Millimeter wave cloud meter clear sky echo recognition method and system based on deep learning

Abstract

The present invention provides a millimeter wave ceilometer clear sky echo recognition method and system based on deep learning. The method converts the acquired millimeter wave ceilometer historical observation data into a Numpy array format of a preset dimension; cleans and visualizes the data in the Numpy array, and draws labels for the visualized echo reflectivity factor element image; normalizes the data in the Numpy array to obtain a target Numpy array; trains a U‑NET model according to the data and labels in the target Numpy array to obtain a target model; obtains the millimeter wave ceilometer observation data, inputs it into the target model after processing, and outputs the reflectivity echo after quality control; replaces the clear sky echo label in the reflectivity echo after quality control with a no-echo label, and decomposes the annotations in the time column to restore the quality control results of the millimeter wave ceilometer vertical observation every minute.

Description

Technical Field

[0001] The present invention belongs to the technical field of millimeter wave nephelometer clear sky echo recognition, and in particular relates to a millimeter wave nephelometer clear sky echo recognition method and system based on deep learning. Background Art

[0002] Millimeter wave cloud meters have many applications in the field of meteorology. With their unique technical advantages, they play an important role in weather forecasting, climate research, and disastrous weather monitoring. They not only improve the accuracy of meteorological observations, but also provide strong technical support for climate research, environmental monitoring, and aviation safety. With the advancement of technology and the deepening of applications, millimeter wave cloud meters will play an increasingly important role in the field of meteorology, mainly including:

[0003] Observation of cloud structure and characteristics: Millimeter wave cloud meters obtain cloud information by emitting millimeter wave signals and receiving their echoes. This radar system can provide high-resolution cloud structure information, including parameters such as the size, distribution and velocity of cloud droplets and cloud particles. Due to the short wavelength of millimeter waves, millimeter wave cloud radar can provide more detailed observations than traditional radars, which is crucial for understanding the microscopic and macroscopic characteristics of clouds; Improving weather forecast accuracy: Millimeter wave cloud meters can continuously monitor the vertical profile changes of clouds, which is of great significance to make up for the shortcomings of remote sensing methods and improve the accuracy of weather forecasts. It can detect particles with diameters much smaller than the radar wavelength, so that it can describe the physical structure inside the cloud; Monitoring of extreme weather events: Millimeter wave cloud meters can detect and predict the occurrence of extreme meteorological disasters, such as severe thunderstorms and other disastrous weather. It can also effectively detect clouds that have not yet formed precipitation, especially when the cloud droplets in the cloud are relatively small and contain less water, which is more difficult for centimeter-wave radars; Cloud physics and artificial weather modification research: The millimeter-wave cloud meter provides important basic data for cloud physics effect research, artificial weather modification, numerical simulation and other research. It can penetrate the cloud layer and describe the physical structure inside the cloud, which is of great value for scientific research and practical applications; Environmental and climate research: Clouds have an important influence on the transmission and balance process of radiation energy and are a key factor in studying climate. The millimeter-wave cloud meter can provide macroscopic information of clouds and their radiation characteristics, which is crucial to understanding the role of clouds in the climate system; Safe flight and airport operations: The millimeter-wave cloud meter can provide vertical structural information of clouds, fog and sandstorms, which is of great significance for safe flight and operation of airports. It can monitor low-altitude clouds and fog and help airport managers make more accurate decisions.

[0004] However, due to the influence of ground reflection, super-refraction, signal attenuation, Doppler effect, birds, insects and signal interference, the millimeter-wave cloud meter observation data may often contain a lot of clear sky echo data, which sometimes seriously interferes with the identification of normal meteorological echoes. It is necessary to eliminate them to ensure the accuracy and reliability of meteorological observation results. Summary of the invention

[0005] Based on this, an embodiment of the present invention provides a deep learning-based millimeter-wave cloud meter clear sky echo recognition method and system, aiming to provide more accurate observation data for millimeter-wave cloud radar data users.

[0006] A first aspect of an embodiment of the present invention provides a method for identifying clear sky echoes of a millimeter wave nephelometer based on deep learning, the method comprising:

[0007] Acquire historical observation data of a millimeter-wave nephelometer, and convert the historical observation data of the millimeter-wave nephelometer into a Numpy array format of a preset dimension, wherein the historical observation data of the millimeter-wave nephelometer is intercepted in the order of time columns and height rows from low to high, and the observation time of each time column is marked. In addition, the Numpy array contains the speed, echo reflectivity factor, and linear depolarization ratio of each millimeter-wave nephelometer at different times and different heights;

[0008] The data in the Numpy array is cleaned and visualized. At the same time, labels are drawn for the visualized echo reflectivity factor element image, where the horizontal axis of the visualized image is time and the vertical axis is height. The labels include no-echo labels, weather echo labels, and clear sky echo labels.

[0009] Under the condition that the dimension size remains unchanged, the data in the Numpy array is normalized to obtain the target Numpy array;

[0010] According to the data and labels in the target Numpy array, the U-NET model is trained to obtain a target model, wherein in the process of training the U-NET model, the meteorological echo retention rate and the clear sky echo recognition rate are used as evaluation functions;

[0011] Obtaining millimeter wave cloud meter observation data, converting it into a Numpy array format of a preset dimension, and after data cleaning, inputting it into the target model, and outputting a reflectivity echo after quality control;

[0012] The clear sky echo label in the reflectivity echo after quality control is replaced with a no-echo label, and the annotations in the time column are decomposed to restore the quality control results of the millimeter-wave cloud meter's vertical observation every minute.

