CN112633227A

CN112633227A - Automatic identification method and system for Zhang Heng I induction magnetometer data lightning whistle sound wave

Info

Publication number: CN112633227A
Application number: CN202011623352.4A
Authority: CN
Inventors: 王桥; 袁静
Original assignee: National Institute of Natural Hazards
Current assignee: National Institute of Natural Hazards
Priority date: 2020-12-30
Filing date: 2020-12-30
Publication date: 2021-04-09
Anticipated expiration: 2040-12-30
Also published as: CN112633227B

Abstract

The invention discloses a method for automatically identifying data lightning whistle sound waves of a Zhanghenyi induction magnetometer, which comprises the following steps: sample collection (SCM data is used as a Fourier spectrum), graying and scale reduction processing, fuzzy convolution processing, L-form convolution processing and SVM model classification and identification; the invention also discloses a Zhang Heng I induction magnetometer data lightning whistle sound wave automatic identification device and system corresponding to the method; the invention develops exploration and research of automatically recognizing lightning whistle sound waves on SCM data of Zhangheng satellite I; fuzzy convolution kernels and L-shaped convolution kernels are designed according to the characteristics of frequency spectrum data and the L-shaped characteristics of the lightning whistle sound waves, so that the lightning morphological characteristics in a frequency spectrum map are further enhanced, and the robustness and the recognition performance of the characteristics are greatly improved; and an SVM classifier is adopted to classify the features, so that the recognition performance of the SVM classifier is over 94% in precision, recall rate and sorting capacity.

Description

Automatic identification method and system for Zhang Heng I induction magnetometer data lightning whistle sound wave

Technical Field

The invention relates to the field of lightning whistle sound wave identification in space physical research, in particular to a Zhang Heng I induction magnetometer data lightning whistle sound wave automatic identification method and system.

Background

Since the 80's 20 th century, exploration of electromagnetic satellite-based seismic monitoring has resulted in the detection of a large amount of electromagnetic anomaly information (Shen et al, 2011; Larkina, 1983; myrev et al, 1989; Parrot and Mogilevsky, 1989; pullines and Legen "Ka, 2003; Molchanov, 1993; Cai et al, 2007), providing a new observation means for electromagnetic studies. Lightning is a natural phenomenon which occurs most frequently in a plurality of meteorological activities, and Brooks (1925) estimates that 1800 thunderstorms exist at any moment in the world by using thunderstorm day data, so that a result (Brooks,1925) that the global lightning frequency is 100 fls-1 (fl is short for flash and represents the number of times lightning occurs) is obtained. High current and strong electromagnetic radiation are easily generated in the lightning discharge process, the peak value of a radiation field in the ionized layer can reach or even exceed 15V/m (Marshall et al, 2010), and the peak value greatly exceeds the ionizing and exciting light radiation thresholds of certain components in the atmosphere, so that the ionized layer is obviously influenced, and the electromagnetic space-time characteristics of the ionized layer are obviously changed (Xiao et al, 2013; Huang and Gu, 2003; Davis and Lo, 2008). Since lightning is able to penetrate the ionosphere, researchers have resorted to lightning to obtain information about the physics of the geospatial space. For example: zahlava et al analyzed the results of measurements of longitudinal dependence of DEMETER and RBSP on whistling mode waves of the magnetic layer inside the earth, which showed strong longitudinal dependence of whistling mode waves inside the plasma layer (Z a hlava et al, 2018). Bayuptai et al analyzed the dispersion of the lightning whistle observed by AKEBONO, which discusses the time the lightning whistle travels along the track versus the electron density distribution (bayuptai et al, 2012), indicating that the dispersion trend of the lightning whistle is a powerful method of determining the overall electron density distribution in a plasma layer. Oike et al analyzed the frequency of occurrence of lightning whistles detected by the akabonono satellites as well as the spatial and temporal variations of the lightning activity observed on the ground (Oike et al, 2014), indicating that the occurrence of lightning whistles in the ionosphere is closely related to the lightning activity and the electron density distribution around the earth. Cliverd et al (2002) use the global lightning location network (WWLLN) to determine the location of the source of the lightning, then remotely sense the electron density distribution frequency based on the propagation time of the lightning whistle sound from the source point to the observation point, and derive a function of the electron density along the propagation path by propagation theory (Oike et al, 2014). It can be seen that understanding the physical parameters and positions of the lightning whistle sound waves is the basis for developing and utilizing lightning, and the characteristic of studying space environments such as an ionized layer and the like by means of the lightning whistle sound waves is an important technical means.

At present, most of data used for researches on lightning whistle sound waves are limited, and for a large amount of lightning whistle sound wave data, researches on space-time distribution rules and related parameters are less developed, and two reasons mainly exist: 1. sufficient lightning whistle sound events are required; 2. it is a very challenging task to rely on manual acquisition of lightning whistle events from a large volume of electromagnetic observation data. The magnetic field observation data carries complex and various disturbance signals of space electromagnetism and the like, such as satellite platform disturbance, lightning whistle, VLF ground emission source, magnetic storm and the like. There are a lot of lightning whistle sound waves in satellite data, and the existing research mainly focuses on some obvious magnetic storms and disturbance of ionosphere by VLF ground emission source (Yang et al, 2020; Liao et al, 2019; Zhao et al, 2019), and there is little research on how to automatically identify and locate lightning events from data. With the formation of a large satellite data pool, how to automatically identify the lightning whistle sound waves from a large amount of satellite observation data also becomes an important prerequisite for deep analysis. It is necessary to discuss a feasible automatic recognition method of lightning whistle sound waves to break through the bottleneck of research.

Lightning whistles originate from lightning discharges and have unique spectral characteristics characterized by decreasing frequency and amplitude over time due to differences in attenuation of high and low frequency components. This dispersion spectrum is called "dispersion" and becomes broad when the path length is long or the electron density along the transmission path is dense, so if these values can be derived along the propagation path, dispersion can be a very useful parameter to derive the electron density distribution in the plasma layer as shown in fig. 1, 2. Helliwell (1965) classifies the morphology of lightning whistle sound waves into 9 classes according to the degree of dispersion.

The current research on the automatic identification of lightning whistle waves is mainly directed to waveform data acquired by an AKEBONO satellite (bayupita et al, 2012), data acquired by an Arase satellite (Ali Ahmad et al, 2019) and spectral data acquired by a demoter satellite (Parrot et al, 2019).

