A method for evaluating super-resolution images of meteorological elements
Through super-resolution reconstruction and convolutional neural network evaluation technology, the problems of limited resolution and inconsistent evaluation standards are solved, the accuracy and uniformity of meteorological image quality evaluation are improved, and the high meteorological analysis and forecasting are ensured. Quality input data.
Patent Information
- Application Number
- CN202411239331.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-05
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-09-05
AI Technical Summary
Due to limited resolution and inconsistent evaluation standards, traditional meteorological images have missed meteorological information, which affects the accuracy of analysis and forecasting, and the image quality is difficult to compare and optimize.
Super-resolution reconstruction technology is used to generate high-resolution images from images from multiple meteorological channels, and quality evaluation is carried out through convolutional neural networks, including image clarity, structural similarity and meteorological element accuracy assessment, to generate practical calculation points, Ensure data accuracy.
It improves the accuracy and uniformity of meteorological images quality assessment, ensures that high-quality input data is used for meteorological analysis and forecasting, improves the quality of input data and public trust of the forecast model, and optimizes resource allocation and disaster response capabilities.
Smart Images

Figure CN118761909B_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of meteorological image processing and relates to a method for evaluating meteorological element super-resolution images. Background Art
[0002] With the development of meteorological forecasting technology, the requirements for the resolution of meteorological images are getting higher and higher. Traditional meteorological images are limited by the acquisition equipment, the diversity of meteorological observation equipment, different measurement environments, lack of stability of instruments, and data acquisition and transmission are restricted by multiple factors such as communication conditions and observation environment. At the same time, most weather image observation equipment is deployed on mobile terminals or edge devices with weak computing power and low memory, such as mobile phones, small automatic observation stations, perception cameras, etc., resulting in a series of low-quality observation images and the inability to directly obtain high-resolution images.
[0003] The defects and drawbacks of existing technologies for meteorological image quality assessment are mainly reflected in the following aspects:
[0004] Existing meteorological images usually come from remote sensing technologies such as satellites and radars. Although these technologies can cover a wide geographical area, their resolution is limited. This resolution limitation may cause important meteorological information to be missed, affecting the accuracy of meteorological analysis and forecasts.
[0005] There is no unified evaluation standard for meteorological image quality, and different institutions and studies may use different evaluation methods and standards. This lack of uniformity not only makes it difficult to compare the results, but also hinders the standardization and optimization of evaluation methods. Summary of the invention
[0006] In view of the above problems existing in the prior art, the present invention provides a meteorological element super-resolution image evaluation method for solving the above technical problems.
[0007] In order to achieve the above purpose and other purposes, the technical solution adopted by the present invention is as follows:
[0008] The present invention provides a method for evaluating a super-resolution image of meteorological elements, the method comprising the following steps:
[0009] Step 1: The image evaluation device receives meteorological images corresponding to the same observation time point from various meteorological channels;
[0010] Step 2: applying super-resolution reconstruction technology to the meteorological images corresponding to the same observation time point of each meteorological channel, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel;
[0011] Step 3: Conduct quality assessment on high-resolution meteorological images from each meteorological channel at the same observation time point;
[0012] Step 4: Based on the quality assessment results of the high-resolution meteorological images of each meteorological channel corresponding to the same observation time point, the practicality of the meteorological images of each meteorological channel corresponding to the same observation time point is evaluated to obtain the calculated practicality score of the meteorological images of each meteorological channel corresponding to the same observation time point;
[0013] Step 5: Use the meteorological image corresponding to the meteorological channel ranked first in terms of the calculation utility score for meteorological analysis and forecasting at the observation time point.