[0013] Furthermore, in the step of performing data cleaning processing on the data in the Numpy array, all times with observed data are stored in chronological order, and each time is marked; all times without observed data are filled with the first default missing value and spliced ​​after the corresponding data; when the acquisition height is lower than the preset height, the data between the acquisition height and the preset height is filled with the second default missing value; when the acquisition height is higher than the preset height, only the data within the preset height is retained.

[0014] Furthermore, the U-NET model includes an encoding part and a decoding part, the encoding part includes a convolution layer and a pooling layer, the decoding part includes a splicing layer, a convolution layer and a sub-pixel convolution layer, and the reduction ratio of the sub-pixel convolution layer is set to 2.

[0015] Furthermore, the step of drawing labels on the visualized echo reflectivity factor element image includes:

[0016] According to the height where the clear sky echo exists, a basic area with a regular shape to be labeled is determined;

[0017] Acquire a pixel value range belonging to the meteorological echo, determine a first pixel point in the visualized echo reflectivity factor element image according to the pixel value range, and combine adjacent first pixel points to form a first area;

[0018] Acquire a first interval distance between adjacent first areas, and determine whether the first interval distance is less than a preset distance;

[0019] If it is determined that the first interval distance is less than a preset distance, combining adjacent first areas to form a second area;

[0020] Modify the base area according to the boundary between the first area and the second area to obtain a target area;

[0021] The no-echo label, the weather echo label, and the clear-sky echo label are marked based on the target area.

[0022] Furthermore, the step of modifying the base area according to the boundary between the first area and the second area to obtain the target area includes:

[0023] Determine boundaries of the first area, the second area, and the base area respectively, and calculate a second spacing distance between areas between the boundaries according to the boundaries of the first area, the second area, and the base area;

[0024] Determine whether the second interval distance is greater than a threshold;

[0025] If it is determined that the second interval distance is greater than a threshold, dividing the second interval distance according to a preset interval distance to obtain a plurality of divided areas;

[0026] Acquire pixel points having echo reflectivity data in each of the divided areas, and determine a first target pixel point closest to a boundary of the base area;

[0027] Determine, according to the first target pixel, a second target pixel among the first target pixel points;

[0028] The second target pixel points are connected, and a target line is generated in an area formed by a connection line of the second target pixel points, the first area, the second area, and a boundary of the base area, and an area enclosed by the target line is the target area.

[0029] Furthermore, the step of determining the second target pixel point in the first target pixel point according to the first target pixel point includes:

[0030] Step 1: In each of the divided areas, select the first target pixel point on the leftmost or rightmost side as an endpoint;

[0031] Step 2: Generate rays according to the endpoints;

[0032] Step 3: With the vertical direction as the starting position of the ray and the endpoint as the center point, control the ray to rotate clockwise or counterclockwise to determine the first target pixel point that the ray passes through for the first time;

[0033] Step 4: determine whether the first target pixel point that the ray passes through for the first time is the first target pixel point on the leftmost or rightmost side;

[0034] Step 5: if it is determined that the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point, the first target pixel point on the ray is determined as the second target pixel point;

[0035] Step 6: if it is determined that the first target pixel point that the ray passes through for the first time is not the first target pixel point on the leftmost or rightmost side, then the first target pixel point that the ray passes through for the first time is re-determined as the endpoint;

[0036] Step seven, looping through steps two to six until all of the second target pixel points are determined.

[0037] Furthermore, the step of connecting the second target pixel points, generating a target line in an area formed by the connecting line of the second target pixel points, the first area, the second area, and the boundary of the base area, and the area enclosed by the target line as the target area includes:

[0038] Generate a first target line in the middle part between the first area, the second area and the connecting line, extend both ends of the first target line, intersect with the boundaries of the base area respectively, and form the target line;

[0039] A preset number of adjustment points are arranged on the target line, and the closer the distance between the first area, the second area and the connecting line is, the more the adjustment points are arranged.

[0040] A second aspect of an embodiment of the present invention provides a millimeter wave nephelometer clear sky echo recognition system based on deep learning, which is used to implement the millimeter wave nephelometer clear sky echo recognition method based on deep learning according to the first aspect, and the system includes:

[0041] A conversion module, used to obtain historical observation data of the millimeter wave nephelometer, and convert the historical observation data of the millimeter wave nephelometer into a Numpy array format of a preset dimension, wherein the Numpy array contains the velocity, echo reflectivity factor and linear depolarization ratio of each millimeter wave nephelometer at different times and different altitudes;

[0042] The data cleaning module is used to clean the data in the Numpy array and visualize the cleaned data. At the same time, it draws labels for the visualized echo reflectivity factor element image, where the horizontal axis of the visualized image is time and the vertical axis is height. The labels include no-echo labels, weather echo labels, and clear sky echo labels.

[0043] The normalization processing module is used to normalize the data in the Numpy array to obtain the target Numpy array while ensuring that the dimension size remains unchanged;

[0044] A training module, used for training a U-NET model according to the data and labels in the target Numpy array to obtain a target model, wherein in the process of training the U-NET model, a meteorological echo retention rate and a clear sky echo recognition rate are used as evaluation functions;

[0045] An input module is used to obtain millimeter wave cloud meter observation data, convert it into a Numpy array format of a preset dimension, and after data cleaning, input it into the target model and output the reflectivity echo after quality control;

[0046] The replacement module is used to replace the clear sky echo label in the reflectivity echo after quality control with a no-echo label, and decompose the annotations in the time column to restore the quality control results of the millimeter wave cloud meter's vertical observation every minute.

