A wind speed identification method, device, equipment and medium based on spectrum data

Abstract

A wind speed identification method, device, equipment and medium based on spectrum data, including spectrum preprocessing of original spectrum data, forming an image matrix from spectrum data after spectrum preprocessing under each range gate, performing convolution processing on the image matrix to perform signal enhancement on spectrum data on each range gate, and obtaining an enhanced image matrix; segmenting wind spectrum data and background data in the enhanced image matrix to obtain a segmented image matrix; binarizing the segmented image matrix using different binarization thresholds, and determining a target binarized image matrix according to the area of ​​the connected area in each binarized image matrix; screening out target feature areas in the target binarized image matrix that meet the wind spectrum feature label, and performing wind spectrum solution on the spectrum data under the range gate whose distance in the weighted target feature area is greater than the first distance threshold, and obtaining the solved wind speed. The accuracy of wind speed identification is improved.

Description

Technical Field

[0001] The present invention relates to the field of laser radar technology, and in particular to a wind speed identification method, device, equipment and medium based on spectrum data. Background Art

[0002] Coherent Doppler Wind Lidar (CDWL) plays an increasingly important role in atmospheric science and meteorology. However, for the application of long-range coherent Doppler wind lidar, the accurate calculation of the radial wind spectrum in the frequency domain has always been a challenge. Some early scholars, through theoretical analysis, derived the theoretical limit relationship between radar carrier-to-noise ratio and wind speed recognition accuracy, that is, the Cramer-Rao lower limit, as well as theoretical analysis and comparison of various algorithms for spectrum estimation based on frequency domain signals. However, actual signals face more complex situations, especially the spectrum data of a single range gate is often affected by abnormal interference and strong random noise during the detection process, resulting in large errors in wind spectrum estimation due to strong interference spectrum signals even when the theoretical signal-to-noise ratio is sufficient.

[0003] Recently, many scholars have begun to try to reduce the noise of long-distance signals through more advanced algorithms to improve the recognition and analysis capabilities of weak echo signals. By using classic signal processing techniques such as wavelet transform (WT), empirical mode decomposition (EMD), variational mode decomposition (VAD), singular value decomposition (SVD), and fusion algorithms with iterative algorithms such as honey badger algorithm and neural network, the echo signals of coherent Doppler wind laser radar are processed. These methods have improved the recognition and analysis accuracy of signals to a certain extent, but there are still some problems. On the one hand, the parameter selection is subjective and the stability of the algorithm needs to be further improved. On the other hand, since the single signal of the wind laser radar is very weak and the coherence of multiple signals in the time domain is poor, it is generally adopted to convert each signal into the frequency domain and then perform multiple cumulative averaging processes to improve the signal strength. Therefore, these algorithms that process in the time domain will generate a huge amount of computational workload under a large number of cumulative times. Some people have also proposed targeted denoising algorithms for some specific interference noises, such as DC interference noise for BPD, TIANWENWEI, etc. Denoising was performed by establishing a noise model, thereby improving performance. This hardware noise can also be eliminated through hardware improvements.

[0004] In general, there are two routes for effective identification of noisy signals. One is based on noise characteristics, as mentioned in the previous study. Regardless of EMD, VAD, or SVD, noise reduction is achieved by using the high-frequency random characteristics of noise to identify and eliminate it by signal decomposition. Alternatively, the characteristics of a specific noise are known, and the noise is inferred based on the relationship between the signal and the noise using a mathematical model, and then eliminated. The other route is based on the expected signal characteristics. For example, in the early research on microwave wind profile radar, PunyaThitimajshima et al. considered the spatial relationship between wind spectra in continuous range gates, used the relative continuity criterion, and used neural networks from multiple candidate features to identify the local maximum value that is most likely to reflect the radial velocity of the atmosphere. Feature recognition and target extraction is a hot and active topic in the field of image processing, and some researchers have conducted statistical research on wind spectrum characteristics based on observational data. However, there are currently few cases of wind spectrum recognition in meteorology, especially wind laser radar.

[0005] For a single range gate spectrum, when the signal-to-noise ratio is high, the estimator can estimate the true frequency value close to 100% and is independent of the size of the intercepted frequency range. However, when the signal-to-noise ratio is small, the performance of the estimator is related to the intercepted frequency range. When the intercepted frequency range contains the average power of the signal, the smaller the k value that characterizes the size of the intercepted frequency range, the higher the probability of a good estimate by the estimator, and the expectation and standard deviation of a good estimate are close to those under the condition of a small k value and a large k value. If the impact of a bad estimate on the estimate expectation and standard deviation is taken into account, a small k value can significantly reduce the estimate standard deviation and improve the estimate accuracy, and significantly increase the probability of a valid estimate within a certain tolerance estimate deviation range (such as the deviation between the estimate and the true value is not greater than 1 point). When the intercept range is not symmetrical about the true value, there is also a deviation between the estimate expectation and the true value, and the size of this deviation is related to the degree of deviation between the center of the intercept range and the true value of the signal, but generally there is a minimum point at a smaller k value. For lidar signals, their signal strength generally always decays with the propagation of the path, and the signal-to-noise ratio of the signal continues to decline. At close distances, since the signal-to-noise ratio is sufficient, the true frequency value of the signal can be obtained by estimating the signal frequency within the entire frequency range through the DSP estimator. However, at long distances, as the signal-to-noise ratio decreases, since the true signal is not at the center of the entire frequency range, estimating through the estimator within a larger frequency range will not only result in worse estimation accuracy but also produce greater estimation deviations. Summary of the invention

[0006] The main technical problem solved by the present invention is to realize more accurate wind speed identification at a long distance.