[0014] In a possible design, super-resolution reconstruction technology is applied to the meteorological images corresponding to the same observation time point in each meteorological channel. The specific process is as follows:
[0015] Extract meteorological image pairs corresponding to the same historical observation time point from the meteorological database, and abbreviate them as meteorological image pairs, wherein each meteorological image pair contains a set of low-resolution meteorological images and a set of high-resolution meteorological images;
[0016] And perform standardized scaling processing on each meteorological image pair so that the pixel value of each meteorological image in each meteorological image pair is maintained in the range of [0,1];
[0017] Extract various types of convolutional neural network models from the meteorological database, input low-resolution meteorological images within each meteorological image pair into each type of convolutional neural network model, thereby outputting high-resolution meteorological images corresponding to the outputs of each type of convolutional neural network model for the low-resolution meteorological images within each meteorological image pair;
[0018] And obtain the RGB value of each pixel in each color channel in the output high-resolution meteorological image corresponding to each type of convolutional neural network model in each meteorological image pair , and simultaneously obtain the RGB value of each pixel in each color channel in each high-resolution meteorological image within each meteorological image pair , where i is the number of each meteorological image pair, i=1,2,..n, f is the number of each type of convolutional neural network model, f=1,2,..z, z is the total number of types corresponding to the convolutional neural network model, p is the number of each pixel, p=1,2,..q, m is the number of each color channel, m=m1,m2,m3, m1,m2 and m3 refer to the red channel, green channel and blue channel respectively;
[0019] Further calculations were performed to obtain the mean square error of each type of convolutional neural network model corresponding to the meteorological image pairs. , n and q represent the total number of meteorological image pairs and the total number of pixels, respectively.
[0020] In a possible design, high-resolution meteorological images corresponding to the same observation time point are generated for each meteorological channel. The specific process is as follows:
[0021] The output high-resolution meteorological images corresponding to the low-resolution meteorological images in each meteorological image pair and each type of convolutional neural network model are simply recorded as the reconstructed images corresponding to each type of convolutional neural network model of each meteorological image pair, and the high-resolution meteorological images in each meteorological image pair are recorded as the original images of each meteorological image pair;
[0022] And according to the image editing software, the bit depth of the reconstructed image of each meteorological image pair corresponding to each type of convolutional neural network model and the original image of each meteorological image pair are checked respectively, where the bit depth is divided into 8 bits and 16 bits;
[0023] The meteorological image pairs whose bit depth of the reconstructed images of each type of convolutional neural network model is consistent with the bit depth of the original image are selected as sample meteorological image pairs, and the bit depth of each sample meteorological image pair is obtained. ,c is the number of each sample meteorological image pair;
[0024] The possible maximum pixel value of the original image in each sample meteorological image pair is thus determined ;
[0025] The similarity between the original image in each sample meteorological image pair and the reconstructed image of each sample meteorological image pair corresponding to each type of convolutional neural network model is obtained by comprehensive calculation. , and arrange them in descending order, and select the convolutional neural network model type with the highest similarity of the corresponding bit depth of each sample meteorological image pair as the reference convolutional neural network model type of the meteorological image with the same bit depth as each sample meteorological image pair;
[0026] According to the image editing software, the meteorological images corresponding to the same observation time point of each meteorological channel are checked respectively, and the bit depth of the meteorological images corresponding to the same observation time point of each meteorological channel is obtained, which is input into the reference convolutional neural network model type of meteorological images with consistent bit depth of each sample meteorological image, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel.
[0027] In one possible design, the quality assessment includes image clarity assessment, structural similarity assessment, and meteorological element accuracy assessment.
[0028] In a possible design, the image clarity of high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is evaluated. The specific process is as follows:
[0029] Import the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel into the rectangular coordinate system to obtain the position coordinates of each pixel point of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel , k is the number of each meteorological channel, k=1,2,...b, b is the total number of meteorological channels;
[0030] Calculate the contrast of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , is the pixel value of the position coordinate of the pth pixel point in the high-resolution meteorological image of the kth meteorological channel corresponding to the same observation time point;
[0031] Extract the set two-dimensional Laplacian operator convolution kernel from the meteorological database , and then calculate the Laplace value of each pixel of the high-resolution meteorological image corresponding to the same observation time point in each meteorological channel ;
[0032] Further, the non-reference image clarity evaluation value of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is obtained. ;
[0033] Comprehensive analysis obtains the image clarity assessment index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , where a1 and a2 represent the weighting factors of the predefined contrast and the definition evaluation value of the reference-free image, respectively, and a1+a2=1, They represent the predefined contrast reference difference and the non-reference image sharpness assessment difference, respectively. They are the set contrast threshold and the no-reference image clarity evaluation threshold respectively.