[0047] A third aspect of an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the deep learning-based millimeter wave nephelometer clear sky echo identification method provided in the first aspect.

[0048] A fourth aspect of an embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, wherein when the processor executes the program, the deep learning-based millimeter wave nebulometer clear sky echo identification method provided in the first aspect is implemented.

[0049] A method and system for identifying clear sky echoes from a millimeter wave ceilometer based on deep learning is provided in an embodiment of the present invention. The method obtains historical observation data of a millimeter wave ceilometer and converts the historical observation data of the millimeter wave ceilometer into a Numpy array format of a preset dimension; performs data cleaning on the data in the Numpy array and visualizes the cleaned data, and at the same time, draws labels for the visualized echo reflectivity factor element image; normalizes the data in the Numpy array while ensuring that the dimension size remains unchanged to obtain a target Numpy array; trains a U-NET model based on the data and labels in the target Numpy array to obtain a target model; obtains millimeter wave The ceilometer observation data is converted into the Numpy array format of preset dimensions, and after data cleaning, it is input into the target model to output the reflectivity echo after quality control; the clear sky echo label in the reflectivity echo after quality control is replaced with a no-echo label, and the annotations in the time column are decomposed to restore the quality control results of the millimeter-wave ceilometer's vertical observation every minute. Specifically, relying on artificial intelligence image segmentation technology, the designed target model not only has good clear sky echo quality control capabilities, but also has very good universality. It can be applied to the quality control of other millimeter-wave ceilometer detection results without modification. At the same time, it has good quality control effect, high accuracy, and fast running speed, and can provide users with more timely and accurate millimeter-wave ceilometer real meteorological echoes. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 A flowchart of a method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning provided in Embodiment 1 of the present invention;

[0051] Figure 2 This is the visualized millimeter-wave cloud radar reflectivity echo map;

[0052] Figure 3 This is the visualized multi-factor map of millimeter-wave cloud radar observations;

[0053] Figure 4 A schematic diagram for drawing labels for the visualized Z-element image;

[0054] Figure 5 A schematic diagram for visualizing labels;

[0055] Figure 6 It is a structural diagram of the U-NET model;

[0056] Figure 7 A structural block diagram of a millimeter wave cloud meter clear sky echo recognition system based on deep learning provided in Embodiment 3 of the present invention;

[0057] Figure 8 This is a structural block diagram of an electronic device provided in Embodiment 4 of the present invention. DETAILED DESCRIPTION

[0058] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive.

[0059] It should be noted that when an element is referred to as being "fixed to" another element, it may be directly on the other element or there may be a central element. When an element is considered to be "connected to" another element, it may be directly connected to the other element or there may be a central element at the same time. The terms "vertical", "horizontal", "left", "right" and similar expressions used herein are for illustrative purposes only.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which the present invention belongs. The terms used herein in the specification of the present invention are only for the purpose of describing specific embodiments and are not intended to limit the present invention. The term "and / or" used herein includes any and all combinations of one or more of the related listed items.

[0061] Embodiment 1

[0062] According to an embodiment of the present invention, an embodiment of a method for identifying clear sky echoes of a millimeter wave nebulometer based on deep learning is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0063] In the first embodiment, a method for identifying clear sky echoes of a millimeter wave nephelometer based on deep learning is provided, which can be used in electronic devices, such as computers. Figure 1 , Figure 1 A flowchart of an implementation method of a millimeter wave nephometer clear sky echo recognition method based on deep learning provided in Embodiment 1 of the present invention is shown, which specifically includes steps S01 to S06.

[0064] Step S01, obtaining historical observation data of a millimeter wave cloud meter, and converting the historical observation data of the millimeter wave cloud meter into a Numpy array format of a preset dimension.

[0065] The historical observation data of the millimeter-wave nephelometer are intercepted in the order of time columns and height rows from low to high, and the observation time of each time column is marked. In addition, the Numpy array contains the speed, echo reflectivity factor and linear depolarization ratio of each millimeter-wave nephelometer at different times and heights.

[0066] In this embodiment, in order to facilitate the subsequent training of the U-NET model, the historical observation data of the millimeter wave nephelometer is converted into a two-dimensional Numpy array format of a fixed size. Specifically, most of the millimeter wave nephelometers currently established in various regions have an observation frequency of 1 minute, but a small number of millimeter wave nephelometers have an observation frequency of 5 seconds. In order to be compatible with the observation data of most nephelometers, the observation frequency of the nephelometer is temporarily set to 1 minute.

[0067] Further determine the size of the observation data dimension required for deep learning. Specifically, in terms of time, there are 1440 minutes per day, that is, each nephelometer has 1440 vertical observations per day; in the vertical direction, each nephelometer can detect an altitude of more than 15-18km, and the height difference between every two data points is 30 meters. 512 height rows of data in the vertical direction are intercepted, and the actual observation height is 512×30=15360 meters.

[0068] When there are three millimeter-wave nephelometers working simultaneously in a certain area, the observation data of the three millimeter-wave nephelometers are intercepted. It should be noted that the millimeter-wave nephelometer observation data provides a variety of meteorological elements. In the embodiment of the present invention, the velocity (V, unit: m / s), echo reflectivity factor (Z, unit: dBZ) and linear depolarization ratio (LDR, unit: dB) are used. The observation data of these three elements are intercepted from low to high in the order of 1440 time columns and 512 height rows, and the observation time of each time column is marked. Please refer to Figure 2 , is the visualized millimeter wave cloud radar reflectivity echo map, unit: dBZ, this map reflects the echo reflectivity factor, please refer to Figure 3 , which is a visualized multi-factor map of millimeter-wave cloud radar observation, from top to bottom are V, Z and LDR, and the height is intercepted from 0km to 4km.