[0007] According to the first aspect, an embodiment provides a method for wind speed identification based on spectrum data, comprising:

[0008] Acquire original spectrum data collected by the laser radar, wherein the original spectrum data includes spectrum data under a preset number of range gates, the spectrum data includes wind spectrum data and background data, the signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet a preset continuity condition;

[0009] Performing spectrum preprocessing on the original spectrum data, forming an image matrix from the spectrum data after spectrum preprocessing under each range gate, and performing convolution processing on the image matrix using a preset first convolution kernel to perform signal enhancement on the spectrum data on each range gate, thereby obtaining an enhanced image matrix; wherein the size of the first convolution kernel is determined according to the number of the range gates;

[0010] Using a preset segmentation operator to segment the wind spectrum data and background data in the enhanced image matrix to obtain a segmented image matrix; wherein the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data;

[0011] Among a plurality of preset binarization thresholds, a binarization threshold is selected in order from small to large. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in order from near to far until the first spectrum data meeting the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The binarized image matrix is ​​obtained by calculation according to the area of ​​the connected area and the spectrum area corresponding to the connected area. If the area ratio of the connected area meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped; otherwise, the next binarization threshold is continuously selected until the multiple preset binarization thresholds are traversed; wherein, a first range gate and a second range gate in the connected area are determined, the first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area, and the area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the frequency spectrum area corresponding to the connected area;

[0012] Filter out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the target feature area after weighting processing, and perform wind speed solution on the spectrum data under the range gate whose distance in the target feature area after weighting processing is greater than the preset first distance threshold to obtain the solved wind speed.

[0013] In some embodiments, the wind spectrum feature label includes that the wind spectrum is slender, the wind spectrum extends from a short distance to a long distance, and the position of the wind spectrum data in the spectrum data under each of the range gates meets a preset position requirement.

[0014] In some embodiments, the probability distribution of the original spectrum data is a Gaussian distribution, the segmentation operator is expressed as se2=[-1 / 6; 0; 1 / 3; 0; -1 / 6], and the expression of the segmented image matrix is ​​as follows:

[0015] ,

[0016] in, represents the segmented image matrix, represents the enhanced image matrix.

[0017] In some embodiments, the number of the range gates is 6, the first convolution kernel is expressed as se1=[1 / 6,1 / 6,1 / 6,1 / 6,1 / 6,1 / 6], and the expression of the enhanced image matrix is ​​as follows:

[0018] ,

[0019] in, represents the enhanced image matrix, Represented as the image matrix.

[0020] In some embodiments, the performing spectrum preprocessing on the original spectrum data to form an image matrix from the spectrum data after the spectrum preprocessing under each range gate includes:

[0021] Normalizing the spectrum data under each range gate in the original spectrum data, and forming a preliminary image matrix from the spectrum data after the normalization processing under each range gate;

[0022] The preliminary image matrix is ​​filtered using a mean filter convolution kernel to obtain a final image matrix.

[0023] In some embodiments, the expression of the preliminary image matrix is ​​as follows:

[0024] ,

[0025] in, represents the preliminary image matrix, G represents the total number of frequency points, M represents the total number of range gates, Represents the spectrum data of the i-th frequency point under the j-th range gate, Represents the spectrum data corresponding to the maximum frequency point under the jth range gate, Represents the spectrum data corresponding to the minimum frequency point under the j-th range gate. The range of i is [1, G] and the range of j is [1, M].

[0026] In some embodiments, the method further includes: using a maximum likelihood discrete spectrum peak estimation method to estimate the wind speed of the spectrum data under a range gate whose distance in the original spectrum data is less than the first distance threshold, to obtain the wind speed of the spectrum data.

[0027] According to the second aspect, an embodiment provides a wind speed identification device based on spectrum data, comprising:

[0028] A spectrum data acquisition module, used to acquire original spectrum data collected by the laser radar, wherein the original spectrum data includes spectrum data under a preset number of range gates, the spectrum data includes wind spectrum data and background data, the signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet preset continuity conditions;

[0029] A signal enhancement module, used to perform spectrum preprocessing on the original spectrum data, form an image matrix from the spectrum data after spectrum preprocessing under each range gate, and perform convolution processing on the image matrix using a preset first convolution kernel to perform signal enhancement on the spectrum data on each range gate, so as to obtain an enhanced image matrix; wherein the size of the first convolution kernel is determined according to the number of the range gates;

[0030] An image segmentation module, used to segment the wind spectrum data and background data in the enhanced image matrix using a preset segmentation operator to obtain a segmented image matrix; wherein the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data;

[0031] The image binarization module is used to select a binarization threshold from a plurality of preset binarization thresholds in ascending order. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in the order of the range gates from near to far until the first spectrum data meeting the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The binarized image matrix is ​​obtained by calculating the area of ​​the connected area and the spectrum area corresponding to the connected area. The connected area ratio of the image matrix is ​​obtained by calculating the area ratio of the connected area of ​​the image matrix. If the area ratio of the connected area meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped; otherwise, the next binarization threshold is continuously selected until the multiple preset binarization thresholds are traversed; wherein, a first range gate and a second range gate in the connected area are determined, the first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area, and the area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the frequency spectrum area corresponding to the connected area;

[0032] The wind speed identification module is used to screen out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the target feature area after weighting processing, and perform wind speed calculation on the spectrum data under the range gate whose distance in the target feature area after weighting processing is greater than the preset first distance threshold to obtain the calculated wind speed.

[0033] According to a third aspect, an embodiment provides a wind speed identification device based on spectrum data, including:

[0034] Memory, used to store programs;

[0035] The processor is used to implement the wind speed identification method by executing the program stored in the memory.

[0036] According to a fourth aspect, an embodiment provides a computer-readable storage medium, on which a program is stored, and the program can be executed by a processor to implement a wind speed identification method.