[0034] In a possible design, the structural similarity of high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is evaluated. The specific process is as follows:
[0035] A meteorological channel is randomly selected from each meteorological channel as a target meteorological channel, thereby screening out a high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point and a high-resolution meteorological image of each other meteorological channel corresponding to the same observation time point, and the high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point is simply recorded as the meteorological image corresponding to the target meteorological channel, and similarly, the high-resolution meteorological images of each other meteorological channel corresponding to the same observation time point are simply recorded as the meteorological images corresponding to each other meteorological channel;
[0036] Synchronously obtain the brightness value of each pixel in the meteorological image corresponding to the target meteorological channel And the brightness value of each pixel in the meteorological image corresponding to other meteorological channels , r is the number of each other meteorological channel, r=1,2,...w, w represents the total number of other meteorological channels, and further calculates the average brightness of the meteorological image corresponding to the target meteorological channel , brightness variance And the average brightness of the corresponding meteorological images of other meteorological channels , brightness variance ;
[0037] Evaluate the structural similarity evaluation index of the target meteorological channel corresponding to the meteorological image ,in are predefined constants, is the brightness covariance of the meteorological images corresponding to each other meteorological channel and the meteorological image corresponding to the target meteorological channel. The specific formula is as follows:
[0038] ;
[0039] In the same way, the structural similarity evaluation index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point is obtained. .
[0040] In a possible design, the accuracy of meteorological elements is evaluated for high-resolution meteorological images of each meteorological channel at the same observation time point. The specific process is as follows:
[0041] Extracting a first estimated value of each meteorological element from the meteorological image corresponding to the target meteorological channel, and extracting other estimated values of each meteorological element from the meteorological images corresponding to other meteorological channels;
[0042] Then, the mean absolute error of each meteorological element of the target meteorological channel corresponding to the meteorological image and the meteorological image corresponding to other meteorological channels is obtained by calculation. , Root Mean Square Error , correlation coefficient and the coefficient of determination , y is the number of each meteorological element, y=1,2,... , is the total number of meteorological elements;
[0043] and will and Perform maximum normalization processing to obtain the average absolute error of each meteorological element between the meteorological image corresponding to the target meteorological channel and the meteorological images corresponding to other meteorological channels. and RMS error ;
[0044] Evaluate the accuracy of meteorological elements in the target meteorological channel corresponding to the meteorological image ,and are the corresponding indicator weights, and ;
[0045] According to the calculation method of the meteorological element accuracy evaluation index of the meteorological image corresponding to the target meteorological channel, the meteorological element accuracy evaluation index of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is calculated in the same way. .
[0046] In a possible design, the practicality of meteorological images corresponding to the same observation time point of each meteorological channel is evaluated. The specific evaluation process is as follows:
[0047] The image clarity evaluation index of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is compared with the image clarity evaluation index interval corresponding to each image clarity score value stored in the meteorological database, so as to obtain the image clarity score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel, and the structural similarity score value and meteorological element accuracy score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel are obtained in the same way according to the acquisition method thereof;
[0048] The image clarity score, structural similarity score and meteorological element accuracy score of the high-resolution meteorological images of each meteorological channel at the same observation time point are added together, and the result of the addition is divided by 3 to obtain the computational utility score of the meteorological images of each meteorological channel at the same observation time point.
[0049] Another aspect of the present invention provides a meteorological element super-resolution image evaluation device, including a processor, a memory and a communication bus;
[0050] The memory stores a computer-readable program executable by the processor;
[0051] The communication bus realizes the connection and communication between the processor and the memory;
[0052] When the processor executes the computer-readable program, the processor implements the steps of a meteorological element super-resolution image assessment method as described in any one of the present inventions.
[0053] On the other hand, the present invention provides a computer-readable storage medium, characterized in that: the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in a meteorological element super-resolution image assessment method as described in any one of the present invention.
[0054] As described above, the meteorological element super-resolution image evaluation method provided by the present invention has at least the following beneficial effects:
[0055] The present invention provides a meteorological element super-resolution image evaluation method, which is used for receiving meteorological images corresponding to the same observation time point from various meteorological channels by an image evaluation device, and applying super-resolution reconstruction technology to the meteorological images corresponding to the same observation time point of various meteorological channels, generating high-resolution meteorological images corresponding to the same observation time point of various meteorological channels, and then performing quality evaluation on the high-resolution meteorological images corresponding to the same observation time point of various meteorological channels, and based on the quality evaluation results of the high-resolution meteorological images corresponding to the same observation time point of various meteorological channels, performing practicality evaluation on the meteorological images corresponding to the same observation time point of various meteorological channels, and obtaining the computational practicality score of the meteorological images corresponding to the same observation time point of various meteorological channels. On the one hand, high-quality meteorological images are the basis for effective meteorological analysis and accurate forecasting. Through strict evaluation and control of image quality, the accuracy of the data used can be ensured at the visual and technical levels, thereby improving the input data quality of the weather forecast model. On the other hand, meteorological image quality evaluation is not only a technical requirement for improving the accuracy of meteorological forecasts, but also an important means for optimizing resource allocation, enhancing disaster response capabilities, improving public trust and satisfaction, promoting scientific research and technological development, and complying with laws and regulations and standards. Therefore, continuous improvement of meteorological image quality evaluation is crucial to the development of modern meteorological services. BRIEF DESCRIPTION OF THE DRAWINGS
[0056] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.