[0069] Step S02, cleaning the data in the Numpy array, and visualizing the cleaned data, and at the same time, drawing labels for the visualized echo reflectivity factor element image.

[0070] Specifically, all times with observation data are stored in chronological order, and each time is marked; all times without observation data are filled with the first default missing value and spliced ​​after the corresponding data; when the collection height is lower than the preset height, the data between the collection height and the preset height are filled with the second default missing value; when the collection height is higher than the preset height, only the data within the preset height is retained.

[0071] It can be understood that when the time column is less than 1440 time columns, all time columns with observation data are stored in order, and each time column is also marked. The remaining time columns without observation data are all filled with default missing values ​​and spliced ​​after the data column. When the height row is less than 512 rows, the data between 512 rows and the highest row are filled with default missing values; if the height row exceeds 512 rows, since there is no clear sky echo above 15km, the height rows above 15km can be exempted from quality control and directly spliced ​​with the 512 height row data after quality control. Exemplarily, the millimeter wave cloud meter observation data of a certain day is decoded and spliced ​​into a file of 1440 time columns. Since the height row of the station on that day exceeds 512 rows, the V, Z and LDR data of 512 height rows from low to high are intercepted to become a Numpy array with a dimension size of (1440, 512, 3).

[0072] In addition, Python programming tools are used to visualize the cleaned data, where the horizontal axis of the visualized image is time, the vertical axis is height, and the labels include no-echo labels, weather echo labels, and clear-sky echo labels. Specifically, based on existing manual recognition experience, labeling software is used to draw labels for the visualized Z-element image. The entire image is divided into three categories: Category 1 is the blank area, including default missing values, values ​​without any cloud echoes, etc. Category 2 is weather echoes, including cloud echoes, fog echoes, aerosol echoes, etc., among which cloud echoes also include rain echoes, snow echoes, hail echoes, etc.; the third category is clear-sky echoes, including all other echoes except categories 1 and 2. Use 0, 1, and 2 as no-echo labels, weather echo labels, and clear-sky echo labels, respectively. Please refer to Figure 4 , for a diagram of drawing labels for the visualized Z-element image, see Figure 5 , is a schematic diagram for visualizing labels, where blue represents weather echo, green represents clear sky echo, and white represents no echo.

[0073] Step S03, normalizing the data in the Numpy array while ensuring that the dimension size remains unchanged, to obtain a target Numpy array.

[0074] Specifically, all non-default missing values ​​and unobserved data are recorded as 9, and all other meteorological echoes and non-meteorological echoes are normalized to values ​​between 0 and 1 according to their respective observation elements to form a new Numpy value, and the dimension size is still (1440, 512, 3).

[0075] Step S04, training the U-NET model according to the data and labels in the target Numpy array to obtain a target model, wherein in the process of training the U-NET model, the meteorological echo retention rate and the clear sky echo recognition rate are used as evaluation functions.

[0076] It should be noted that the labels in the target Numpy array and the corresponding V, Z and LDR data are divided into training set, validation set and test set according to the ratio of 8:1:1. Among them, V, Z and LDR are the millimeter wave nephelometer observation data of the same day, which are used as the input of the U-NET model, and the labels are the classification labels of the same day. All training sets, validation sets and test sets are made into TFRecord format.

[0077] In this embodiment, by using the meteorological echo retention rate and the clear sky echo recognition rate as evaluation functions, the trained target model can be made more accurate. If too many echoes are retained, the clear sky echo recognition rate will decrease; if the clear sky echo recognition rate is high, too many echoes will be eliminated, resulting in a low meteorological echo retention rate.

[0078] Specifically, the clear sky echo recognition rate is equal to the number of clear sky echoes correctly identified in the data set divided by the number of clear sky echoes marked in the data set, and the meteorological echo retention rate is equal to the number of meteorological echoes retained in the data set divided by the number of meteorological echoes marked in the data set. For example, there are 300 distance libraries with echo data in the data set, of which 100 distance libraries are marked as clear sky echoes, and 200 distance libraries are marked as meteorological echoes. The distance library identified as clear sky echoes overlaps with the distance library marked as clear sky echoes by 82, and the clear sky echo recognition rate is 82%; the distance library identified as meteorological echoes overlaps with the distance library marked as meteorological echoes by 160, and the meteorological echo retention rate is 80%.

[0079] For more details, see Figure 6, is a schematic diagram of the structure of the U-NET model. The U-NET model includes an encoding part and a decoding part. The encoding part is mainly responsible for feature input, feature extraction and image dimension reduction, including convolution layer and pooling layer; the decoding part is mainly responsible for splicing and upsampling, including splicing layer, convolution layer and subpixel convolution layer. Among them, the purpose of subpixel layer convolution is to increase the resolution of the channel by reducing the number of channels. The reduction ratio r is set to 2, that is, the number of channels is reduced by 4 times, and the number of grid points is also increased by 4 times. In the feature input part, the three elements of V, LDR and Z are fused into a 3-channel data set with a grid number of 1440×512×3. After the data set enters the model, it undergoes 4 down sampling operations to perform deep feature extraction. The number of convolution kernels in the 4 down sampling processes is 64, 128, 256 and 512 respectively, the convolution kernel size is 3×3, and it contains 1 pooling layer with a stride of 2. In the decoding part, CR-Unet performs two convolution operations (the number of convolution kernels is 512) and then performs four upsampling operations to restore the original data size. Upsampling includes 1 splicing layer, 1 sub-pixel convolution layer and 2 convolution layers. The number of convolution kernels in the convolution process is 256, 128, 64 and 32 respectively, and the convolution kernel size is also 3×3. From the beginning to this step, the activation function is all Relu. The last step uses a 1×1 convolution kernel and uses the Softmax activation function to output the classification image. The output values ​​are 0, 1 and 2, representing meteorological echo, clear sky echo and no echo respectively. The epoch of the entire model is set to 300, and the batch size is 16 (Batch-size). The model is compiled and trained based on TensorFlow2.6, and the NVIDIA Tesla V100S 32GB graphics processor is used for computing acceleration.

