to claim 1 or 8, wherein the working band of the dual-polarized weather radar data is S-band, C-band or X-band.

Images