[0057] Figure 1 It is a schematic diagram of the connection of each step of the method of the present invention.
[0058] Figure 2 It is a schematic diagram of the structure of the device shown in the present invention. DETAILED DESCRIPTION
[0059] The above contents in combination with the implementation of the present invention are merely examples and explanations of the concept of the present invention. The technical personnel in the relevant technical field may make various modifications or supplements to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they shall all fall within the protection scope of the present invention.
[0060] Example 1
[0061] See also Figure 1 As shown, a method for evaluating meteorological element super-resolution images comprises the following steps:
[0062] Step 1: The image evaluation device receives meteorological images corresponding to the same observation time point from various meteorological channels;
[0063] It should be added that meteorological channels include but are not limited to satellites, radar systems, ground observation stations, weather balloons and aerial observations.
[0064] Step 2: applying super-resolution reconstruction technology to the meteorological images corresponding to the same observation time point of each meteorological channel, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel;
[0065] On the basis of the above embodiment, super-resolution reconstruction technology is applied to the meteorological images corresponding to the same observation time point of each meteorological channel. The specific process is as follows:
[0066] Extract meteorological image pairs corresponding to the same historical observation time point from the meteorological database, and abbreviate them as meteorological image pairs, wherein each meteorological image pair contains a set of low-resolution meteorological images and a set of high-resolution meteorological images;
[0067] And perform standardized scaling processing on each meteorological image pair so that the pixel value of each meteorological image in each meteorological image pair is maintained in the range of [0,1];
[0068] Extract various types of convolutional neural network models from the meteorological database, input low-resolution meteorological images within each meteorological image pair into each type of convolutional neural network model, thereby outputting high-resolution meteorological images corresponding to the outputs of each type of convolutional neural network model for the low-resolution meteorological images within each meteorological image pair;
[0069] And obtain the RGB value of each pixel in each color channel in the output high-resolution meteorological image corresponding to each type of convolutional neural network model in each meteorological image pair , and simultaneously obtain the RGB value of each pixel in each color channel in each high-resolution meteorological image within each meteorological image pair , where i is the number of each meteorological image pair, i=1,2,..n, f is the number of each type of convolutional neural network model, f=1,2,..z, z is the total number of types corresponding to the convolutional neural network model, p is the number of each pixel, p=1,2,..q, m is the number of each color channel, m=m1,m2,m3, m1,m2 and m3 refer to the red channel, green channel and blue channel respectively;
[0070] It should be added that the various convolutional neural network models include but are not limited to SRCNN architecture, ESPCN architecture, SRGAN architecture and EDSR architecture;
[0071] Further calculations were performed to obtain the mean square error of each type of convolutional neural network model corresponding to the meteorological image pairs. , n and q represent the total number of meteorological image pairs and the total number of pixels, respectively.
[0072] On the basis of the above embodiment, a high-resolution meteorological image corresponding to the same observation time point is generated for each meteorological channel. The specific process is as follows:
[0073] The output high-resolution meteorological images corresponding to the low-resolution meteorological images in each meteorological image pair and each type of convolutional neural network model are simply recorded as the reconstructed images corresponding to each type of convolutional neural network model of each meteorological image pair, and the high-resolution meteorological images in each meteorological image pair are recorded as the original images of each meteorological image pair;
[0074] And according to the image editing software, the bit depth of the reconstructed image of each meteorological image pair corresponding to each type of convolutional neural network model and the original image of each meteorological image pair are checked respectively, where the bit depth is divided into 8 bits and 16 bits;
[0075] The meteorological image pairs whose bit depth of the reconstructed images of each type of convolutional neural network model is consistent with the bit depth of the original image are selected as sample meteorological image pairs, and the bit depth of each sample meteorological image pair is obtained. ,c is the number of each sample meteorological image pair;
[0076] If the bit depth of the original image of a certain meteorological image is identified to be 8 bits, and the bit depths of the reconstructed images of the corresponding meteorological images corresponding to each type of convolutional neural network model are all 8 bits, they are uniformly divided into 8-bit groups. If there is a corresponding meteorological image pair corresponding to a certain type of convolutional neural network model whose bit depth of reconstructed image is 16 bits, the output high-resolution meteorological image of this type of convolutional neural network model is discarded.