[0080] Step S05, obtaining millimeter wave cloud meter observation data, converting it into a Numpy array format of a preset dimension, and after data cleaning processing, inputting it into the target model, and outputting a reflectivity echo after quality control.

[0081] Among them, the dimension size of the target model output is (1440, 512).

[0082] Step S06, replace the clear sky echo label in the reflectivity echo after quality control with a no-echo label, and decompose the annotations in the time column to restore the quality control results of the millimeter wave cloud meter's vertical observation every minute.

[0083] Specifically, all red clear sky echoes are changed to white without echoes, so that the entire Numpy array only has meteorological echoes, no echoes or missing data, and the quality control is completed. The Numpy array is re-visualized to obtain a millimeter cloud radar map without clear sky echoes. At the same time, the system can also take echoes of any period for quality control as needed. It only needs to arrange the data time columns in order, and then fill the excess parts into 1440 time columns for quality control calculation and visualization.

[0084] In summary, the method for identifying clear sky echoes of a millimeter-wave ceilometer based on deep learning in the above-mentioned embodiment of the present invention obtains historical observation data of the millimeter-wave ceilometer and converts the historical observation data of the millimeter-wave ceilometer into a Numpy array format of a preset dimension; performs data cleaning on the data in the Numpy array and visualizes the cleaned data, and at the same time, draws labels for the visualized echo reflectivity factor element image; normalizes the data in the Numpy array while ensuring that the dimension size remains unchanged to obtain a target Numpy array; trains a U-NET model according to the data and labels in the target Numpy array to obtain a target model; obtains ... The cloud instrument observation data is converted into the Numpy array format of preset dimensions, and after data cleaning processing, it is input into the target model to output the reflectivity echo after quality control; the clear sky echo label in the reflectivity echo after quality control is replaced with a no-echo label, and the annotations in the time column are decomposed to restore the quality control results of the millimeter-wave cloud meter's vertical observation every minute. Specifically, relying on artificial intelligence image segmentation technology, the designed target model not only has good clear sky echo quality control capabilities, but also has very good universality. It can be applied to the quality control of other millimeter-wave cloud meter detection results without modification. At the same time, it has good quality control effect, high accuracy, and fast running speed, and can provide users with more timely and accurate millimeter-wave cloud meter real meteorological echoes.

[0085] Embodiment 2

[0086] Embodiment 2 of the present invention also provides a method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning. The difference from Embodiment 1 of the present invention is that in the process of drawing labels for the visualized echo reflectivity factor element image, auxiliary drawing of labels is performed. Specifically, according to the height at which the clear sky echo exists, a basic area with a regular shape to be drawn with labels is determined. For example, the height at which the clear sky echo exists is assumed to be 1km~15km, and the basic area can be a rectangular frame within the height range of 1km~15km.

[0087] Obtaining a pixel value range belonging to the meteorological echo, wherein different radar reflectivities correspond to different pixel values, determining a first pixel point in the visualized echo reflectivity factor element image according to the pixel value range, and combining adjacent first pixel points to form a first area;

[0088] Obtaining a first interval distance between adjacent first areas, and determining whether the first interval distance is less than a preset distance;

[0089] If it is determined that the first interval distance is less than the preset distance, the adjacent first areas are combined to form a second area, wherein the process of combining the adjacent first areas is to first determine the nearest line segment of the adjacent first areas, and based on the line segment, generate straight lines parallel to the line segment on both sides thereof, and control the two straight lines to translate in a direction away from the line segment until the area formed between the two straight lines covers any one of the two adjacent first areas, and finally the area enclosed by the two adjacent first areas and the two straight lines is the second area;

[0090] According to the boundary of the first area and the second area, the base area is modified to obtain the target area. Specifically, the boundaries of the first area, the second area and the base area are determined respectively, and according to the boundaries of the first area, the second area and the base area, a second spacing distance of the areas between the boundaries is calculated, wherein the spacing distance between the first areas, between the second areas or between the first area and the second area is mainly calculated, and the spacing distance refers to the shortest distance between two areas;

[0091] Determine whether the second interval distance is greater than a threshold;

[0092] If it is determined that the second interval distance is greater than the threshold, the second interval distance is divided according to the preset interval distance to obtain a plurality of divided areas. It can be understood that, taking a rectangular area as an example, the length of the rectangular area is 4 cm, the preset interval distance is 1 cm, and the rectangular area is divided into four divided areas by three vertical lines;

[0093] Obtaining pixel points with echo reflectivity data in each divided area, and determining a first target pixel point closest to a boundary of the base area;

[0094] According to the first target pixel point, a second target pixel point in the first target pixel point is determined to adjust the boundary contour according to the distribution trend of the second target pixel point. Specifically, in step 1, in each divided area, the leftmost or rightmost first target pixel point is selected as an endpoint;

[0095] Step 2: Generate a ray based on the endpoint, that is, take the endpoint as the starting point and send out a ray z;