According to Eckersley's theorem (Gurnett et al, 1990), the time and frequency of lightning whistle waves satisfy the formula (1):

wherein D is the lightning Whistle sonic dispersion constant; t is the time of arrival at frequency f and t0 is the time of lightning trigger. Bayuptati applies this theorem to waveform data acquired by an akabonono satellite for lightning identification (2012). The traditional automatic identification method of the lightning whistle sound wave is realized by identifying a straight line in a power spectrum, as shown in figure 1: t1 is the time of arrival at the frequency f1, t2 is the time of arrival at the frequency f2, and D is calculated as shown in equation (2):

it is found from fig. 1 that the lightning whistle sound wave is a straight line having a certain slope, and the equations (1) and (2) merely represent a simple relationship between the frequency of the lightning whistle sound wave and time.

(Ali Ahmad et al, 2019) considers that it is difficult to detect other lightning types summarized by (Helliwell et al, 1965) by the above method, and by referring to the definition of the lightning type of (Helliwell et al, 1965), the types of lightning whistle waves of Arase satellites are corrected again, and different mode characteristics are established according to different types. The identification method comprises the following steps: firstly, FFT spectrum calculation is carried out, then a Gaussian convolution kernel is adopted for carrying out fuzzy processing on a spectrogram, a Laplace operator is used for edge extraction, and then an image is segmented by an image segmentation algorithm, as shown in FIG. 3; and equally dividing the image into a plurality of grids as shown in fig. 4; finally, features are extracted by adopting a Bresenham's linear detection algorithm (Bresenham et al, 1977) and then input into a decision tree for recognition, and the recognition precision is only 75%. The algorithm has several problems:

(1) in the identification process, a large amount of image preprocessing operations are needed, such as parameter selection of a Gaussian convolution kernel, parameter selection of a Laplace operator, parameter selection of a segmentation algorithm and the like, and the selection of the parameters has a large influence on feature extraction by adopting a Bresenham's algorithm.

(2) The Bresenham's algorithm requires setting parameters, and the selection of the parameters also affects the robustness of the features.

(3) The evaluation index of the algorithm only adopts precision as the evaluation index, and the index cannot evaluate the recall capability and the sequencing capability of the classifier.

The lightning whistle wave identification is based on data from an AKEBONO satellite and an Arase satellite, respectively. The identification of lightning whistle waves (as shown in figure 5) based on the zhangheng-1 satellite data was not investigated.

Zhangheng satellite I is the first satellite in China to observe electromagnetic information related to earthquake activities. The revisiting period is 5 days, about 15 tracks of observation are carried out every day, the observation of a detailed inspection mode is carried out in a south latitude and a north latitude of 65 degrees, in a region of China continent and periphery 1000km and two seismic zones (a Pacific seismic zone and an Eurasia seismic zone) all over the world, the other regions are inspection modes, the carried SCM obtains the induction magnetic field data of an ionosphere through a Faraday electromagnetic induction law, and only the power spectrum data is obtained in the inspection modes. The satellite is observed in orbit for more than 2 years, and a large amount of waveform and power spectrum data of a global electromagnetic field are collected, wherein 3 components X/Y/Z of the SCM comprise 3 frequency bands ULF/ELF/VLF, the frequency point range ULF is 1Hz-200Hz, and ELF: 200Hz-2.2kHz, VLF 12.5Hz-25.6kHz, the sampling rate of original data is 51.2kHz, the frequency point interval ULF of power spectrum data is 0.25Hz, ELF is 2.5Hz, VLF:12.5Hz, the detailed review mode VLF waveform data 80ms contains 4096 points (Wang et al, 2018; Fan et al, 2018; Wang et al, 2018), producing about 10G of data per day, how it is particularly critical and urgent to automatically identify a lightning whistle wave event from such voluminous observation data.

At present, research on automatic identification of lightning whistle sound waves based on Zhang Heng satellites is not developed, the robustness of an automatic identification algorithm of the lightning whistle sound waves based on other constellations is poor, and how to create an automatic identification method of the lightning whistle sound waves based on Zhang Heng satellites so as to improve the robustness and the identification performance of characteristics belongs to one of the current important research and development subjects.

Disclosure of Invention

The invention aims to provide a lightning whistle sound wave automatic identification method based on Zhang Heng satellite, which can improve the robustness and identification performance of the characteristics.

In order to solve the technical problems, the invention adopts the following technical scheme:

a Zhang Heng I induction magnetometer data lightning whistle sound wave automatic identification method comprises the following steps:

collecting samples: the method comprises the following steps of (1) making an SCM data into a Fourier spectrogram, segmenting the spectrogram to complete sample image collection according to the obvious L morphological characteristics of lightning whistle sound waves in the spectrogram, and collecting a lightning spectrogram and a non-lightning sample spectrogram;

graying and scaling: carrying out graying processing and scale reduction on the frequency spectrum image so as to reduce the calculation dimension and strengthen the lightning characteristics;

fuzzy convolution processing: designing a fuzzy convolution kernel, and carrying out convolution calculation on the image to filter out the influence of a large amount of step edge information;

l form convolution processing: designing an L-form convolution kernel based on the L-form characteristics of the lightning whistle sound waves, and performing convolution processing on the image to further enhance the L-form characteristics in the image;

classification and identification of SVM models: and inputting the enhanced image into a Support Vector Machine (SVM) for training and classification recognition to obtain a recognition result.

As a further improvement of the invention, the SCM data is mainly waveform data of VLF wave band of SCM load of Zhang balance one satellite three orbits, a sliding window is designed in the collection process, the width of the sliding window is 10 s and the step size is 2s, and the sliding window is adopted to intercept data from power spectrum data at 1000Hz-6000 Hz.

Further, the graying processing is as follows:

Gray＝RGB.R×0.3+RGB.G×0.59+RGB.B×0.11

where RGB represents the original sample spectrogram, rgb.r is the pixel values of its red channel, rgb.g and rgb.b are the values of the green and blue channels, respectively, and Gray is a grayed-out spectrogram.

Further, the scaling process scales down to 50 × 50. The method mainly considers that the detailed information of the small-scale reverse image of the image is richer and the overall shape information is weaker; the overall shape information of the large-scale image is richer and the detailed information is very weak, which just accords with the image characteristics of the lightning whistle sound wave, namely the overall L-shaped morphological characteristics are presented.

Further, the fuzzy convolution kernel is a full 1 convolution template with the scale of 5 × 5, and the template mainly plays a role in performing noise reduction processing on an image.

Further, the L-shaped convolution kernel is a scale 9 × 9L-shaped convolution kernel.