[0077] The possible maximum pixel value of the original image in each sample meteorological image pair is thus determined ;
[0078] The similarity between the original image in each sample meteorological image pair and the reconstructed image of each sample meteorological image pair corresponding to each type of convolutional neural network model is obtained by comprehensive calculation. , and arrange them in descending order, and select the convolutional neural network model type with the highest similarity of the corresponding bit depth of each sample meteorological image pair as the reference convolutional neural network model type of the meteorological image with the same bit depth as each sample meteorological image pair;
[0079] According to the image editing software, the meteorological images corresponding to the same observation time point of each meteorological channel are checked respectively, and the bit depth of the meteorological images corresponding to the same observation time point of each meteorological channel is obtained, which is input into the reference convolutional neural network model type of meteorological images with consistent bit depth of each sample meteorological image, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel.
[0080] Step 3: Conduct quality assessment on high-resolution meteorological images from each meteorological channel at the same observation time point;
[0081] Based on the above embodiment, the quality assessment includes image clarity assessment, structural similarity assessment and meteorological element accuracy assessment.
[0082] On the basis of the above embodiment, the image clarity evaluation is performed on the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel, and the specific process is as follows:
[0083] Import the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel into the rectangular coordinate system to obtain the position coordinates of each pixel point of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel , k is the number of each meteorological channel, k=1,2,...b, b is the total number of meteorological channels;
[0084] Calculate the contrast of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , is the pixel value of the position coordinate of the pth pixel point in the high-resolution meteorological image of the kth meteorological channel corresponding to the same observation time point;
[0085] Extract the set two-dimensional Laplacian operator convolution kernel from the meteorological database , and then calculate the Laplace value of each pixel of the high-resolution meteorological image corresponding to the same observation time point in each meteorological channel ;
[0086] Further, the non-reference image clarity evaluation value of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is obtained. ;
[0087] Comprehensive analysis obtains the image clarity assessment index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , where a1 and a2 represent the weighting factors of the predefined contrast and the definition evaluation value of the reference-free image, respectively, and a1+a2=1, They represent the predefined contrast reference difference and the non-reference image sharpness assessment difference, respectively. They are the set contrast threshold and the no-reference image clarity evaluation threshold respectively.
[0088] On the basis of the above embodiment, the structural similarity evaluation is performed on the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel. The specific process is as follows:
[0089] A meteorological channel is randomly selected from each meteorological channel as a target meteorological channel, thereby screening out a high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point and a high-resolution meteorological image of each other meteorological channel corresponding to the same observation time point, and the high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point is simply recorded as the meteorological image corresponding to the target meteorological channel, and similarly, the high-resolution meteorological images of each other meteorological channel corresponding to the same observation time point are simply recorded as the meteorological images corresponding to each other meteorological channel;
[0090] Synchronously obtain the brightness value of each pixel in the meteorological image corresponding to the target meteorological channel And the brightness value of each pixel in the meteorological image corresponding to other meteorological channels , r is the number of each other meteorological channel, r=1,2,...w, w represents the total number of other meteorological channels, and further calculates the average brightness of the meteorological image corresponding to the target meteorological channel , brightness variance And the average brightness of the corresponding meteorological images of other meteorological channels , brightness variance ;
[0091] Evaluate the structural similarity evaluation index of the target meteorological channel corresponding to the meteorological image ,in are predefined constants, , , D is the bit depth of the meteorological image corresponding to the target meteorological channel, is 0.01, is 0.01;
[0092] These are two small constants added to avoid the denominator being zero;
[0093] is the brightness covariance of the meteorological images corresponding to each other meteorological channel and the meteorological image corresponding to the target meteorological channel. The specific formula is as follows:
[0094] ;
[0095] In the same way, the structural similarity evaluation index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point is obtained. .
