CN115511011B - Radar data correction method and system based on countermeasure generation network model - Google Patents

Abstract

The invention discloses a radar data correction method and system based on a confrontation generation network model, which belongs to the field of radar detection. A method for correcting radar data based on a confrontation generation network model comprises the following steps: inputting radar data to be corrected, to be corrected The radar data is the radial wind velocity field measured by the laser wind radar; the data to be corrected is reconstructed by gray scale conversion; Obscuration mask processing, obtain the masked image and mask; judge whether the degree of occlusion of the mask is within the preset threshold range; input the masked image and mask to the trained confrontation generation network model; conduct radar Data correction process, and output the correction result, the invention can output the corrected radar data in combination with the correction model, and timely and accurately provide various meteorological elements for atmospheric scientific research.

Description

A radar data correction method and system based on an adversarial generative network model

technical field

The invention relates to the field of radar detection, more specifically, to a radar data correction method and system based on a confrontation generation network model.

Background technique

Due to the propagation characteristics of electromagnetic waves and the obstruction of hard targets such as tall buildings near the radar, the beam blocking phenomenon often occurs in the actual detection of radar, resulting in low quality of radar data. Especially for radars deployed in urban areas, the echo data is more susceptible to beam blocking. Even with a slight degree of blocking, the electromagnetic waves emitted cannot fully propagate forward, resulting in weak or completely blocked echoes, making the radar data accuracy is affected. Since the development of radar detection technology, it has become one of the most important remote sensing technologies and is an indispensable existence in the field of atmospheric science. Relying on radar detection technology, we can obtain the temporal and spatial distribution of various meteorological elements. made a great contribution. How to solve the beam blocking phenomenon in the radar detection process is the key problem to improve the radar detection technology.

At present, the correction of radar beam blocking mainly relies on three methods: digital elevation model (Digital ElevationModel, DEM), recognition algorithm based on echo probability characteristics, and beam blocking recognition algorithm based on spatial correlation. The inventors of the present invention have found through research that: the digital elevation model can perform echo correction on areas where beam blocking occurs based on actual surveying and mapping data, but this method has certain limitations, because with social and economic development and urban construction, new buildings The continuous appearance of objects may cause beam blocking, and its surveying and mapping data are difficult to update in real time. Therefore, the beam blocking correction scheme based on the digital elevation model is difficult to guarantee the accuracy of the correction. The beam blocking identification algorithm based on spatial correlation does not require digital elevation data, is not affected by terrain and atmospheric refraction conditions, and can better identify and correct beam blocking. However, when large-scale beam blocking occurs, it cannot use adjacent Because the strong correlation of the large-span space of the radar echo cannot be guaranteed, and when the zero-degree bright band appears, and the zero-degree bright band is used for correction, the correction amount may be too high.

Contents of the invention

1. technical problem to be solved

Aiming at the problems existing in the prior art, the purpose of the present invention is to provide a radar data correction method and system based on the confrontation generation network model, which can output the corrected radar data according to the radar data and in combination with the correction model, so as to contribute to the atmospheric science in time. Research accurately provides various meteorological elements.

2. Technical solutions

In order to solve the above problems, the present invention adopts the following technical solutions.

A radar data correction method based on a confrontation generation network model, comprising the following steps:

Detect masked parts based on radar data;

Train the confrontation generation network model according to the radar data; the confrontation generation network model corrects the radar data;

Select predetermined parameters to classify radar data, and establish multi-radar data correction model;

Obtain real-time radar data;

Match the radar data with the established multi-radar data correction system to obtain the matched correction system; use the correction system to identify and correct the beam blocking of the real-time radar data.

Real-time meteorological elements are retrieved from the corrected radar data and output.

Further, detecting the masked part according to the radar data includes: converting the radar data into a grayscale image, judging whether the value of each pixel is within a preset threshold range, if the value is outside the threshold range, then the point is Masked area, the output is a mask; if the value is within the threshold range, the point is a valid area, and the output is a damaged image.

Further, training the confrontation generation network model according to the radar data includes: selecting a certain amount of radar data to extract masks in batches, judging whether the degree of occlusion of the masks is within a preset threshold range, and selecting masks within the threshold range, According to the different masking degrees of the mask, it is classified as a mask set. A certain amount of radar data is selected, and it is judged whether the shielding degree of the mask is within the preset threshold range, and the radar data within the threshold range are selected as the background set. The mask set and the background set are respectively divided into a training set, a verification set and a test set of the mask and a training set, a verification set and a test set of the background according to a preset ratio. Use the data to train the confrontational generative network model, and the trained model includes the edge reconstruction network and the image restoration network. The edge reconstruction network reconstructs the edge information of the missing area according to the input radar data and mask information, and the image inpainting network reconstructs the image of the missing area under the adjustment of the edge information to obtain the repaired image.

Further, select predetermined parameters to classify the radar data, and establish a multi-radar data correction model, including: training a correction model suitable for different data according to different radar detection data; inputting the radar data to be corrected into the correction system, using The probability algorithm classifies and judges the radar data to be corrected, and selects the appropriate correction model.