[0037] According to the wind speed identification method, device, equipment and medium based on spectrum data of the above-mentioned embodiment, the problem of huge computational complexity in time domain signal processing is avoided by performing spectrum preprocessing on the original spectrum data collected by the laser radar. At the same time, the image matrix composed of the spectrum data after spectrum preprocessing is subjected to signal enhancement and segmentation processing to obtain the segmented image matrix, determine the target binary image matrix corresponding to the segmented image matrix, screen out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area, and obtain the target feature area after weighting processing. Through the above-mentioned image processing, a more accurate target feature area can be screened out at a long distance, and the wind speed is solved by solving the spectrum data under the distance gate with a distance greater than the first distance threshold in the target feature area after weighting processing to obtain the solved wind speed, thereby improving the accuracy of wind speed identification at a long distance and a low signal-to-noise ratio. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a flow chart of wind speed identification based on spectrum data according to an embodiment of the present application;

[0039] Figure 2 A spectrum diagram of raw spectrum data of an embodiment;

[0040] Figure 3 A flowchart of performing spectrum preprocessing on original spectrum data in an embodiment, and forming an image matrix from spectrum data after spectrum preprocessing under each range gate;

[0041] Figure 4 A spectrum diagram of an original low signal-to-noise ratio spectrum signal of an embodiment;

[0042] Figure 5 A spectrum diagram of the spectrum after the adjacent range gates are superimposed and enhanced in one embodiment;

[0043] Figure 6 A graph showing changes in the proportion of white pixels in an image after binarization processing using different binarization thresholds at the maximum detection distance gate number under different speed deviation requirements in an embodiment;

[0044] Figure 7 A graph showing the change of the maximum detection distance gate number under different speed deviation requirements when using wind spectrum continuous operators of different lengths for detection in an embodiment;

[0045] Figure 8 A frequency spectrum diagram of connected regions in a binarized image matrix in one embodiment;

[0046] Fig. 9A comparison diagram of an inversion effect diagram of directly performing wind speed calculation on original spectrum data and an inversion effect diagram of performing wind speed calculation on original spectrum data using a wind speed identification method in an embodiment;

[0047] Fig.10 A reference diagram of performance indicators of a laser radar according to an embodiment;

[0048] Fig.11 The measured spectrum data of an embodiment and the data processed by the image using the wind speed identification method;

[0049] Fig.12 The measured spectrum data, signal-to-noise ratio and radial wind speed inversion results within half an hour of an embodiment;

[0050] Fig.13 The signal-to-noise ratio of measured data and the radial wind speed inversion effect statistics within half an hour in an embodiment;

[0051] Fig.14 Spectral images of different stages after image processing in a wind speed identification process of an embodiment;

[0052] Fig.15 A schematic diagram of the structure of a wind speed identification device based on spectrum data according to an embodiment;

[0053] Fig.16 The figure is a schematic diagram of the structure of a wind speed identification device based on spectrum data according to an embodiment. DETAILED DESCRIPTION

[0054] The present invention is further described in detail below by specific embodiments in conjunction with the accompanying drawings. Wherein similar elements in different embodiments adopt associated similar element numbers. In the following embodiments, many detailed descriptions are for making the present application better understood. However, those skilled in the art can easily recognize that some features can be omitted in different situations, or can be replaced by other elements, materials, methods. In some cases, some operations related to the present application are not shown or described in the specification, this is to avoid the core part of the present application being overwhelmed by too much description, and for those skilled in the art, it is not necessary to describe these related operations in detail, and they can fully understand the related operations according to the description in the specification and the general technical knowledge in the art.

[0055] In addition, the features, operations or characteristics described in the specification can be combined in any appropriate manner to form various implementations. At the same time, the steps or actions in the method description can also be interchanged or adjusted in a manner that is obvious to those skilled in the art. Therefore, the various sequences in the specification and the drawings are only for the purpose of clearly describing a certain embodiment and are not meant to be a required sequence, unless otherwise specified that a certain sequence must be followed.

[0056] The serial numbers assigned to the components herein, such as "first", "second", etc., are only used to distinguish the objects described and do not have any order or technical meaning. The "connection" and "coupling" mentioned in this application, unless otherwise specified, include direct and indirect connections (couplings).

[0057] Please refer to Figure 1 In an embodiment of the present invention, a method for wind speed identification based on spectrum data is provided, including steps S10 to S50, which are described in detail below.

[0058] Step S10: Obtaining the original spectrum data collected by the laser radar.

[0059] In some embodiments, the original spectrum data includes spectrum data under a preset number of range gates. The spectrum data includes wind spectrum data and background data. The signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet the preset continuous conditions.

[0060] Please refer to Figure 2 In some embodiments, since the laser radar characterizes the wind speed by detecting the group velocity of atmospheric aerosols, the distribution of atmospheric aerosols on the earth has spatiotemporal continuity, the movement of atmospheric aerosol particles obeys the laws of atmospheric kinematics, and the speed of atmospheric aerosols has continuity in both space and time. Therefore, when the range resolution is sufficient, the wind speed change between adjacent range gates will not be too large, which is manifested in the spectrum as the peak values ​​of the signal spectra of adjacent range gates are close, that is, the wind speed under the adjacent range gates meets the preset continuity condition. However, it should be specially noted that the change in wind speed is affected by the actual situation and it is difficult to accurately predict.

[0061] Step S20: spectrally preprocess the original spectrum data, form an image matrix from the spectrum data after spectrum preprocessing under each range gate, and use a preset first convolution kernel to convolve the image matrix to perform signal enhancement on the spectrum data on each range gate to obtain an enhanced image matrix.

[0062] Please refer to Figure 3In some embodiments, step S20 performs spectrum preprocessing on the original spectrum data, and the spectrum data after spectrum preprocessing under each range gate forms an image matrix, including steps S21 to S22, which are described in detail below.

[0063] Step S21: normalizing the spectrum data under each range gate in the original spectrum data, and forming a preliminary image matrix from the normalized spectrum data under each range gate.

[0064] In some embodiments, the expression of the preliminary image matrix is ​​as follows:

[0065] ,

[0066] in, represents the preliminary image matrix, G represents the total number of frequency points, M represents the total number of range gates, Represents the spectrum data of the i-th frequency point under the j-th range gate, Represents the spectrum data corresponding to the maximum frequency point under the jth range gate, Represents the spectrum data corresponding to the minimum frequency point under the j-th range gate. The range of i is [1, G] and the range of j is [1, M].

[0067] Step S22: Filter the preliminary image matrix using a mean filter convolution kernel to obtain a final image matrix.

[0068] In some embodiments, the preliminary image matrix is ​​convolved with a mean filter convolution kernel se0 to filter the preliminary image matrix, thereby reducing high-frequency noise in the preliminary image matrix and maintaining edge features of the image matrix while smoothing the preliminary image matrix.