Furthermore, in the SVM model classification recognition, a polynomial kernel function k (x) is adopted by the SVM_i，x_j)：

Wherein x is_jFeature vector, x, representing the jth image_iA feature vector representing an ith image; gamma and r are parameters to be adjusted, and the SVM is from an SVC library of Python; where the optimal parameter d is 13, and both γ and r use the default parameters under the library.

Further, in the classification and identification of the SVM model, the adopted evaluation indexes are as follows: accuracy, recall, F1 values, and AUC-ROC were identified.

The invention also provides a device for automatically identifying the data lightning whistle sound wave of the Zhanghenyi first induction magnetometer, which comprises:

a sample collection module: the method is used for making an SCM data into a Fourier spectrogram, segmenting the spectrogram to finish sample image collection according to the obvious L morphological characteristics of lightning whistle sound waves in the spectrogram, and collecting a lightning spectrogram and a non-lightning sample spectrogram;

graying and scale processing module: the system is used for carrying out graying processing and scale reduction on the frequency spectrum image so as to reduce the calculation dimension and strengthen the lightning characteristics;

a fuzzy convolution processing module: the method is used for designing a fuzzy convolution kernel and carrying out convolution calculation on the image so as to filter the influence of a large amount of step edge information;

an L form convolution processing module: the method is used for designing an L-shaped convolution kernel based on the L-shaped characteristics of the lightning whistle sound waves, and performing convolution processing on the image to further enhance the L-shaped characteristics in the image;

an SVM model classification and identification module: and the method is used for inputting the enhanced image into a Support Vector Machine (SVM) for training and classification recognition to obtain a recognition result.

The invention also provides a Zhangheng I induction magnetometer data lightning whistle sound wave automatic identification system, which is characterized by comprising the following components: one or more processors; a storage device for storing one or more programs which, when executed by the one or more processors, cause the one or more processors to implement the aforementioned Zhang Hei Induction magnetometer data lightning Whistle Sound wave automatic identification method.

By adopting the technical scheme, the invention at least has the following advantages:

1. on the basis of satellite SCM data of Zhang Heng I, a machine learning-based automatic identification framework of the lightning whistle sound waves is explored, fuzzy convolution kernels and L-shaped convolution kernels are designed according to the characteristics of frequency spectrum data and the L-shaped characteristics of the lightning whistle sound waves, the lightning shape characteristics in a frequency spectrum graph are further enhanced, and the robustness and the identification performance of the characteristics are greatly improved.

2. The results of a large number of experiments carried out by the invention show that: the automatic lightning event identification method provided by the invention is effective, and the identification effect reaches more than 94% on indexes of precision, recall rate, F1 value (F1 score) and receiver operating characteristic Curve Area (AUC).

Drawings

The foregoing is only an overview of the technical solutions of the present invention, and in order to make the technical solutions of the present invention more clearly understood, the present invention is further described in detail below with reference to the accompanying drawings and the detailed description.

FIG. 1 is an example of an AKEBONO satellite lightning geometry;

FIG. 2 is a DEETER satellite lightning morphology legend (Parrot et al, 2015; Parrot et al, 2019);

fig. 3 is a graph of the segmentation results (Ali Ahmad et al, 2019);

FIG. 4 is a grid line plot (Ali Ahmad et al, 2019);

FIG. 5 is a lightning morphology diagram in a time-frequency plot of SCM power spectrum data for Zhang Heng satellite number one;

FIG. 6 is a sample collection flow chart;

fig. 7 is a sample illustration: (a) lightning sample; (b) no lightning sample;

FIG. 8 is a recognition algorithm (solid lines are model training streams; dashed lines are model recognition streams);

FIG. 9 is a color spectrum (with lightning);

FIG. 10 is a grayed-out spectrogram (with lightning);

FIG. 11 is a raw scale graph;

FIG. 12 is a small scale view (and an enlarged detail view);

FIG. 13 is a fuzzy convolution map (a) a fuzzy convolution kernel; (b) a fuzzy convolution effect;

FIG. 14 is an L-feature convolution map (a) an L-feature convolution kernel; (b) convolution effect;

figure 15 is a lightning recognition diagram: (a) correctly identifying; (b) error identification; (c) a grayed image;

figure 16 is a non-lightning recognition diagram: (a) correctly identifying; (b) error identification; (c) a grayness map (b); (d) a fuzzy convolution map; (e) an L-form convolution kernel processing graph;

fig. 17 is a box plot of accuracy, recall and F1: (a) lightning (b) non-lightning;

FIG. 18 is an AUC box plot;

FIG. 19 is an original gray scale map containing lightning information;

FIG. 20 is an image after processing with a fuzzy convolution kernel;

FIG. 21 is a feature distribution trend graph; (a) is a characteristic distribution of the original grey-scale map of lightning and non-lightning samples; FIG. 21(b) feature distributions of lightning and non-lightning samples after fuzzy convolution processing; where 1 represents a lightning sample, -1 is a non-lightning sample;

FIG. 22 is an image feature (image with lightning) extracted by different convolution kernels: (a) is an original gray map, and (b) is an image processed by a fuzzy convolution kernel; (c) b is processed by L convolution kernel with 3 × 3 scale; (d) b is the image processed by the L convolution kernel with the scale of 5 multiplied by 5; (e) b is the image processed by L convolution kernel with the scale of 7 multiplied by 7; (f) b is the image processed by the L convolution kernel with the scale of 9 multiplied by 9; (g) is the image after b is processed by L convolution kernel of dimension 11 × 11; (h) b is the image processed by L convolution kernel of scale 13 × 13; (i) b is the image processed by L convolution kernel with the scale of 15 multiplied by 15; (j) is the locally enlarged image at the frame in (c);

FIG. 23 is an image feature (non-lightning containing image) extracted by different convolution kernels, where (a) is the original gray scale image and (b) is the image after fuzzy convolution kernel processing; (c) b is processed by an L convolution kernel with the scale of 3 multiplied by 3; (d) b is the image processed by the L convolution kernel with the scale of 5 multiplied by 5; (e) b is the image processed by L convolution kernel with the scale of 7 multiplied by 7; (f) b is the image processed by the L convolution kernel with the scale of 9 multiplied by 9; (g) b is the image processed by L convolution kernel of 11 × 11 scale; (h) b is the image processed by L convolution kernel of scale 13 × 13; (i) b is the image processed by L convolution kernel with the scale of 15 multiplied by 15;

FIG. 24 is a schematic of image feature reduction to two dimensions: (a) is the image after the fuzzy convolution kernel processing; (b) is the image after the L convolution kernel of 3 x 3 of the scale processes a; (c) is the image after the L convolution kernel processing a with the dimension of 5 multiplied by 5; (d) is the image after the L convolution kernel of the scale 7 multiplied by 7 processes a; (e) is the image after the L convolution kernel processing a of the scale 9 multiplied by 9; (f) is the image after the L convolution kernel processing a of the scale 11 multiplied by 11; (g) is the image after the L convolution kernel processing a of the scale 13 multiplied by 13; (h) is the image after the L convolution kernel processing a with the scale of 15 multiplied by 15

FIG. 25 is the AUC of a classifier trained on different features;

FIG. 26 is a distribution of three metrics for classifiers trained on different features: (a) identifying a lightning sample; (b) non-lightning samples are identified.