[0096] On the basis of the above embodiment, the accuracy of meteorological elements is evaluated for the high-resolution meteorological images corresponding to the same observation time point in each meteorological channel. The specific process is as follows:
[0097] Extract the first estimated value of each meteorological element from the meteorological image corresponding to the target meteorological channel , and extract other estimated values corresponding to each meteorological element from the corresponding meteorological images of other meteorological channels ;
[0098] Then, the mean absolute error of each meteorological element of the target meteorological channel corresponding to the meteorological image and the meteorological image corresponding to other meteorological channels is obtained by calculation. , Root Mean Square Error , correlation coefficient and the coefficient of determination , y is the number of each meteorological element, y=1,2,... , is the total number of meteorological elements;
[0099] It should be added that , , , are the first estimated mean of the meteorological elements in the meteorological image corresponding to the target meteorological channel and the other estimated means of the meteorological elements in the meteorological image corresponding to the rth other meteorological channel, ;
[0100] and will and Perform maximum normalization processing to obtain the average absolute error of each meteorological element between the meteorological image corresponding to the target meteorological channel and the meteorological images corresponding to other meteorological channels. and RMS error ;
[0101] It should be added that , , and They represent the maximum mean absolute error threshold and the maximum root mean square error threshold respectively.
[0102] Evaluate the accuracy of meteorological elements in the target meteorological channel corresponding to the meteorological image ,and are the corresponding indicator weights, and ;
[0103] According to the calculation method of the meteorological element accuracy evaluation index of the meteorological image corresponding to the target meteorological channel, the meteorological element accuracy evaluation index of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is calculated in the same way. .
[0104] Step 4: Based on the quality assessment results of the high-resolution meteorological images of each meteorological channel corresponding to the same observation time point, the practicality of the meteorological images of each meteorological channel corresponding to the same observation time point is evaluated to obtain the calculated practicality score of the meteorological images of each meteorological channel corresponding to the same observation time point;
[0105] On the basis of the above embodiment, the practicability of the meteorological images corresponding to the same observation time point of each meteorological channel is evaluated. The specific evaluation process is as follows:
[0106] The image clarity evaluation index of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is compared with the image clarity evaluation index interval corresponding to each image clarity score value stored in the meteorological database, so as to obtain the image clarity score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel, and the structural similarity score value and meteorological element accuracy score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel are obtained in the same way according to the acquisition method thereof;
[0107] The image clarity score, structural similarity score and meteorological element accuracy score of the high-resolution meteorological images of each meteorological channel at the same observation time point are added together, and the result of the addition is divided by 3 to obtain the computational utility score of the meteorological images of each meteorological channel at the same observation time point.
[0108] Step 5: Use the meteorological image corresponding to the meteorological channel ranked first in terms of the calculation utility score for meteorological analysis and forecasting at the observation time point.
[0109] Example 2
[0110] See also Figure 2 As shown, a meteorological element super-resolution image evaluation device includes a processor, a memory and a communication bus;
[0111] The memory stores a computer-readable program executable by the processor;
[0112] The communication bus realizes the connection and communication between the processor and the memory;
[0113] When the processor executes the computer-readable program, the processor implements the steps of a meteorological element super-resolution image assessment method as described in any one of the present inventions.
[0114] Example 3
[0115] A computer-readable storage medium, characterized in that: the computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement the steps in a meteorological element super-resolution image assessment method as described in any one of the present invention.
[0116] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0117] It should be understood that determining B based on A does not mean determining B only based on A. B can also be determined based on A and / or other information.
[0118] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.
[0119] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.