[0069] In some embodiments, a convolution process is performed on the image matrix using a preset first convolution kernel to perform signal enhancement on the spectrum data on each range gate to obtain an enhanced image matrix, wherein the size of the first convolution kernel is determined according to the number of range gates.

[0070] In some embodiments, when the number of range gates is 6, the first convolution kernel is expressed as se1=[1 / 6,1 / 6,1 / 6,1 / 6,1 / 6,1 / 6].

[0071] In some embodiments, the expression of the enhanced image matrix is ​​as follows:

[0072] ,

[0073] in, represents the enhanced image matrix, Represented as an image matrix.

[0074] Please refer to Figure 4 and Figure 5 In some embodiments, the noise is random white noise. The fluctuation caused by the noise can be smoothed by superimposing the frequency spectra of adjacent range gates. Since the signal spectra of adjacent range gates are close in position, the signal power spectrum approximately obeys the Gaussian distribution in the frequency domain and has a certain spectrum width. Therefore, the image matrix can be convolved by the first convolution kernel. By superimposing the frequency spectra of adjacent range gates, the wind spectrum ranges between adjacent range gates can be enhanced, the continuity of the signal spectrum range gate is improved, and the contrast between the signal and the noise background is increased.

[0075] In some embodiments, it is worth noting that although the signal spectrum is continuous on the range gate, since the signal frequencies of adjacent range gates are actually not consistent, the signals superimposed and enhanced by each range gate may deviate from the actual signal frequency to a certain extent, thereby affecting the estimation deviation in the subsequent wind speed estimation. Therefore, it is necessary to select a suitable length of se1 to balance the range gate signal enhancement and the degree of center deviation.

[0076] Please refer to Figure 6 and Figure 7 In some embodiments, the effective detection bandwidth of the wind laser radar system is 350Mhz, the sampling frequency is 500Mhz, and a 1024-point fast Fourier transform is used to achieve time-frequency conversion, wherein the frequency shift frequency is 80Mhz. By analyzing the data of the distribution of the real wind speed and signal-to-noise ratio with distance at a distance resolution of 30m, it can be seen that when the binarization threshold is 5%, the estimator can achieve a wind speed accuracy of 0.5m / s or 1m / s to obtain the farthest detection distance. Under this accuracy requirement, it is not sensitive to the length of se1. When the length of se1 is greater than 5, it can basically meet the requirement of obtaining the farthest detection distance. A se1 that is too long will increase the subsequent calculation requirements. Therefore, se1 can be determined as se1=[1 / 6,1 / 6,1 / 6,1 / 6,1 / 6,1 / 6].

[0077] In some embodiments, Figure 6 and Figure 7 IPA0.5 refers to the maximum distance that can be detected after image preprocessing when the wind speed accuracy requirement is 0.5m / s, IPA1 refers to the maximum distance that can be detected after image preprocessing when the wind speed accuracy requirement is 1m / s, and IPA0.1 refers to the maximum distance that can be detected after image preprocessing when the wind speed accuracy requirement is 0.1m / s. Figure 6 and Figure 7The 0.5 mark means the maximum distance that can be detected without image preprocessing when the wind speed accuracy requirement is 0.5m / s, the 1 mark means the maximum distance that can be detected without image preprocessing when the wind speed accuracy requirement is 1m / s, and the 0.1 mark means the maximum distance that can be detected without image preprocessing when the wind speed accuracy requirement is 0.1m / s. Image preprocessing refers to the steps of processing the image in steps S10 to S50.

[0078] Step S30: using a preset segmentation operator to segment the wind spectrum data and the background data in the enhanced image matrix to obtain a segmented image matrix.

[0079] In some embodiments, the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data. When the probability distribution of the original spectrum data is a Gaussian distribution, the segmentation operator is expressed as se2=[-1 / 6; 0; 1 / 3; 0; -1 / 6]. The change characteristics of the Gaussian distribution of the wind spectrum data in the frequency domain and the uniformity characteristics of the background data are utilized, and the segmentation operator is used to perform gradient processing on the enhanced image matrix to segment the wind spectrum data and the background data in the enhanced image matrix to obtain the segmented image matrix.

[0080] In some embodiments, the expression of the segmented image matrix is ​​as follows:

[0081] ,

[0082] in, represents the segmented image matrix, Represents the enhanced image matrix.

[0083] Step S40: A binarization threshold is selected from a plurality of preset binarization thresholds in ascending order. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in order from near to far until the first spectrum data that meets the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The connected area area ratio of the binarized image matrix is ​​calculated according to the area area of ​​the connected area and the spectrum area corresponding to the connected area. If the connected area area ratio meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped. Otherwise, the selection of the next binarization threshold is continued until the plurality of preset binarization thresholds are traversed.

[0084] In some embodiments, the position of the wind spectrum data in the spectrum data that meets the preset wind spectrum conditions is [1, K (DSP)], where 1 refers to the first range gate. When the signal-to-noise ratio is sufficient, the position of the wind spectrum data can be obtained by the DSP algorithm, that is, K (DSP). Starting from this position, the connected region analysis method is used to solve the corresponding connected region C K , where the connected area C K The area of ​​the region is S(C K ). Determine the first range gate and the second range gate in the connected area. The first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area. Calculate the area between the first range gate and the second range gate in the binary image matrix and use it as the spectrum area S (C) corresponding to the connected area. NK ).

[0085] In some embodiments, the wind spectrum refers to the Doppler frequency shift of the radial wind speed corresponding to the frequency spectrum data detected by the laser radar, and the wind speed itself corresponds to the group motion of atmospheric particles, and the atmospheric particles also have random thermal motion. Since the random thermal motion wind spectrum is Gaussian distributed on the spectrum, the spectrum data that meets the preset wind spectrum conditions refers to the peak part of the wind spectrum data of the Gaussian distribution, wherein the peak part of the wind spectrum data corresponds to the Doppler frequency shift of the group motion of atmospheric particles.