Detailed Description

Zhang Heng I satellite has filed a large amount of induction type Magnetometer (SCM) data, and exploring the algorithm of automatically recognizing lightning whistle sound wave from it has important research significance for further summarizing the space-time change rule of space weather lightning events. The embodiment provides a Zhangheng I induction magnetometer data lightning whistle sound wave automatic identification method, which comprises the following steps:

The following is a detailed description thereof:

1 data Collection (sample Collection)

The data mainly comes from waveform data of VLF band of SCM load of Zhang Heyi satellite in three orbits of 2019 and month 8, the collection process is shown in FIG. 6, and a power spectrum diagram is obtained by Fourier transform of the data. According to the data: most lightning whistle sound waves have an L shape in the 1000Hz-6000Hz region and do not last for more than 20s (fisher et al, 2010), so this embodiment designs a sliding window with a width of 10 2s and a step length of 2s, and intercepts data from the power spectrum data by using the sliding window for 1000Hz-6000Hz, and obtains 8316 data: of which 316 lightning whistle sound data and 8000 non-lightning whistle sound data, a sample example is shown in fig. 7.

2 lightning whistle sound wave automatic identification algorithm

The deep learning model is the mainstream model of machine learning at present, but still needs a large amount of data samples as a basis. However, the amount of lightning data collected in the present embodiment is relatively small, and based on this, a traditional recognition algorithm technology is still used to explore a recognition scheme of lightning whistle sound waves, where the scheme includes a training process and a recognition process, as shown in fig. 8: the solid line is the training procedure and the dashed line is the recognition procedure. The main purpose of the training process is to obtain a lightning recognition model, and the main purpose of the recognition process is to apply the lightning recognition model. The training process comprises the following steps: graying processing, scale processing, fuzzy convolution processing, L-form convolution processing and SVM model training. The identification process comprises the following steps: graying, scaling, fuzzy convolution, L-form convolution and lightning identification by using SVM model.

2.1 graying

The original sample spectrogram is shown in fig. 9, and the lightning identification is observed to be mainly based on the L morphological characteristics, so that the influence of color is eliminated by using a graying process. The effect of graying the image of fig. 9 according to the formula (3) is shown in fig. 10

Gray＝RGB.R×0.3+RGB.G×0.59+RGB.B×0.11 (3)

2.2 Scale treatment

The image is observed to have larger scale and more obvious detail, but the overall morphological characteristics are not outstanding enough, and the smaller scale is easy to bulge the overall morphological characteristics. As shown in fig. 11 and 12: fig. 11 is the original scale and fig. 12 shows the effect of reducing it to 1/4, and the L-shaped feature is found more clearly than in fig. 11. Therefore, the present embodiment chooses to reduce the scale to 50 × 50, which not only can highlight the morphological features but also reduces the data dimension for better calculation.

2.3 fuzzy convolution

Observing fig. 13 shows that the image has a more severe striped texture, indicating that the image has a distinct step edge, and the edge information is not favorable for identifying the L-shaped feature. Therefore, the fuzzy convolution kernel is designed to perform convolution processing on the graph 12 in the embodiment, so as to weaken the step edge. The convolution kernel design is shown in fig. 13(a), and is a full 1 template with a scale of 5 × 5, and the convolution processing is performed on fig. 12 using this, and the processing result is shown in fig. 13 (b).

2.4L feature extraction

The lightning whistle sound wave takes a distinct L shape, and in order to further highlight the lightning characteristic, a convolution kernel having a size of 9 × 9L is designed according to the characteristics of the L shape of lightning, as shown in fig. 14(a) (as to why the scale of the convolution kernel is 9 × 9, which will be analyzed in detail in the discussion section), and the result of processing using this convolution kernel to fig. 13(b) is shown in fig. 14 (b).

2.5 SVM classification

And inputting the convolved image into an SVM classifier for training and recognition. SVM is a supervised classifier based on statistical theory (Sanchez and David, 2003). The categories in this embodiment are mainly two: lightning and non-lightning. First, the image data is divided into a training set and a test set.

The images of the training set are represented as: x is the number of₁，x₂，…，x_nWith the corresponding feature label y₁，y₂，…，y_n， y_i∈[-1，1]

-1 indicates that the sample is a non-lightning sample, and 1 indicates that there is a lightning event in the sample.

The main goal of the linear SVM classifier is to find a hyperplane w ═ w (w)₁，w₂，…，w_nWhich can classify-1 and 1. The mathematical model of the hyperplane is shown in equation (4):

wx+b＝0 (4)

where x is the sample point on the hyperplane and b is the offset.

If the lightning sample correctly falls in the lightning classification area, the method has the formula (5)

y_i(wx_i+b)≥+1 (5)

If the non-lightning sample is correctly located in the non-lightning classification area, the method has the formula (6)

y_i(wx_i+b)≤-1 (6)

Both equation (5) and equation (6) may be combined into one equation (7):

y_i(wx_i+b)-1≥0 (7)

there are many hyperplanes that perform classification tasks, and the hyperplane found by SVMs maximizes the separation in feature space. Its mathematical model can be expressed as formula (8):

max 1/||w|| s.t.y_i(wx_i+b)-1≥0(i＝1，2，...，n) (8)

the goal of training the model is to find w that satisfies the above equation.

Since in most cases the samples are not linearly separable, and thus the hyperplane satisfying the above condition does not exist, the SVM provides a kernel function k (x) for this problem of non-linear separability_i，x_j) Instead of equation (5), the data can be mapped to a high-dimensional space such that it is linearly separable in the high-dimensional space (Sanchez and David, 2003). The kernel function in this embodiment is a polynomial kernel function (9):

wherein x is_jFeature vector, x, representing the jth image_iA feature vector representing an ith image; gamma and r are parameters to be adjusted, and the SVM is from an SVC library of Python; where the optimal parameter d is 13, and γ and r all use the default parameters under the library.