Further, obtaining real-time radar data includes: establishing communication with the output end of the radar system, and receiving the radar data output from the output end in real time.

Further, the radar data is matched with the established multi-radar data correction system to obtain the matched correction system; the used correction system is used to identify and correct the beam blocking of the real-time radar data.

3. Beneficial effect

Compared with the prior art, the present invention has the advantage of being able to output corrected radar data according to the radar data and in combination with the correction model, and timely and accurately provide various meteorological elements for atmospheric scientific research.

Description of drawings

Fig. 1 is a flow chart of a method for correcting radar data based on an adversarial generation network model provided by an embodiment of the present invention;

FIG. 2 is a schematic diagram of a method for establishing multi-radar data correction provided by an embodiment of the present invention;

Fig. 3 is the training flowchart of the confrontation generating network model provided by the embodiment of the present invention;

Fig. 4 is a flow chart of the radar data correction system based on the confrontation generation network model provided by the embodiment of the present invention;

Fig. 5 is the radial wind velocity field image to be corrected provided by the embodiment of the present invention;

Fig. 6 is the reconstructed radial wind velocity field image obtained after grayscale conversion provided by the embodiment of the present invention;

FIG. 7 is an image after masking and masking processing provided by an embodiment of the present invention;

Fig. 8 is the corrected radial wind velocity field image provided by the embodiment of the present invention.

Detailed ways

The following will be combined with the accompanying drawings in the embodiments of the present invention; the technical solutions in the embodiments of the present invention will be clearly and completely described; obviously; the described embodiments are only some embodiments of the present invention; rather than all embodiments, based on The embodiments in the present invention; all other embodiments obtained by those skilled in the art without creative work; all belong to the protection scope of the present invention.

In the description of the present invention, it should be noted that the orientations or positional relationships indicated by the terms "upper", "lower", "inner", "outer", "top/bottom" etc. are based on the orientations shown in the drawings Or positional relationship is only for the convenience of describing the present invention and simplifying the description, but does not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only, and should not be understood as indicating or implying relative importance.

In the description of the present invention, it should be noted that, unless otherwise specified and limited, the terms "installed", "set with", "sleeved/connected", "connected", etc. should be understood in a broad sense, such as " Connection" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection or an indirect connection through an intermediary, and it can be an internal connection between two components. connectivity. Those of ordinary skill in the art can understand the specific meanings of the above terms in the present invention in specific situations.

Example 1:

Please refer to Figure 1, a radar data correction method based on the confrontation generative network model, including the following steps:

S1. Input the radar data to be corrected;

Wherein, in this embodiment, the radar data to be corrected refers to the radial wind velocity field measured by the laser wind radar, and the input is the radial wind velocity field to be corrected in 256×256×3 RGB colors.

S2. Reconstruct the data to be corrected by gray scale conversion, and the reconstructed radial wind velocity field image after gray scale conversion is shown in Figure 6;

It specifically includes: converting the radial wind velocity field map into a grayscale image, and then sequentially judging whether the value of each pixel is within the predetermined threshold range, if the value is outside the threshold range, then the point is regarded as having beam blocking and the If the pixel value of the point is set to 255, it is a masked area; if the value is within the threshold range, then there is no beam blocking at this point and the pixel value of this point is set to 0, which is an effective area.

S3. Use the occlusion detection program to perform occlusion judgment on the reconstructed data, and perform occlusion mask processing on the data to be corrected according to the obtained occlusion judgment result, to obtain the image and mask after occlusion, as shown in FIG. 7 ;

In one embodiment, the reconstructed grayscale image in S2 is used to perform occlusion detection. If the pixel value of a certain point is 255, the point is marked as a mask; otherwise, it is marked as an effective area. After the occlusion detection is completed, Generate the mask and output the masked image, respectively.

S4, judging whether the shading degree of the mask is within a preset threshold range;

Use the mask generated in S3 to calculate the degree of shading (the ratio of the size of the mask to the size of the radial wind velocity field to be corrected), where the image of the radial wind velocity field to be corrected is shown in Figure 5, if the shading degree is within the predetermined threshold range If the data to be corrected this time is available, input it to S5; otherwise, this correction will end.

S5. Input the masked image and mask to the trained confrontation generation network model;

In one embodiment, the mask generated in S4 and the masked image are input into the trained confrontation generation network model, and the confrontation generation network model includes an edge reconstruction network and an image restoration network. The edge reconstruction network can reconstruct the edge of the missing part of the image according to the edge information of the input masked image. Then the image inpainting network repairs the image in the missing region under the adjustment of the edge information, so that the inpainting effect is better.

S6. Carry out the radar data correction process, and output the correction result, as shown in FIG. 8 .

In one embodiment, the radial wind velocity field covered by the mask is repaired by using the confrontation generation network model, so as to obtain the reconstructed radial wind velocity field, and complete the correction of the radar data.

The confrontation generation network model is mainly based on the existing information of the input radial wind velocity field to be corrected to restore the missing part of the image. Therefore, after the mask and the masked image are obtained by the confrontation generation network model, the image can be repaired by the confrontation generation network model, so as to obtain the reconstructed radial wind velocity field.