[0086] In some embodiments, the expression of the connected region area ratio of the binarized image matrix is ​​as follows:

[0087] ,

[0088] in, represents the ratio of the area of ​​the connected region, S(C K ) represents the area of ​​the connected region, S(C NK ) represents the spectrum area corresponding to the connected region, represents the area of ​​a single range gate, Indicates the number of range gates that constitute the connected area, that is, the difference between the first range gate and the second range gate.

[0089] In some embodiments, the preset first area ratio condition is that the area ratio of the connected regions is less than or equal to 5%.

[0090] Please refer to Figure 8In some embodiments, the connected area is the area where the wind spectrum data is located when the range gate is 40 to the range gate is 100. Since the first range gate is the smallest range gate in the connected area and the second range gate is the largest range gate in the connected area, the first range gate in the connected area is determined to be 40, and the second range gate in the connected area is determined to be 100. The area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the spectrum area S (C NK ), that is, to obtain the area of ​​a single range gate , and obtain the difference between the first range gate and the second range gate as the number of range gates in the connected area, then the spectrum area corresponding to the connected area is 60 .

[0091] Step S50: Filter out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the weighted target feature area, and perform wind speed calculation on the spectrum data under the range gate whose distance in the weighted target feature area is greater than the preset first distance threshold to obtain the calculated wind speed.

[0092] In some embodiments, the wind spectrum feature label includes that the wind spectrum is slender, the wind spectrum extends from a short distance to a long distance, and the position of the wind spectrum data in the spectrum data under each range gate meets the preset position requirements. The target feature area that meets the wind spectrum feature label in the target binary image matrix is ​​screened out, and the target feature area is weighted to weaken the influence of the area outside the target feature area in the wind spectrum inversion, and the target feature area after weighting is obtained. For the spectrum data under the range gate whose distance in the target feature area after weighting is greater than the preset first distance threshold, the maximum likelihood discrete spectrum peak estimation method is used to solve the wind speed of the spectrum data to obtain the solved wind speed.

[0093] In some embodiments, the original spectrum data collected by the laser radar includes spectrum data under different distance gates, wherein the signal-to-noise ratio of the spectrum data under a close distance gate is relatively high, and the signal-to-noise ratio of the spectrum data decreases as the distance gate increases. For the spectrum data under a distance gate whose distance in the original spectrum data is less than the first distance threshold, the maximum likelihood discrete spectrum peak estimation method is directly used to estimate the wind speed of the spectrum data to obtain the wind speed of the spectrum data. For the spectrum data under a distance gate whose distance in the original spectrum data is greater than the first distance threshold, the original spectrum data is processed through steps S10 to S50 to obtain a target feature area after weighted processing, and the wind speed of the spectrum data under a distance gate whose distance in the target feature area after weighted processing is greater than the preset first distance threshold is solved to obtain the solved wind speed.

[0094] In some embodiments, the maximum likelihood discrete spectrum peak estimation method is directly used to estimate the wind speed of the spectrum data under the close range gate with a signal-to-noise ratio greater than -14dB to obtain the wind speed of the spectrum data. The blind areas under the first few range gates in the original spectrum data are ignored.

[0095] In some embodiments, the original spectrum data is image preprocessed by using the wind spectrum with high signal-to-noise ratio at a short distance, according to the prior characteristics of reasonable wind spectrum such as spatial continuity of the wind spectrum, and the possible range of the wind spectrum in the long-distance low signal-to-noise ratio area is pre-intercepted. By increasing the weight of the pre-intercepted spectrum signal, the interference of the noise spectrum outside the pre-intercepted area of ​​the wind spectrum is reduced, and the accuracy and probability of correct solution of the wind spectrum are improved. Preprocessing the data in the frequency domain avoids the huge amount of calculation in the time domain signal processing, and can be further processed based on other advanced algorithms such as time series or frequency domain. The method is used to invert the wind spectrum through simulation and measured data. The results show that for the case data, the data after image preprocessing can increase the effective detection distance of the radar by more than 1 km when the wind spectrum wind speed is inverted from the spectrum data by MLDSP. The data accuracy efficiency can be increased from about 10% to more than 40% under low signal-to-noise ratio conditions, and from 40% to more than 80%. This shows that the algorithm can improve the accuracy and reliability of radar wind spectrum inversion, so as to improve the problem of difficulty in solving the wind spectrum of long-distance low signal-to-noise ratio spectrum signals of coherent Doppler wind lidar.

[0096] In some embodiments, image preprocessing is performed directly on the data in the frequency domain, thereby avoiding the huge amount of calculation in time domain signal processing. At the same time, conventional or more advanced spectrum estimation algorithms can still be used on the basis of preprocessing, retaining the scalability of more advanced algorithms. By performing inversion analysis on the image preprocessed data, it is found that compared with the unprocessed data, the image processed data can improve the accuracy of wind speed recognition at a lower signal-to-noise ratio, thereby obtaining wind speed detection at a longer distance.

[0097] Please refer to Fig. 9 In some embodiments, in 20 random simulations, a comparison is presented between an inversion effect diagram of directly solving the wind speed on the original spectrum data and an inversion effect diagram of solving the wind speed on the original spectrum data using the wind speed identification method from step S10 to step S50.

[0098] Among them, through Fig. 9 (a) and Fig. 9 From the comparison between (b) in FIG. 1 and FIG. 2 , it can be seen that the deviation comparison image DWbIMG of the wind speed value and the true value obtained by the wind speed identification method from step S10 to step S50 is obviously better than the inversion effect image DWbML of directly solving the wind speed of the original spectrum data. Fig. 9As shown in (c), at close range and high signal-to-noise ratio, the mean MDWbIMG of the deviation between the wind speed value and the true value obtained by the wind speed identification method of steps S10 to S50 is close to the mean MDWbML of the deviation between the wind speed value and the true value obtained by the wind speed identification method without steps S10 to S50 and both are close to 0, but when the long-distance signal-to-noise ratio drops to a certain extent, for example, after 3.5 km, MDWbML begins to diverge sharply and become more than 1 m / s, and although MDWbIMG also begins to increase, it can still be stabilized within 0.3 m / s, and it is not until 5 km that MDWbIMG further diverges and becomes more than 0.5 m / s. Fig. 9 As shown in (d), the average value SDWbIMG of the deviation between the wind speed value and the true value obtained by the wind speed identification method from step S10 to step S50 increases with the distance, which is also significantly lower than the average value SDWbML of the deviation between the wind speed value and the true value obtained by the wind speed identification method without step S10 to step S50. Among them, the effective distance of SDWbML less than 1m / s is about 3.4km, while the effective distance of SDWbIMG can reach more than 5km. The above random simulation results further illustrate that the wind speed identification method from step S10 to step S50 in the frequency domain pre-extracts the approximate range of the wind spectrum, which has an improved effect on the accuracy of wind speed inversion and the effective identification distance.