3 experiments and analyses

3.1 protocol

Because the lightning samples are not balanced with the non-lightning samples, according to the traditional machine learning practice, the proportion of positive and negative samples cannot exceed 1: 3. the number of lightning samples in this embodiment is 316, and therefore 1000 non-lightning samples corresponding thereto should be selected. Whereas in practice the present embodiment obtains 8848 non-lightning samples. In order to test whether the proposed lightning recognition algorithm is valid, the following experimental scheme is made.

Data set: the data set consists of 316 lightning samples and 1000 non-lightning samples. The 1000 non-lightning samples are randomly selected from 8848 non-lightning samples, which results in 316 lightning sample sets and 1000 non-lightning sample sets.

Training set: randomly taking 80% from the lightning sample set as training samples and also randomly taking 80% from the non-lightning sample set as training samples, the process results in a training set of lightning samples and a training set of non-lightning samples.

Feature extraction: aiming at a training set, three different feature extraction methods are used for extracting image features of a lightning sample training set and a non-lightning training sample set. The embodiment provides three image feature extraction modes:

(1) and original Gray level image characteristics are expressed by Gray.

(2) The characteristics of the original Gray-scale image after the fuzzy convolution processing are represented by Gray _ Blur.

(3) Firstly, fuzzy convolution processing is carried out on an original Gray level image, and then image characteristics after L characteristic convolution processing are adopted and are represented by Gray _ Blur _ L.

Training process: and respectively training the SVM models by using different characteristics to obtain three different SVM lightning recognition models.

And (3) test set: and taking the rest samples in the lightning sample set as test samples, and taking the rest samples in the non-lightning sample set as training samples, and obtaining a test set of the lightning samples and a test set of the non-lightning samples in the process.

Feature extraction: in the same manner as the feature extraction above.

The identification process comprises the following steps: and respectively putting the three different characteristics into different SVM models for identification, and outputting an identification result.

Evaluating the recognition effect: four indexes are adopted for evaluating the recognition effect: precision (Precision), Recall (Recall), F1 value, and ROC Area (AUC).

The number of experiments: 10000 times

3.2 SVM parameter selection

The SVM in the present algorithm comes from the SVC library of Python. The selected kernel function is a polynomial kernel as shown in equation (9). The optimal parameter d obtained by cross-validation is 13, and the default parameters under the library are used for the rest.

3.3 evaluation index

Four different evaluation indexes are adopted to evaluate the effect of each recognition, the lightning is a positive sample, the non-lightning is a negative sample, and the definition of the four indexes is explained next.

First, the symbol definitions are shown in Table 1

TABLE 1 symbol definitions

The definition of recognition accuracy (Precision) is shown in equation (10):

Precision＝TP/(FP+TP) (10)

TP + FP, which is the total lightning sample and is the number of positive pictures in the predicted pictures; TP is the number of pictures for which the positive class is also predicted. The meaning of the method is the proportion of the number of correctly predicted pictures to the total number of positive type predictions, and generally, the larger the index is, the better the index is.

The Recall (Recall) is defined as shown in equation (11):

Recall＝TP/(TP+FN) (11)

wherein TP + FN represents the number of pictures which totally meet the picture marking; TP is the number of the pictures of which the positive class is predicted as the positive class; the meaning of the method is to determine the number of the positive class predicted as the positive class picture occupying all the labeled pictures, and generally, the larger the index is, the better the index is.

The F1 value (F1-Score) is defined as shown in formula (12):

F1＝2/(1/Precision+1/Recall) (12)

the meaning is as follows: in general, Precision is high, Recall is low, Recall is high, and Precision is low. Index F1-score, taking into account the harmonic values of Precision and Recall. When Recall is unchanged, the larger the Precision, the smaller the 1/Precision, and thus the larger the F1. The same principle is that: when Precision is constant, the larger Recall, the smaller 1/Recall and thus the larger F1. The larger the index 1, the better.

AUC-ROC: higher values of the area under the ROC curve mean better sorting performance of the classification.

3.4 evaluation strategy

On a training set and a test set of each experiment, three different feature extraction methods are used for extracting image features, three different classifiers are obtained through training, and the performance of each classifier is evaluated by adopting precision, recall rate, F1SCORE and ROC-AUC. Because the training set and the test set at each time are different, the effect of the lightning recognition algorithm provided by the embodiment is difficult to fully evaluate by using the single four evaluation indexes, so that the experiment is carried out 10000 times, and the following evaluation strategy is formulated on the basis of the four evaluation indexes:

(1) and displaying a part of recognition results.

(2) And (3) evaluating the overall recognition accuracy: and (3) evaluating strategies for averaging the evaluation indexes of 10000 experiments.

(3) Evaluation of stability of recognition effect: and for the evaluation indexes of 10000 experiments, a boxed graph is adopted to evaluate the stability of 10000 classifications.

(4) Identification effect difference evaluation: in order to evaluate whether the classification effect caused by different characteristics has obvious difference, T test is adopted to evaluate the difference of the classification effect of different characteristics. A threshold p of 0.05, i.e., less than 0.05, indicates a significant difference, and a threshold p of 0.05 indicates no significant difference.

3.4.1 partial identification results presentation

The partial recognition results are shown in fig. 15 and 16: fig. 15 is a partial result of identifying a lightning image, (a) is a lightning image that can be correctly identified, and (b) is a lightning image that identifies an error. All of (a) and (b) are mistakenly identified because the L morphological feature in (b) is much weaker than the "fizzing" feature on the right, as shown in (c), when the main information of the picture is reflected on the fizzing feature, the classifier has a false judgment; fig. 16 is a partial result of identifying a non-lightning image, (a) is a result of being able to correctly identify a non-lightning image, and (b) is a non-lightning image that identifies an error. The reason for the recognition error is that multiple "roasing" appearing in the image causes an artifact of the L morphological feature, as shown in (c) (d) (e), the classifier has a false positive.

3.4.2 Total recognition accuracy evaluation

The section mainly aims at 10000 times of Precision, Recall, F1score and AUC-ROC averaging to measure lightning identification effects under different image characteristics. The results are shown in tables 2 and 3.