[0096] Step 3: With the vertical direction as the starting position of the ray and the end point as the center point, control the ray to rotate clockwise or counterclockwise to determine the first target pixel point that the ray passes through for the first time;

[0097] Step 4: determine whether the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point;

[0098] Step 5: if it is determined that the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point, the first target pixel point on the ray is determined as the second target pixel point;

[0099] Step 6: if it is determined that the first target pixel point that the ray first passes through is not the first target pixel point on the leftmost side or the first target pixel point on the rightmost side, then the first target pixel point that the ray first passes through is re-determined as the endpoint;

[0100] Step 7, looping through Step 2 to Step 6 until all second target pixels are determined;

[0101] Connect the second target pixel points, generate a target line in the area formed by the connecting line of the second target pixel points, the first area, the second area and the boundary of the base area, and the area enclosed by the target line is the target area. It should be noted that the first target line is generated in the middle part between the first area, the second area and the connecting line, and the two ends of the first target line are extended to intersect with the boundary of the base area respectively to form the target line;

[0102] A preset number of adjustment points are set on the target line, and the closer the distance between the first area, the second area and the connecting line is, the more adjustment points are set;

[0103] Based on the target area, no-echo labels, weather echo labels and clear sky echo labels are marked, and the above method is used to assist users in quick marking.

[0104] Embodiment 3

[0105] See also Figure 7 , Figure 7 It is a structural block diagram of a millimeter wave nephelometer clear sky echo identification system based on deep learning provided in Example 3 of the present invention. The millimeter wave nephelometer clear sky echo identification system 200 based on deep learning is used to implement the above-mentioned embodiments and preferred implementation modes, and the descriptions that have been made will not be repeated. As used below, the term "module" can implement a combination of software and / or hardware for a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, the implementation in hardware, or a combination of software and hardware, is also possible and conceivable.

[0106] Specifically, the millimeter wave nephometer clear sky echo recognition system 200 based on deep learning includes: a conversion module 21, a data cleaning module 22, a normalization processing module 23, a training module 24, an input module 25 and a replacement module 26, wherein:

[0107] A conversion module 21 is used to obtain historical observation data of the millimeter wave nephelometer, and convert the historical observation data of the millimeter wave nephelometer into a Numpy array format of a preset dimension, wherein the Numpy array contains the velocity, echo reflectivity factor and linear depolarization ratio of each millimeter wave nephelometer at different times and different heights;

[0108] The data cleaning module 22 is used to perform data cleaning on the data in the Numpy array and visualize the cleaned data. At the same time, a label is drawn for the visualized echo reflectivity factor element image, wherein the horizontal axis of the visualized image is time, the vertical axis is height, and the label includes a no-echo label, a weather echo label, and a clear sky echo label, wherein all times with observation data are stored in chronological order, and each time is marked; all times without observation data are filled with the first default missing value and spliced ​​after the corresponding data; when the acquisition height is lower than the preset height, the data between the acquisition height and the preset height is filled with the second default missing value; when the acquisition height is higher than the preset height, only the data within the preset height is retained;

[0109] A normalization processing module 23 is used to normalize the data in the Numpy array while ensuring that the dimension size remains unchanged to obtain a target Numpy array;

[0110] A training module 24 is used to train a U-NET model according to the data and labels in the target Numpy array to obtain a target model, wherein in the process of training the U-NET model, the meteorological echo retention rate and the clear sky echo recognition rate are used as evaluation functions, the U-NET model includes an encoding part and a decoding part, the encoding part includes a convolution layer and a pooling layer, the decoding part includes a splicing layer, a convolution layer and a sub-pixel convolution layer, and the reduction ratio of the sub-pixel convolution layer is set to 2;

[0111] An input module 25 is used to obtain millimeter wave cloud meter observation data, convert it into a Numpy array format of a preset dimension, and after data cleaning, input it into the target model to output a reflectivity echo after quality control;

[0112] The replacement module 26 is used to replace the clear sky echo label in the reflectivity echo after quality control with a no-echo label, and decompose the annotations in the time column to restore the quality control results of the millimeter wave cloud meter's vertical observation every minute.

[0113] Further, in some optional embodiments of the present invention, the data cleaning module 22 includes:

[0114] A first determining unit is used to determine a basic area with a regular shape to be labeled according to a height where a clear sky echo exists;

[0115] A second determination unit is used to obtain a pixel value range belonging to the meteorological echo, determine a first pixel point in the visualized echo reflectivity factor element image according to the pixel value range, and combine adjacent first pixel points to form a first area;

[0116] A first judging unit, configured to obtain a first interval distance between adjacent first areas, and judge whether the first interval distance is less than a preset distance;

[0117] a combining unit, configured to combine adjacent first areas to form a second area if it is determined that the first interval distance is less than a preset distance;

[0118] a modifying unit, configured to modify the base area according to a boundary between the first area and the second area to obtain a target area;

[0119] The marking unit is used to mark the no-echo label, the weather echo label and the clear sky echo label based on the target area.