[0099] Please refer to Fig.10 In some embodiments, various performance indicators of the laser radar are pre-set, and observations are started at 18:37 on the evening of June 18, 2024, on the roof of the second floor of the observation building of the low-latitude observation station of Harbin Institute of Technology (Shenzhen) in Boluo County, Huizhou City, Guangdong Province. The laser radar performs continuous radial detection at an elevation angle of 3 degrees to the north, and the laser radar acquisition system accumulates and averages the spectral data 10,000 times per second, and outputs the accumulated spectral data and radial wind speed information.

[0100] In some embodiments, a single measurement result obtained at 18:37:29 is as follows: Fig.11 As shown in (a) in the figure, since the wind spectrum has correlation within a certain continuous time, and the noise spectrum is random white noise, the signal-to-noise ratio of the wind spectrum can be further improved by further averaging the continuous second-level spectrum data. The average result of 100 consecutive second-level spectrum data from 18:37:29 to 18:39:09 is shown in Fig.11 As shown in (b) in the figure, the obvious noise spectrum is further suppressed, and the wind spectrum becomes more clearly distinguishable visually. Fig.11 (c) shows the result of image preprocessing of the spectrum data at 18:37:29. Fig.11 The signal strength of the wind spectrum area in (b) is effectively enhanced. Fig.11 (b) and Fig.11 (c) in the figure shows that image preprocessing can enhance the accuracy of extracting wind spectrum candidate areas at low signal-to-noise ratios, and effectively reduce the spectral range of wind spectrum inversion when the wind spectrum candidate areas are accurate, thereby effectively increasing the probability of accurate radial wind speed being solved. Fig.11 (d) to Fig.11 As shown in (f) in the figure, the solved radial wind speed is superimposed on the spectrum. The white dots are the results of normal ML DSP solution, and the black dots are the results of MLDSP algorithm solution after image preprocessing of spectrum data.

[0101] In some embodiments, Fig.11 (a) and Fig.11 (b) shows the partial second-level spectrum data of continuous detection for half an hour and the spectrum data after sliding average of 100 consecutive second-level spectrum data. Fig.11 (c) and Fig.11 (d) in the figure are the signal-to-noise ratio results of the second-level and average spectrum data solved by the ML DSP algorithm within half an hour. Fig.12 As shown in (a) in the figure, the signal-to-noise ratio of the averaged data can be improved by more than 5dB. Fig.11 (d) and Fig.12 As shown in (e) in the figure, when the signal-to-noise ratio of the second-level spectrum data solved by the MLDSP algorithm is lower than the 8dB threshold, that is, after about 3km, the wind speed solution results begin to make a lot of errors. The average data improves the signal-to-noise ratio by suppressing noise, as shown in Fig.11 (e) and Fig.12 In (f), the radial wind speed calculated by ML DSP can be stable up to 6 km at some times, but sometimes it shows obvious errors after 4.5 km, which is easy to be affected by the overall background. In terms of the corresponding time, these areas are actually caused by the low signal-to-noise ratio. After image preprocessing, the radial wind speed calculated by ML DSP algorithm on the second-level data is shown in the figure below. Fig.12 (g) is obviously cleaner than the one without image preprocessing, which means that there are much fewer noise points. And the probability of erroneous results is higher after 4.5 km. After 4.5 km, due to the lower signal-to-noise ratio, even the wind speed result calculated by averaging data will produce anomalies due to the influence of non-random noise. Image preprocessing of the average data before solving further avoids anomalies. The solved radial wind speed is consistent with the average wind speed on the entire path. Fig.11 The spectrum of the stroke spectrum shown in (e) is more consistent, as shown in Fig.12 As shown in (h) in the figure, there is no abnormality in the wind speed solution results during the entire measurement time and the entire path distance.

[0102] Please refer to Fig.13In some embodiments, the meanings of the English abbreviations in the figure are respectively explained as follows: SNR1: signal-to-noise ratio of single data, SNRA: signal-to-noise ratio after averaging multiple data (data averaging can reduce noise and improve signal-to-noise ratio, and is used as the true value to monitor the effects of different algorithms), mML: 1-hour average of wind speed results obtained by maximum likelihood estimation of single data, mAML: 1-hour average of wind speed results obtained by maximum likelihood estimation after averaging multiple data, mIMG: 1-hour average of wind speed results obtained by image preprocessing of single data, mAIMG: 1-hour average of wind speed results obtained by image preprocessing after averaging multiple data, stML: 1-hour average of wind speed results obtained by maximum likelihood estimation of single data. The standard deviation of the 1-hour wind speed results, stAML: the standard deviation of the 1-hour wind speed results obtained by maximum likelihood estimation after averaging multiple data, stIMG: the standard deviation of the 1-hour wind speed results estimated by image preprocessing of a single data, stAIMG: the standard deviation of the 1-hour wind speed results estimated by image preprocessing after averaging multiple data, TpML: the correct probability of the 1-hour wind speed data obtained by maximum likelihood estimation after averaging multiple data, TpAML: the correct probability of the 1-hour wind speed data obtained by maximum likelihood estimation after averaging multiple data, TpIMG: the correct probability of the 1-hour wind speed data obtained by image preprocessing of a single data.