TABLE 2.10000 average Effect after experiments

TABLE 3 mean values of ROC area AUC

It can be found by observing tables 2 and 3 that:

(1) the classification precision of the features extracted by the fuzzy convolution kernel is obviously superior to the effect of the original gray level image. For example, Precision for Gray features identified is 0.732, while Precision for classes using fuzzy convolution kernels is 0.915; meanwhile, the classification Precison after the characteristics are extracted by adopting an L-shaped convolution kernel is up to 0.945.

(2) The classification recall rate of the features extracted by the fuzzy convolution kernel is obviously superior to the effect of the original gray level image. For example, the value of Recall of Gray is 0.610, while the value of Recall after fuzzy convolution is 0.911, and the classification Recall rate after feature extraction by using an L convolution kernel is obviously the highest and is 0.974.

(3) The F1score of the feature extracted by the fuzzy convolution kernel is also superior to the effect of the original gray image. For example, Gray's F1score has a value of 0.664, while the value after fuzzy convolution is 0.912, and the F1score after feature extraction using the L convolution kernel is highest at 0.958.

(4) The classifier after extracting features by using the convolution kernel has better sorting capability than the effect of the original gray image, as shown in table 3, the AUC of the original gray image classification is 0.882, the AUC of the fuzzy convolution classification is 0.975, and the AUC of the L convolution kernel is 0.989.

3.4.3 evaluation of stability of recognition Effect

In order to evaluate whether the performance of the classifier is stable, 10000 pieces of data per index were plotted using a box plot, and the results are shown in fig. 17 and 18, which mainly qualitatively evaluate the stability of the classification performance.

(1) Distribution of 10000 sets of data of lightning recognition accuracy (Precision) is shown in a Precision map of fig. 17 (a): the horizontal axis is the different image characteristics and the vertical axis is the accuracy. The following are found: the distribution of the non-lightning recognition accuracy of the Gray _ Blur and Gray _ Blur _ L features constitutes a box which is lower than the lightning recognition accuracy of the Gray feature. This phenomenon illustrates that the Gray _ Blur and Gray _ Blur _ L features make the performance of the classifier more stable than the Gray features. The case corresponding to the Gray _ Blur _ L characteristic is the lowest, and the characteristic extraction method is explained to enable the performance of the classifier to be the most stable. Observing the median line at the same time further finds: the middle bit lines of the boxes corresponding to Gray _ Blur _ L and Gray _ Blur are higher than those of the boxes of Gray, which indicates that the precision large probability of Gray _ Blur _ L and Gray _ Blur is higher than that of Gray characteristics; the accuracy of the Gray _ Blur _ L feature learning lightning classifier is optimal. The same properties as described above are also shown in the distribution of the Recall and F1score indices.

(2) Observing the distribution data of the non-lightning recognition accuracy in the above-described comparative manner, as shown in fig. 17(b), it was found that: the data boxes of the lightning identification Precision of the Gray characteristics are higher than the non-lightning identification Precision of the Gray _ Blur and Gray _ Blue _ L image characteristics, and the median lines are higher than the median lines of the Precision boxes of the Gray level images, so that the Gray _ Blue _ L characteristics have better Precision and smaller fluctuation than the original Gray level characteristics when the non-lightning identification is carried out.

(3) Also observed is a box plot of 10000 values of the AUC indicator for the classifier, as shown in fig. 18: the horizontal axis represents different feature extraction methods, and the vertical axis represents AUC values. The following are found: the bin of the AUC for the Gray feature is much higher than the bin of the Gray _ Blur feature and the bin of the convolution kernel feature for the Gray _ Blue _ L feature, where the AUC bin of the Gray _ Blue _ L feature is not only minimal but also highest in number of bits. It is shown that the classifier trained by the Gray _ Blue _ L feature is not only less fluctuating but also has the best sorting capability.

In summary, the classifier obtained by the Gray _ Blur _ L image feature learning not only has a more reliable classification effect, but also is more stable than the classification performance of the original Gray feature.

3.4.4 evaluation of significance difference in recognition Effect

In order to test whether the performance of the classifier has obvious differences under different image characteristics, a T test method is adopted to carry out quantitative evaluation on the significance differences, the higher the significance P value is, the smaller the significance difference is, usually the threshold value is 0.05, and the meaning is that if the difference is less than 0.05, the significant difference is considered to exist; if the difference is more than 0.05, no obvious difference exists between the two experiments. The results are shown in tables 4, 5, 6 and 7.

(1) Focusing on Precision index for identifying lightning, performing T test on 10000 precisions of Gray feature classification, 10000 precisions of Gray _ Blur feature classification and 10000 precisions of Gray _ Blur _ L feature classification in pairs to obtain different test values, as shown in Table 4, wherein the Precision of classifiers trained by different features is 0, for example, the P value of the T test is 0 when performing T test on data of Gray and Gray _ Blur, and the data shows that: there is a clear difference between the obtained classification accuracy of the Gray feature and the Gray _ bur feature. For example, a P-value of 1 for T-tests on Gray and Gray, a P-value of 1 indicates no significant difference between the data. And performing T test on the recognition accuracy of the Gray _ Blur _ L characteristic and the recognition accuracy of the other two characteristics to obtain that the P values are both 0, and also showing that the classifier trained by the Gray _ Blur _ L characteristic has obvious difference in the accuracy of the lightning classification compared with the classifiers trained by the other two characteristics.

(2) Observing the Recall index of the lightning recognition, the P value of the T test between the Recall rate of the Gray characteristic and the Recall rate of the Gray _ Blur characteristic is 0, which shows that the classifiers of the lightning recognition learned by the two classifiers have obvious difference in the Recall rate. The effect of identifying the lightning of the Gray _ Blur _ L characteristic and the T-test values of the other two characteristics are both 0, which also shows that the lightning classification effect has obvious difference in recall rate due to the three different characteristics.

(3) By observing the F1score index in the above manner, a comparison similar to the above can be obtained. There is a significant difference between the F1 values of the classifiers learned for the three features. And observing the index AUC, wherein the significance test P between the AUCs of the classifiers learned by the three characteristics is 0, which shows that the sorting capability of the classifiers learned by the three different characteristics has obvious difference.

(4) The above conclusions also exist on the recognition performance of non-lightning pictures.

In a word, the lightning identification is carried out by adopting the Gray characteristics, the Gray _ Blur characteristics and the Gray _ Blur _ L characteristics, the identification effect is obviously different in precision, recall rate, F1 value and AUC value, and the classification effect of the Gray _ Blur _ L characteristics is optimal.