Claims
1. A method for evaluating meteorological element super-resolution images, characterized in that: The method comprises the following steps: Step 1: The image evaluation device receives meteorological images corresponding to the same observation time point from various meteorological channels; Step 2: applying super-resolution reconstruction technology to the meteorological images corresponding to the same observation time point of each meteorological channel, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel; The super-resolution reconstruction technology is applied to the meteorological images corresponding to the same observation time point in each meteorological channel. The specific process is as follows: Extract meteorological image pairs corresponding to the same historical observation time point from the meteorological database, and abbreviate them as meteorological image pairs, wherein each meteorological image pair contains a set of low-resolution meteorological images and a set of high-resolution meteorological images; And perform standardized scaling processing on each meteorological image pair so that the pixel value of each meteorological image in each meteorological image pair is maintained in the range of [0,1]; Extract various types of convolutional neural network models from the meteorological database, input low-resolution meteorological images within each meteorological image pair into each type of convolutional neural network model, thereby outputting high-resolution meteorological images corresponding to the outputs of each type of convolutional neural network model for the low-resolution meteorological images within each meteorological image pair; And obtain the RGB value of each pixel in each color channel in the output high-resolution meteorological image corresponding to each type of convolutional neural network model in each meteorological image pair , and simultaneously obtain the RGB value of each pixel in each color channel in each high-resolution meteorological image within each meteorological image pair , where i is the number of each meteorological image pair, i=1,2,..n, f is the number of each type of convolutional neural network model, f=1,2,..z, z is the total number of types corresponding to the convolutional neural network model, p is the number of each pixel, p=1,2,..q, m is the number of each color channel, m=m1,m2,m3, m1,m2 and m3 refer to the red channel, green channel and blue channel respectively; Further calculations were performed to obtain the mean square error of each type of convolutional neural network model corresponding to the meteorological image pairs. , n and q represent the total number of meteorological image pairs and the total number of pixels, respectively; Generate high-resolution meteorological images corresponding to the same observation time point for each meteorological channel. The specific process is as follows: The output high-resolution meteorological images corresponding to the low-resolution meteorological images in each meteorological image pair and each type of convolutional neural network model are simply recorded as the reconstructed images corresponding to each type of convolutional neural network model of each meteorological image pair, and the high-resolution meteorological images in each meteorological image pair are recorded as the original images of each meteorological image pair; And according to the image editing software, the bit depth of the reconstructed image of each meteorological image pair corresponding to each type of convolutional neural network model and the original image of each meteorological image pair are checked respectively, where the bit depth is divided into 8 bits and 16 bits; The meteorological image pairs whose bit depth of the reconstructed images of each type of convolutional neural network model is consistent with the bit depth of the original image are selected as sample meteorological image pairs, and the bit depth of each sample meteorological image pair is obtained. ,c is the serial number of each sample meteorological image pair; The possible maximum pixel value of the original image in each sample meteorological image pair is thus determined ; The similarity between the original image in each sample meteorological image pair and the reconstructed image of each sample meteorological image pair corresponding to each type of convolutional neural network model is obtained by comprehensive calculation. , and arrange them in descending order, and select the convolutional neural network model type with the highest similarity of the corresponding bit depth of each sample meteorological image pair as the reference convolutional neural network model type of the meteorological image with the same bit depth as each sample meteorological image pair; Using image editing software, the meteorological images corresponding to the same observation time point of each meteorological channel are checked respectively, the bit depth of the meteorological images corresponding to the same observation time point of each meteorological channel is obtained, and the bit depth is input into the reference convolutional neural network model type of each sample meteorological image for meteorological images with consistent bit depth, thereby generating high-resolution meteorological images corresponding to the same observation time point of each meteorological channel; Quality assessment includes image clarity assessment, structural similarity assessment, and meteorological element accuracy assessment; The practicality of meteorological images corresponding to the same observation time point in each meteorological channel is evaluated. The specific evaluation process is as follows: The image clarity evaluation index of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is compared with the image clarity evaluation index interval corresponding to each image clarity score value stored in the meteorological database, so as to obtain the image clarity score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel, and the structural similarity score value and meteorological element accuracy score value of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel are obtained in the same way according to the acquisition method thereof; The image clarity score, structural similarity score and meteorological element accuracy score of the high-resolution meteorological image of each meteorological channel at the same observation time point are added together, and the result of the addition is divided by 3 to obtain the calculation utility score of the meteorological image of each meteorological channel at the same observation time point; Step 3: Conduct quality assessment on high-resolution meteorological images from each meteorological channel at the same observation time point; Step 4: Based on the quality assessment results of the high-resolution meteorological images of each meteorological channel corresponding to the same observation time point, the practicability of the meteorological images of each meteorological channel corresponding to the same observation time point is evaluated to obtain the calculated practicability score of the meteorological images of each meteorological channel corresponding to the same observation time point; Step 5: Use the meteorological image corresponding to the meteorological channel ranked first in terms of the calculation utility score for meteorological analysis and forecasting at the observation time point.