[0103] In some embodiments, statistics are collected for half an hour of data results, such as Fig.13 (b) shows the average value of radial wind speed solved within half an hour. The result solved by second-level data without image processing deviates from the results solved by other methods at less than 3 km. This is because the radial wind speed solved by second-level data without image processing is affected by non-completely random noise, and even if the average processing is performed for a longer time, it will deviate from the result solved by the average data. Image processing can reduce the influence of non-random noise, so that the average result of radial wind speed solved by second-level data after image processing is very consistent with the average data with a higher signal-to-noise ratio before 4.5 km. This shows that image preprocessing also has advantages in eliminating non-random random noise. Fig.13As shown in (c) in the figure, the variance of the second-level radial wind speed solved after image processing within half an hour, that is, the fluctuation caused by random noise, also increases more slowly with distance than the result directly solved without image processing. The radial wind speed result solved after image processing with a variance of less than 1m / s (including the fluctuation of the true wind speed itself over time) can be maintained for more than 4km, which is close to the result of directly solving the ML DSP solution with an additional 100 consecutive average spectrum data to 4.5km, while the radial wind speed result solved without image preprocessing with a variance of less than 1m / s can only be maintained at about 2.5km. This shows the advantage of image processing algorithms in suppressing the influence of random noise in wind speed solution. Taking the radial wind speed solved by ML DSP after image preprocessing and then the average spectrum data is taken as the standard, it is considered that the radial wind speed solved by other algorithm results at the same time and distance is valid if the difference is less than 1m / s. The effective data acquisition rate within half an hour is as follows: Fig.13 As shown in (d), in the low signal-to-noise ratio area after 3 km, the effective acquisition rate of radial wind speed after image processing is about 30% higher than that of directly solving it without image processing. If the effective detection distance is determined by the data acquisition rate of 80%, the effective detection distance of radial wind speed solved directly by ML DSP algorithm without image processing is about 3 km, while the effective detection distance of radial wind speed solved by ML DSP algorithm after image processing can reach 4.3 km.

[0104] Please refer to Fig.14 In some embodiments, Fig.14 (a) is the original spectrum data. Fig.14 (b) in the figure is the initial image matrix. Fig.14 (c) in the figure is the enhanced image matrix. Fig.14 (d) in is the image matrix after segmentation. Fig.14 (e) in is the image matrix after binarization. Fig.14 (f) in is the target binary image matrix, Fig.14 (g) in the figure is the image containing the target feature area. Fig.14 (h) in the figure is an image containing the target feature area after weighted processing.

[0105] Please refer to Fig.15 In an embodiment of the present application, a wind speed identification device based on spectrum data is provided, including a spectrum data acquisition module 10, a signal enhancement module 20, an image segmentation module 30, an image binarization module 40 and a wind speed identification module 50, which are described in detail below.

[0106] The spectrum data acquisition module 10 is used to obtain the original spectrum data collected by the laser radar.

[0107] In some embodiments, the original spectrum data includes spectrum data under a preset number of range gates, the spectrum data includes wind spectrum data and background data, the signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet preset continuity conditions.

[0108] The signal enhancement module 20 is used to perform spectrum preprocessing on the original spectrum data, form an image matrix from the spectrum data after spectrum preprocessing under each range gate, and use a preset first convolution kernel to perform convolution processing on the image matrix to perform signal enhancement on the spectrum data on each range gate to obtain an enhanced image matrix.

[0109] In some embodiments, the size of the first convolution kernel is determined according to the number of range gates.

[0110] The image segmentation module 30 is used to segment the wind spectrum data and the background data in the enhanced image matrix using a preset segmentation operator to obtain a segmented image matrix.

[0111] In some embodiments, the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data.

[0112] The image binarization module 40 is used to select a binarization threshold from a plurality of preset binarization thresholds in ascending order. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in order from near to far until the first spectrum data that meets the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The connected area area ratio of the binarized image matrix is ​​calculated according to the area area of ​​the connected area and the spectrum area corresponding to the connected area. If the connected area area ratio meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped. Otherwise, the selection of the next binarization threshold is continued until the multiple preset binarization thresholds are traversed.

[0113] In some embodiments, a first range gate and a second range gate in a connected area are determined, the first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area, and the area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the spectrum area corresponding to the connected area.

[0114] The wind speed identification module 50 is used to screen out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the target feature area after weighting processing, and perform wind speed calculation on the spectrum data under the range gate whose distance in the target feature area after weighting processing is greater than the preset first distance threshold to obtain the calculated wind speed.

[0115] Please refer to Fig.16 An embodiment of the present application provides a wind speed identification device based on spectrum data, including: a memory 1000, used to store programs; a processor 1100, used to implement a wind speed identification method by executing the programs stored in the memory.

[0116] Those skilled in the art will appreciate that all or part of the functions of the various methods in the above-mentioned embodiments can be implemented by hardware or by computer programs. When all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can be stored in a computer-readable storage medium, and the storage medium can include: read-only memory, random access memory, disk, optical disk, hard disk, etc., and the program is executed by a computer to implement the above-mentioned functions. For example, the program is stored in the memory of the device, and when the program in the memory is executed by the processor, all or part of the above-mentioned functions can be implemented. In addition, when all or part of the functions in the above-mentioned embodiments are implemented by computer programs, the program can also be stored in a storage medium such as a server, another computer, disk, optical disk, flash disk or mobile hard disk, and can be downloaded or copied and saved in the memory of the local device, or the system of the local device is updated, and when the program in the memory is executed by the processor, all or part of the functions in the above-mentioned embodiments can be implemented.

[0117] The above specific examples are used to illustrate the present invention, which is only used to help understand the present invention and is not intended to limit the present invention. For those skilled in the art, according to the concept of the present invention, some simple deductions, modifications or substitutions can be made.