Among them, as shown in Figure 4, a radar data correction system based on the confrontation generative network model includes a grayscale conversion module, an occlusion detection module, a probability algorithm module, a correction model selection module and a radar data correction module.

Among them, as shown in Figure 3, the training process of the confrontation generation network model is as follows:

S1. Use the radar data to extract the corresponding mask to generate a mask set, select relatively complete radar data as the background field and generate a background set;

Wherein, in this embodiment, the radar data refers to the radial wind velocity field measured by the laser wind radar. Use the mask detection module to detect it and extract the corresponding mask to generate a mask set. The shielding detection module screens the radar data according to the preset threshold, and if it is within the threshold range, it is selected as the background field and a background set is generated.

S2. Input the mask set and background field into the confrontation generation network model, and the mask set used by the model performs masking processing on the background field to generate a damaged image and a missing area image. The missing area image is the masked part of the background mask, and then Obtain the inpainted image from the model;

S3. Determine the loss value of the confrontation generation network model according to the background, repaired image, mask and missing region image;

S4. If the loss value is higher than the preset threshold, update the model.

Example 2:

Please refer to Figure 2, a radar data correction method based on the confrontation generative network model, including the following steps:

S1. Input the radar data to be corrected, the radar data to be corrected is the radial wind velocity field measured by the laser wind radar;

S2. Using the maximum probability algorithm to classify and judge the radar data to be corrected, and select a suitable correction model;

S3. Reconstructing the data to be corrected by means of grayscale transformation;

S4. Using the occlusion detection program to perform occlusion judgment on the reconstructed data, and according to the obtained occlusion judgment result, perform occlusion mask processing on the data to be corrected, and obtain the occluded image and mask;

S5, judging whether the shading degree of the mask is within a preset threshold range;

S6. Input the masked image and mask to the trained confrontation generation network model;

S7. Carry out the radar data correction process, and output the correction result.

The above; is only a preferred embodiment of the present invention; but the protection scope of the present invention is not limited thereto; any person familiar with the technical field is within the technical scope disclosed in the present invention; according to the technical solution of the present invention Equivalent replacements or changes thereof and their improved concepts should be covered within the protection scope of the present invention.

Claims (4)

1. A radar data correction method based on the confrontation generation network model, is characterized in that: comprise the following steps: S1. Input the radar data to be corrected, the radar data to be corrected is the radial wind velocity field measured by the laser wind radar, and the radar data to be corrected includes the RGB color input of 256×256×3 to be corrected Radial wind velocity field; S2. Reconstructing the data to be corrected by gray scale conversion; The specific method is: convert the radial wind speed field map into a grayscale image, and then judge whether the value of each pixel point is within the predetermined threshold range, and if the value is outside the threshold range, then the point is regarded as having beam blocking And set the pixel value of this point to 255, which is a masked area; when the value is within the threshold range, then no beam blocking occurs at this point and set the pixel value of this point to 0, which is an effective area; S3. Using the occlusion detection program to perform occlusion judgment on the reconstructed data, and according to the obtained occlusion judgment result, perform occlusion mask processing on the data to be corrected, and obtain the occluded image and mask; The reconstructed grayscale image in S2 is used for occlusion detection. When the pixel value of a point is 255, the point is marked as a mask; otherwise, it is marked as an effective area. After the occlusion detection is completed, the mask and output are generated respectively. masked image; S4. Determine whether the occlusion degree of the mask is within the preset threshold range; use the mask generated in S3 to calculate the occlusion degree; Go to S5; otherwise, end this revision; S5. Input the masked image and the mask to the trained confrontation generation network model, the confrontation generation network model includes an edge reconstruction network and an image repair network, and the edge reconstruction network is based on the edge of the input masked image The information reconstructs the edge of the missing part of the image, and the post-image repair network repairs the image of the missing area under the adjustment of the edge information; S6. Carry out the radar data correction process, and output the correction result. 2. A kind of radar data correction method based on the confrontation generative network model according to claim 1, is characterized in that: the training procedure of described confrontation generative network model comprises the following steps: S1. Use the radar data to extract the corresponding mask to generate a mask set, select relatively complete radar data as the background field and generate a background set; S2. Input the mask set and the background field into the confrontation generation network model, and the model uses the mask set to cover the background field to generate a damaged image and a missing area image, and the missing area image is the mask in the background Covering the part, and then obtaining the repaired image by the model; S3. Determine the loss value of the confrontation generation network model according to the background, repaired image, mask and missing region image; S4. If the loss value is higher than the preset threshold, update the model. 3. A method for correcting radar data based on an adversarial generative network model according to claim 1, characterized in that: before the data to be corrected is reconstructed in a gray scale conversion mode, the radar data to be corrected is classified by a maximum probability algorithm Judgment, select the appropriate revised model. 4. a radar data correction system based on the confrontation generation network model for realizing the radar data correction method based on the confrontation generation network model described in claim 1, is characterized in that: comprise gray-scale conversion module, cover detection module, probability algorithm module, correction model selection module and radar data correction module.

Images