Discussion 4

The experiment shows that the algorithm research of automatically recognizing lightning whistle sound waves is developed based on Zhang Heng satellite SCM data, and the method has a certain effect. The fuzzy convolution kernel and the L-feature convolution kernel in the algorithm scheme have a very important influence on lightning identification. This section will carry out a more thorough discussion and analysis of the effects that it has.

4.1 fuzzy convolution kernels

The experimental result shows that the effect of adopting the fuzzy convolution kernel to extract the image characteristics for lightning classification is superior to the effect of adopting the original spectrogram as the characteristics, and the part can discuss the characteristics from the visual observation angle and the characteristic separable angle.

4.1.1 visual observation Angle

The original spectrogram has a distinct vertical information trace, as shown in fig. 19: the existence of a large amount of vertical information has a great influence on the identification of the L-shape feature, and fig. 20 is an image in which the vertical information is blurred by the blur convolution kernel, and the vertical information is weakened while the L-shape feature is highlighted.

4.1.2 Angle over which the feature can be divided

To explore whether the image characteristics after the fuzzy convolution processing and the characteristics of the original gray-scale image have different separability. This section will use PCA dimensionality reduction to convert the image data of 316 lightning samples and 1000 random non-lightning samples into 2 dimensions and map their distribution in 2-dimensional space to see the separability trend of the features, as shown in fig. 21. Observing fig. 21, it was found that: in the case where a large number of positive and negative samples are gathered together in the original image feature, especially in the circular area in fig. 21(a), these superimposed positive and negative samples increase the difficulty of identifying lightning and non-lightning, resulting in a decrease in the performance of the classifier in identifying lightning; in contrast, in the circular area of fig. 21(b), the number of positive and negative samples gathered together is greatly reduced compared with the former, which improves the distinction between positive and negative samples, increases the distinction degree between lightning and non-lightning image features, and enhances the lightning identification capability of the classifier.

4.2L feature convolution kernel effects on lightning identification

Based on the fuzzy convolution processing image, the lightning classification effect is better by adopting 9 x 9L characteristic convolution kernel processing, and then the recognition effect and the scale selection of the L characteristic convolution kernel are discussed in depth from two aspects.

4.2.1 visual observation Angle

It is found from fig. 22 that the L feature extracted by the blur kernel is more conspicuous than the original image, as shown in fig. 22(a) and 22 (b); further observing, the L feature extracted by using the 3 × 3L convolution kernel is shown in fig. 22(c), and since it is too dark, the rectangular area of fig. 22(c) is enlarged as shown in fig. 22(j), and it is found that most of the detail information is filtered out, and only the main contour of the L feature is retained; as the L convolution kernel scale becomes larger, more and more detail information is around the L feature, as shown in fig. 22(d, e, f); with the scale of the L convolution kernel further increasing, not only the detail intensity around the L feature becomes more and more obvious, but also more details appear at places irrelevant to the L region.

Looking further at FIG. 23, the small scale L convolution kernel can filter out detail information, as shown in FIG. 23(c) (d) (e); with the increasing scale of the L convolution kernel, the detail information becomes more and more obvious, as shown in FIG. 23(f) (g); when the scale is further increased, the amplified interference information generates a form similar to the L feature, such as a red circular area in fig. 23(i), i.e., a large-scale L convolution kernel easily amplifies the noise in the area, and easily brings certain trouble for identification. Therefore, the small-scale L convolution kernel filters out too much detail information around the L feature, the too large L convolution kernel is easy to amplify noise in the non-lightning image, and how to automatically select the L convolution kernel with the proper scale is a problem to be considered further.

4.2.2 characteristic divisible Angle

In order to qualitatively compare separability of different features, firstly, image features of 316 lightning samples and 1000 random non-lightning samples are extracted in different feature extraction modes, feature dimensions are reduced to 2 dimensions by adopting a PCA (principal component analysis) technology, and finally the features are drawn in a two-dimensional space to observe distribution conditions of corresponding features and analyze separability trends of the features, wherein +1 represents the lightning sample data, and-1 represents the non-lightning sample data, as shown in FIG. 24.

As can be seen from fig. 24: the image features extracted by the fuzzy convolution kernel are as shown in fig. 24(a), although most of positive and negative samples are separated as a whole, the positive and negative samples are still overlapped in the middle; fig. 24(b) is a graph showing the effect of two-dimensional distribution of features obtained by processing an image by using a 3 × 3L convolution kernel by first processing the image by using a fuzzy convolution, and the distances between positive and negative samples are increased as compared with fig. 24(a), and the discriminativity between positive and negative classes is better than that of fig. 24(a) as the scale of the L convolution kernel increases, but the difference is not large as compared with fig. 24(b), as shown in fig. 24(c) (d) (f); it is worth mentioning, however, that there is an increasing accumulation of lightning samples within class, indicating that the intra-class difference of lightning samples is smaller and smaller, which is advantageous for the identification of lightning events, as shown in fig. 24 (f); as the scale of the L-feature convolution kernel is further increased, it is found that there is a phenomenon in which a part of the lightning sample and the non-lightning sample overlap each other, as shown by the circular area in fig. 24(g) (h). And this overlapping phenomenon becomes more and more serious as the scale increases, as shown in fig. 24 (h). Furthermore, the non-lightning samples are increasingly scattered in the class, which indicates that the non-lightning samples are increasingly poor in the class, which is not favorable for identifying the non-lightning class.

In conclusion, the appropriate L convolution kernel scale can better realize the classification effect between the lightning samples and the non-lightning samples. Although the classification effect can be improved by too small a scale, the large lightning intraclass difference is not beneficial to the identification of lightning samples; too large a scale easily causes overlap of non-lightning and lightning samples, resulting in smaller and smaller inter-class variance; also leading to large intra-class variance of non-lightning samples; small inter-class variance and large intra-class variance will not facilitate efficient classification by the classifier.

4.2.3 statistical Box Angle

Image features extracted by using L convolution kernels of different scales are utilized, a classifier capable of identifying lightning is trained, an AUC value reflecting the sorting capability of the classifier is calculated, the experiment is repeated for 10000 times, the distribution condition of AUC data of the classifier under the features is obtained, and a box diagram is shown in FIG. 25: the horizontal axis shows different feature extraction methods, and the leftmost image feature extracted by the blur convolution kernel is followed by the 3 × 3L form convolution kernel, the 5 × 5L form convolution kernel, …, and the 15 × 15L form convolution kernel in this order. The bins for the AUC of the original gray scale features were found to be much higher than the bins based on the L-morphology convolution kernel. Especially when the convolution kernel is 13 × 13, its AUC bin is not only minimal but also the highest number of bits. The classifier obtained by training the features extracted by the L convolution kernel with the scale of 13 multiplied by 13 is explained, and the sorting capability of the classifier is not only most stable and optimal.