Claims

1. A wind speed identification method based on spectrum data, characterized in that: include: Acquire original spectrum data collected by the laser radar, wherein the original spectrum data includes spectrum data under a preset number of range gates, the spectrum data includes wind spectrum data and background data, the signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet a preset continuity condition; Performing spectrum preprocessing on the original spectrum data, forming an image matrix from the spectrum data after spectrum preprocessing under each range gate, and performing convolution processing on the image matrix using a preset first convolution kernel to perform signal enhancement on the spectrum data on each range gate, thereby obtaining an enhanced image matrix; wherein the size of the first convolution kernel is determined according to the number of the range gates; Using a preset segmentation operator to segment the wind spectrum data and background data in the enhanced image matrix to obtain a segmented image matrix; wherein the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data; Among a plurality of preset binarization thresholds, a binarization threshold is selected in order from small to large. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in order from near to far until the first spectrum data meeting the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The binarized image matrix is ​​obtained by calculation according to the area of ​​the connected area and the spectrum area corresponding to the connected area. If the area ratio of the connected area meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped; otherwise, the next binarization threshold is continuously selected until the multiple preset binarization thresholds are traversed; wherein, a first range gate and a second range gate in the connected area are determined, the first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area, and the area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the frequency spectrum area corresponding to the connected area; Filter out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the target feature area after weighting processing, and perform wind speed solution on the spectrum data under the range gate whose distance in the target feature area after weighting processing is greater than the preset first distance threshold to obtain the solved wind speed.

2. The wind speed identification method according to claim 1, characterized in that: The wind spectrum characteristic label includes that the wind spectrum is slender, the wind spectrum extends from a short distance to a long distance, and the position of the wind spectrum data in the spectrum data under each of the range gates meets a preset position requirement.

3. The wind speed identification method according to claim 1, characterized in that: The probability distribution of the original spectrum data is Gaussian distribution, the segmentation operator is expressed as se2=[-1 / 6; 0; 1 / 3; 0; -1 / 6], and the expression of the segmented image matrix is ​​as follows: ， in, represents the segmented image matrix, represents the enhanced image matrix.

4. The wind speed identification method according to claim 1, characterized in that: The number of the range gates is 6, the first convolution kernel is expressed as se1=[1 / 6,1 / 6,1 / 6,1 / 6,1 / 6,1 / 6], and the expression of the enhanced image matrix is ​​as follows: ， in, represents the enhanced image matrix, Represented as the image matrix.

5. The wind speed identification method according to claim 1, characterized in that: The performing spectrum preprocessing on the original spectrum data, and forming an image matrix from the spectrum data after the spectrum preprocessing under each range gate, includes: Normalizing the spectrum data under each range gate in the original spectrum data, and forming a preliminary image matrix from the spectrum data after the normalization processing under each range gate; The preliminary image matrix is ​​filtered using a mean filter convolution kernel to obtain a final image matrix.

6. The wind speed identification method according to claim 5, characterized in that: The expression of the preliminary image matrix is ​​as follows: ， in, represents the preliminary image matrix, G represents the total number of frequency points, M represents the total number of range gates, represents the spectrum data of the i-th frequency point under the j-th range gate, Represents the spectrum data corresponding to the maximum frequency point under the jth range gate, Represents the spectrum data corresponding to the minimum frequency point under the j-th range gate. The range of i is [1, G] and the range of j is [1, M].

7. The wind speed identification method according to claim 1, characterized in that: Also includes: The wind speed of the spectrum data under the range gate whose distance in the original spectrum data is less than the first distance threshold is estimated by using the maximum likelihood discrete spectrum peak estimation method to obtain the wind speed of the spectrum data.

8. A wind speed identification device based on spectrum data, characterized in that: include: A spectrum data acquisition module, used to acquire original spectrum data collected by the laser radar, wherein the original spectrum data includes spectrum data under a preset number of range gates, the spectrum data includes wind spectrum data and background data, the signal-to-noise ratio of the spectrum data under each range gate is inversely proportional to the size of the range gate, and the wind speeds under adjacent range gates meet preset continuity conditions; A signal enhancement module, used to perform spectrum preprocessing on the original spectrum data, form an image matrix from the spectrum data after spectrum preprocessing under each range gate, and perform convolution processing on the image matrix using a preset first convolution kernel to perform signal enhancement on the spectrum data on each range gate to obtain an enhanced image matrix; wherein the size of the first convolution kernel is determined according to the number of the range gates; An image segmentation module, used to segment the wind spectrum data and background data in the enhanced image matrix using a preset segmentation operator to obtain a segmented image matrix; wherein the corresponding segmentation operator is determined according to the probability distribution of the original spectrum data; The image binarization module is used to select a binarization threshold from a plurality of preset binarization thresholds in ascending order. Each time a binarization threshold is selected, the segmented image matrix is ​​binarized using the binarization threshold to obtain a binarized image matrix. In the binarized image matrix, starting from the range gate corresponding to the spectrum data with the highest signal-to-noise ratio, the spectrum data under each range gate is traversed in the order of the range gates from near to far until the first spectrum data meeting the preset wind spectrum condition is obtained, and the connected area corresponding to the position of the wind spectrum data in the spectrum data is solved. The binarized image matrix is ​​obtained by calculating the area of ​​the connected area and the spectrum area corresponding to the connected area. The connected area ratio of the image matrix is ​​obtained by calculating the area ratio of the connected area of ​​the image matrix. If the area ratio of the connected area meets the preset first area ratio condition, the binarized image matrix is ​​used as the target binarized image matrix, and the selection of the next binarization threshold is stopped; otherwise, the next binarization threshold is continuously selected until the multiple preset binarization thresholds are traversed; wherein, a first range gate and a second range gate in the connected area are determined, the first range gate is the smallest range gate in the connected area, and the second range gate is the largest range gate in the connected area, and the area between the first range gate and the second range gate in the binarized image matrix is ​​calculated and used as the frequency spectrum area corresponding to the connected area; The wind speed identification module is used to screen out the target feature area that meets the preset wind spectrum feature label in the target binary image matrix, perform weighting processing on the target feature area to obtain the target feature area after weighting processing, and perform wind speed calculation on the spectrum data under the range gate whose distance in the target feature area after weighting processing is greater than the preset first distance threshold to obtain the calculated wind speed.

9. A wind speed identification device based on spectrum data, characterized in that: include: Memory, used to store programs; A processor, configured to implement the wind speed identification method according to any one of claims 1 to 7 by executing the program stored in the memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a program, and the program can be executed by a processor to implement the wind speed identification method according to any one of claims 1 to 7.

Images