2. A meteorological element super-resolution image evaluation method according to claim 1, characterized in that: The image clarity of high-resolution meteorological images corresponding to the same observation time point of each meteorological channel is evaluated. The specific process is as follows: Import the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel into the rectangular coordinate system to obtain the position coordinates of each pixel point of the high-resolution meteorological images corresponding to the same observation time point of each meteorological channel , k is the number of each meteorological channel, k=1,2,...b, b is the total number of meteorological channels; Calculate the contrast of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , is the pixel value of the position coordinate of the pth pixel point in the high-resolution meteorological image of the kth meteorological channel corresponding to the same observation time point; Extract the set two-dimensional Laplacian operator convolution kernel from the meteorological database , and then calculate the Laplace value of each pixel of the high-resolution meteorological image corresponding to the same observation time point in each meteorological channel ; Further, the non-reference image clarity evaluation value of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is obtained. ; Comprehensive analysis obtains the image clarity assessment index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point , where a1 and a2 represent the weighting factors of the predefined contrast and the definition evaluation value of the reference-free image, respectively, and a1+a2=1, They represent the predefined contrast reference difference and the non-reference image sharpness assessment difference, respectively. They are the set contrast threshold and the no-reference image clarity evaluation threshold respectively.
3. The method for evaluating meteorological element super-resolution images according to claim 1, characterized in that: The structural similarity of high-resolution meteorological images corresponding to the same observation time point in each meteorological channel is evaluated. The specific process is as follows: A meteorological channel is randomly selected from each meteorological channel as a target meteorological channel, thereby screening out a high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point and a high-resolution meteorological image of each other meteorological channel corresponding to the same observation time point, and the high-resolution meteorological image of the target meteorological channel corresponding to the same observation time point is simply recorded as the meteorological image corresponding to the target meteorological channel, and similarly, the high-resolution meteorological images of each other meteorological channel corresponding to the same observation time point are simply recorded as the meteorological images corresponding to each other meteorological channel; Synchronously obtain the brightness value of each pixel in the meteorological image corresponding to the target meteorological channel And the brightness value of each pixel in the meteorological image corresponding to other meteorological channels , r is the number of each other meteorological channel, r=1,2,...w, w is the total number of other meteorological channels, in order to further calculate the average brightness of the meteorological image corresponding to the target meteorological channel , brightness variance And the average brightness of the corresponding meteorological images of other meteorological channels , brightness variance ; Evaluate the structural similarity evaluation index of the target meteorological channel corresponding to the meteorological image ,in are predefined constants, is the brightness covariance of the meteorological images corresponding to other meteorological channels and the meteorological images corresponding to the target meteorological channel. The specific formula is as follows: ; In the same way, the structural similarity evaluation index of high-resolution meteorological images of each meteorological channel corresponding to the same observation time point is obtained. .
4. The method for evaluating meteorological element super-resolution images according to claim 1, characterized in that: The accuracy of meteorological elements is evaluated for the high-resolution meteorological images of each meteorological channel at the same observation time point. The specific process is as follows: Extracting a first estimated value of each meteorological element from the meteorological image corresponding to the target meteorological channel, and extracting other estimated values of each meteorological element from the meteorological images corresponding to other meteorological channels; Then, the mean absolute error of each meteorological element of the target meteorological channel corresponding to the meteorological image and the meteorological image corresponding to other meteorological channels is obtained by calculation. , Root Mean Square Error , correlation coefficient and the coefficient of determination , y is the number of each meteorological element, y=1,2,... , is the total number of meteorological elements; and will and Perform maximum normalization processing to obtain the average absolute error of each meteorological element between the meteorological image corresponding to the target meteorological channel and the meteorological images corresponding to other meteorological channels. and RMS error ; Evaluate the accuracy of meteorological elements in the target meteorological channel corresponding to the meteorological image ,and are the corresponding indicator weights, and ; According to the calculation method of the meteorological element accuracy evaluation index of the meteorological image corresponding to the target meteorological channel, the meteorological element accuracy evaluation index of the high-resolution meteorological image corresponding to the same observation time point of each meteorological channel is calculated in the same way. .
5. A meteorological element super-resolution image evaluation device, characterized in that: It is implemented based on a meteorological element super-resolution image assessment method according to any one of claims 1 to 4, and includes a processor, a memory and a communication bus; The memory stores a computer-readable program executable by the processor; The communication bus realizes the connection and communication between the processor and the memory; When the processor executes the computer-readable program, it is implemented to implement a meteorological element super-resolution image assessment method as described in any one of claims 1-4.
6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores one or more programs, and the one or more programs can be executed by one or more processors to implement a meteorological element super-resolution image assessment method as described in any one of claims 1-4.