Similarly, training a classifier based on features extracted by L convolution kernels of different scales to obtain three index values of lightning identification Precision (Precision), recall ratio and F1-value of the classifier, repeating 10000 times of experiments, wherein data distribution is as shown in fig. 26(a), wherein blu represents the features extracted by using fuzzy convolution kernels, L3 represents 3 × 3L-shaped convolution kernels, and so on, and L15 represents 15 × 15L-shaped convolution kernels. The observation shows that: in the process of identifying the lightning image, the classifiers without extracting features by using the L convolution kernels have a higher Recall index box than the box with the L-shaped convolution features, particularly at the position of L9 in the 2 nd graph of FIG. 26(a), which means that the scale of the convolution kernels is 9 x 9, and meanwhile, the median line of the position is higher than that of the boxes at other positions, which indicates that the performance of the classifier trained by using the features extracted by the scale convolution kernels not only fluctuates minimally but also has optimal precision in Recall rate. It was also found that the boxes for the Recall data were in the L3 and L15 positions, the box height became high, and the median line declined. Which means that an L-shaped convolution kernel with too small a scale, e.g. a scale of 3 x 3, and a convolution kernel with too large a scale, e.g. an L-shaped convolution kernel with a scale of 15 x 15, will degrade the performance of the lightning identification. The same properties as described above are also shown on the F1score distribution plot. But the accuracy does not vary much. FIG. 26(b) presents the data for three indicators in identifying non-lightning images: on the precision index, the median line of the L9 position reaches the highest, which indicates that the convolution kernel effect of the scale is the best on the position, and compared with the position, the minimum scale (L3) and the maximum scale (L15) have no ideal effect; on the recall index, the convolution kernel scale has little influence on the recall index; at F1score, the minimum scale convolution kernel works poorly.

In a word, the image features extracted based on the convolution kernel not only have better classification performance, but also have smaller fluctuation than the original gray level, and the classification performance is more stable.

In conclusion, the invention develops exploration and research for automatically identifying the lightning whistle sound wave on SCM data of Zhang Heng satellite I. A fuzzy convolution kernel and an L-shaped convolution kernel are designed according to the characteristics of frequency spectrum data and the L-shaped characteristics of the lightning whistle sound waves, the lightning morphological characteristics in a frequency spectrum diagram are further enhanced, and an SVM classifier is adopted to classify the characteristics, so that the recognition performance, the recall rate and the sequencing capacity of the characteristics are all over 94%.

In addition, this embodiment also provides an automatic identification apparatus of a zhanghenyi induction magnetometer data lightning whistle sound wave corresponding to the above method, including:

This embodiment still provides a No. one response magnetometer data lightning whistle sound wave automatic identification system of Zhanghenyi, includes: one or more processors; a storage device for storing one or more programs which, when executed by the one or more processors, cause the one or more processors to implement the aforementioned Zhang Hei Induction magnetometer data lightning Whistle Sound wave automatic identification method. Since the hardware parts are conventional in the art, they are not described in detail here.

The above description is only for the purpose of illustrating the preferred embodiments of the present invention and is not to be construed as limiting the present invention in any way, and it will be apparent to those skilled in the art that the above description of the present invention can be applied to various modifications, equivalent variations or modifications without departing from the spirit and scope of the present invention.

Claims

1. A Zhang Heng I induction magnetometer data lightning whistle sound wave automatic identification method is characterized by comprising the following steps:

2. The method for automatically identifying data lightning whistle sound waves of Zhang Heyi induction magnetometer No. 1 is characterized in that the SCM data mainly comes from waveform data of VLF wave band of SCM load of Zhang Heyi satellite three orbits, a sliding window is designed in the collection process, the width of the sliding window is 10 s and 2s are steps, and the sliding window is adopted to intercept data from power spectrum data at 1000Hz-6000 Hz.

3. The method for automatically identifying data lightning whistle sound waves of Zhang Heng I induction magnetometer according to claim 1, wherein the graying processing is performed according to the following formula:

Gray＝RGB.R×0.3+RGB.G×0.59+RGB.B×0.11

4. The method for automatically identifying data lightning whistle sound waves of Zhang Heng I induction magnetometer of claim 1, wherein the scale processing is to reduce the scale to 50 x 50.

5. The method for automatically identifying data lightning whistle sound waves of Zhang Heng I induction magnetometer of claim 1, wherein the fuzzy convolution kernel is a full 1 template with a dimension of 5 x 5.

6. The method for automatically identifying data lightning whistle sound waves of Zhang Heng I induction magnetometer of claim 1, wherein the L-shaped convolution kernel is an L-shaped convolution kernel with a scale of 9 x 9.

7. The method for automatically recognizing data lightning whistle sound waves of Zhang Heng I induction magnetometer according to claim 1, wherein in the classification recognition of the SVM model, a polynomial kernel function k (x) is adopted by the SVM_i，x_j)：

Wherein x is_jDenotes the j (th)Feature vector of individual image, x_iA feature vector representing an ith image; gamma and r are parameters to be adjusted, and the SVM is from an SVC library of Python; where the optimal parameter d is 13, and γ and r all use the default parameters under the library.

8. The method for automatically recognizing the data lightning whistle sound wave of the Zhang Heng I induction magnetometer according to claim 1, wherein in the classification recognition of the SVM model, the adopted evaluation indexes are as follows: accuracy, recall, F1 values, and AUC-ROC were identified.

9. Zhang a weighing apparatus response magnetometer data lightning whistle sound wave automatic identification equipment, its characterized in that includes:

a fuzzy convolution processing module: the method is used for designing a fuzzy convolution kernel and carrying out convolution calculation on the image so as to filter out the influence of a large amount of step edge information;

an L form convolution processing module: the system is used for designing an L-shaped convolution kernel based on the L-shaped characteristics of the lightning whistle sound waves, and performing convolution processing on the image to further enhance the L-shaped characteristics in the image;

10. Zhang a weighing apparatus response magnetometer data lightning whistle sound wave automatic identification system, its characterized in that includes:

one or more processors;

a storage device for storing one or more programs,

when executed by the one or more processors, cause the one or more processors to implement the Zhang Heng I Induction magnetometer data lightning Whistle Sound wave automatic identification method according to any one of claims 1 to 8.