[0120] Further, in some optional embodiments of the present invention, the modification unit includes:

[0121] a calculation subunit, configured to respectively determine boundaries of the first area, the second area, and the base area, and calculate a second spacing distance between areas between the boundaries according to the boundaries of the first area, the second area, and the base area;

[0122] A judging subunit, configured to judge whether the second interval distance is greater than a threshold;

[0123] a dividing subunit, configured to divide the second interval distance into a plurality of divided areas according to a preset interval distance if it is determined that the second interval distance is greater than a threshold;

[0124] A first determining subunit is used to obtain pixel points having echo reflectivity data in each of the divided areas, and determine a first target pixel point closest to a boundary of the base area;

[0125] A second determining subunit is configured to determine a second target pixel point among the first target pixel points according to the first target pixel point. Specifically, in step 1, in each of the divided areas, the leftmost or rightmost first target pixel point is selected as an endpoint;

[0126] Step 2: Generate rays according to the endpoints;

[0127] Step 3: With the vertical direction as the starting position of the ray and the endpoint as the center point, control the ray to rotate clockwise or counterclockwise to determine the first target pixel point that the ray passes through for the first time;

[0128] Step 4: determine whether the first target pixel point that the ray passes through for the first time is the first target pixel point on the leftmost or rightmost side;

[0129] Step 5: if it is determined that the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point, the first target pixel point on the ray is determined as the second target pixel point;

[0130] Step 6: if it is determined that the first target pixel point that the ray passes through for the first time is not the first target pixel point on the leftmost or rightmost side, then the first target pixel point that the ray passes through for the first time is re-determined as the endpoint;

[0131] Step 7, looping through Step 2 to Step 6 until all of the second target pixel points are determined;

[0132] a connecting subunit, configured to connect the second target pixel points, generate a target line in an area formed by a connecting line of the second target pixel points, the first area, the second area, and a boundary of the base area, wherein an area enclosed by the target line is the target area, and specifically, generate a first target line in a middle portion between the first area, the second area, and the connecting line, extend both ends of the first target line, and respectively intersect with the boundary of the base area to form the target line;

[0133] A preset number of adjustment points are arranged on the target line, and the closer the distance between the first area, the second area and the connecting line is, the more the adjustment points are arranged.

[0134] Embodiment 4

[0135] Another aspect of the present invention provides an electronic device, see Figure 8 , shown is an electronic device in Embodiment 4 of the present invention, including a memory 20, a processor 10, and a computer program 30 stored in the memory and executable on the processor. When the processor 10 executes the computer program 30, the method for identifying clear sky echoes of a millimeter-wave nephelometer based on deep learning as described above is implemented.

[0136] In some embodiments, the processor 10 may be a central processing unit (CPU), a controller, a microcontroller, a microprocessor or other data processing chip, used to run program codes or process data stored in the memory 20, such as executing access restriction programs.

[0137] Among them, the memory 20 includes at least one type of readable storage medium, and the readable storage medium includes a flash memory, a hard disk, a multimedia card, a card-type memory (for example, an SD or DX memory, etc.), a magnetic memory, a magnetic disk, an optical disk, etc. In some embodiments, the memory 20 can be an internal storage unit of an electronic device, such as a hard disk of the electronic device. In other embodiments, the memory 20 can also be an external storage device of an electronic device, such as a plug-in hard disk equipped on the electronic device, a smart memory card (Smart Media Card, SMC), a secure digital (Secure Digital, SD) card, a flash card (FlashCard), etc. Further, the memory 20 can also include both an internal storage unit of the electronic device and an external storage device. The memory 20 can not only be used to store application software and various types of data of the electronic device, but also can be used to temporarily store data that has been output or is to be output.

[0138] It should be pointed out that Figure 8 The structure shown does not constitute a limitation on the electronic device. In other embodiments, the electronic device may include fewer or more components than those shown in the figure, or combine certain components, or arrange the components differently.

[0139] An embodiment of the present invention further proposes a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned deep learning-based millimeter-wave nephelometer clear sky echo identification method.

[0140] Those skilled in the art will appreciate that the logic and / or steps represented in the flowchart or otherwise described herein, for example, may be considered as an ordered list of executable instructions for implementing logical functions, and may be specifically implemented in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For purposes of this specification, "computer-readable medium" may be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0141] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0142] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or a combination thereof: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0143] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0144] The above embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present invention. It should be pointed out that, for a person of ordinary skill in the art, several modifications and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the present invention patent shall be subject to the attached claims.

Claims

1. A method for identifying clear sky echoes from a millimeter wave nephelometer based on deep learning, characterized in that: The method comprises: Acquire historical observation data of a millimeter-wave nephelometer, and convert the historical observation data of the millimeter-wave nephelometer into a Numpy array format of a preset dimension, wherein the historical observation data of the millimeter-wave nephelometer is intercepted in the order of time columns and height rows from low to high, and the observation time of each time column is marked. In addition, the Numpy array contains the speed, echo reflectivity factor, and linear depolarization ratio of each millimeter-wave nephelometer at different times and different heights; The data in the Numpy array is cleaned and visualized. At the same time, labels are drawn for the visualized echo reflectivity factor element image, where the horizontal axis of the visualized image is time and the vertical axis is height. The labels include no-echo labels, weather echo labels, and clear sky echo labels. Under the condition that the dimension size remains unchanged, the data in the Numpy array is normalized to obtain the target Numpy array; According to the data and labels in the target Numpy array, the U-NET model is trained to obtain a target model, wherein in the process of training the U-NET model, the meteorological echo retention rate and the clear sky echo recognition rate are used as evaluation functions; Obtaining millimeter wave cloud meter observation data, converting it into a Numpy array format of a preset dimension, and after data cleaning, inputting it into the target model, and outputting a reflectivity echo after quality control; The clear sky echo label in the reflectivity echo after quality control is replaced with a no-echo label, and the annotations in the time column are decomposed to restore the quality control results of the millimeter-wave cloud meter's vertical observation every minute.

2. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 1, characterized in that: In the step of performing data cleaning processing on the data in the Numpy array, all times with observed data are stored in chronological order, and each time is marked; all times without observed data are filled with the first default missing value and spliced ​​after the corresponding data; when the acquisition height is lower than the preset height, the data between the acquisition height and the preset height is filled with the second default missing value; when the acquisition height is higher than the preset height, only the data within the preset height is retained.

3. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 2, characterized in that: The U-NET model includes an encoding part and a decoding part, the encoding part includes a convolution layer and a pooling layer, the decoding part includes a splicing layer, a convolution layer and a sub-pixel convolution layer, and the reduction ratio of the sub-pixel convolution layer is set to 2.

4. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 3, characterized in that: The step of drawing labels on the visualized echo reflectivity factor element image comprises: According to the height where the clear sky echo exists, a basic area with a regular shape to be labeled is determined; Acquire a pixel value range belonging to the meteorological echo, determine a first pixel point in the visualized echo reflectivity factor element image according to the pixel value range, and combine adjacent first pixel points to form a first area; Acquire a first interval distance between adjacent first areas, and determine whether the first interval distance is less than a preset distance; If it is determined that the first interval distance is less than a preset distance, combining adjacent first areas to form a second area; Modify the base area according to the boundary between the first area and the second area to obtain a target area; The no-echo label, the weather echo label, and the clear-sky echo label are marked based on the target area.

5. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 4, characterized in that: The step of modifying the base area according to the boundary between the first area and the second area to obtain the target area comprises: Determine boundaries of the first area, the second area, and the base area respectively, and calculate a second spacing distance between areas between the boundaries according to the boundaries of the first area, the second area, and the base area; Determine whether the second interval distance is greater than a threshold; If it is determined that the second interval distance is greater than a threshold, dividing the second interval distance according to a preset interval distance to obtain a plurality of divided areas; Acquire pixel points having echo reflectivity data in each of the divided areas, and determine a first target pixel point closest to a boundary of the base area; Determine, according to the first target pixel, a second target pixel among the first target pixel points; The second target pixel points are connected, and a target line is generated in an area formed by a connection line of the second target pixel points, the first area, the second area, and a boundary of the base area, and an area enclosed by the target line is the target area.

6. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 5, characterized in that: The step of determining the second target pixel point in the first target pixel point according to the first target pixel point comprises: Step 1: In each of the divided areas, select the first target pixel point on the leftmost or rightmost side as an endpoint; Step 2: Generate rays according to the endpoints; Step 3: With the vertical direction as the starting position of the ray and the endpoint as the center point, control the ray to rotate clockwise or counterclockwise to determine the first target pixel point that the ray passes through for the first time; Step 4: determine whether the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point; Step 5: if it is determined that the first target pixel point that the ray passes through for the first time is the leftmost or rightmost first target pixel point, the first target pixel point on the ray is determined as the second target pixel point; Step 6: if it is determined that the first target pixel point that the ray passes through for the first time is not the first target pixel point on the leftmost or rightmost side, then the first target pixel point that the ray passes through for the first time is re-determined as the endpoint; Step seven, looping through steps two to six until all of the second target pixel points are determined.

7. The method for identifying clear sky echoes of a millimeter wave cloud meter based on deep learning according to claim 6, characterized in that: The step of connecting the second target pixel points, generating a target line in an area formed by a connecting line of the second target pixel points, the first area, the second area, and a boundary of the base area, and wherein an area enclosed by the target line is the target area comprises: Generate a first target line in the middle part between the first area, the second area and the connecting line, extend both ends of the first target line, intersect with the boundaries of the base area respectively, and form the target line; A preset number of adjustment points are arranged on the target line, and the closer the distance between the first area, the second area and the connecting line is, the more the adjustment points are arranged.

8. A millimeter wave cloud meter clear sky echo recognition system based on deep learning, characterized in that: The system is used to implement the method for identifying clear sky echoes of a millimeter wave nephelometer based on deep learning as described in any one of claims 1 to 7, the system comprising: A conversion module, used to obtain historical observation data of the millimeter wave nephelometer, and convert the historical observation data of the millimeter wave nephelometer into a Numpy array format of a preset dimension, wherein the Numpy array contains the velocity, echo reflectivity factor and linear depolarization ratio of each millimeter wave nephelometer at different times and different altitudes; The data cleaning module is used to clean the data in the Numpy array and visualize the cleaned data. At the same time, it draws labels for the visualized echo reflectivity factor element image, where the horizontal axis of the visualized image is time and the vertical axis is height. The labels include no-echo labels, weather echo labels, and clear sky echo labels. The normalization processing module is used to normalize the data in the Numpy array to obtain the target Numpy array while ensuring that the dimension size remains unchanged; A training module, used for training a U-NET model according to the data and labels in the target Numpy array to obtain a target model, wherein in the process of training the U-NET model, a meteorological echo retention rate and a clear sky echo recognition rate are used as evaluation functions; An input module is used to obtain millimeter wave cloud meter observation data, convert it into a Numpy array format of a preset dimension, and after data cleaning, input it into the target model and output the reflectivity echo after quality control; The replacement module is used to replace the clear sky echo label in the reflectivity echo after quality control with a no-echo label, and decompose the annotations in the time column to restore the quality control results of the millimeter wave cloud meter's vertical observation every minute.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the millimeter wave cloud meter clear sky echo recognition method based on deep learning as described in any one of claims 1 to 7.

10. An electronic device, characterized in that: It includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the program, it implements the deep learning-based millimeter wave cloud meter clear sky echo recognition method as described in any one of claims 1 to 7.

Images