A cloud particle data processing method based on optimized clustering

Abstract

The present invention relates to a cloud particle data processing method based on optimized clustering, which mainly comprises firstly dividing and screening original cloud particle image data collected by cloud particle detection equipment, then processing all image data by image pre-clustering method, then merging pixel blocks of pixel block image data and extracting cloud particle regions. By means of this method, clustering processing based on multiple pixel block size parameters and multiple search range parameters can be performed on randomly screened cloud particle data, different clustering processing can be performed on original cloud particle data of different collection time periods, and different clustering results of cloud particle data can be analyzed based on ratio discreteness, thereby improving the accuracy and reliability of cloud particle region extraction.

Description

Technical Field

[0001] The invention belongs to the field of image processing and meteorology, and specifically relates to pre-clustering of cloud particle image data obtained by airborne cloud particle detection equipment, and extracting the cloud particle image data on this basis. Background Art

[0002] As the main detection method for modern cloud microphysics technology research in my country, the aircraft cloud penetration experiment is a cutting-edge observation technology that has only been developed in China in recent years. Compared with the observation of clouds using remote sensing equipment such as radar, the cloud information obtained by the cloud particle imager is more accurate and intuitive. The use of airborne detection equipment to obtain cloud particle images has become the basis for cloud microphysics research. Effective quality control of the raw data obtained by the cloud particle imager is the basis for the research of cloud particle shape recognition, cloud microphysics vertical structure analysis, cloud microphysics parameter inversion, and artificial influence on cloud precipitation process.

[0003] At present, it is an important development direction to analyze and process cloud particle image data based on artificial intelligence methods. However, artificial intelligence-based analysis methods often require a large amount of data as support. Although cloud particle image data can meet the data volume requirements of artificial intelligence methods, how to effectively convert massive amounts of raw cloud particle image data into usable cloud particle data becomes a key issue.

[0004] The data collected by cloud particle detection equipment often have problems such as particle breakage, pixel loss, and false targets. These interferences will lead to a decrease in the quality of cloud particle data, affecting the accuracy of subsequent cloud particle classification and identification, cloud microphysical vertical structure analysis, and cloud microphysical parameter inversion, and may even further affect the effectiveness of artificial influence operations in the cloud precipitation process. Summary of the invention

[0005] In view of the above-mentioned deficiencies of the prior art, the present invention proposes a cloud particle data processing method based on optimized clustering, which comprises the following steps:

[0006] A cloud particle data processing method based on optimized clustering mainly includes the following steps:

[0007] Step 1: Divide the original cloud particle image data collected by the cloud particle detection equipment into a data set to obtain a sub-data set D composed of the original cloud particle image data. N and random raw cloud particle image dataset R N , where N is the data set number;

[0008] Step 2: For each random raw cloud particle image dataset R obtained in step 1 NAll image data in the image pre-clustering method are processed, and different pre-clustered cloud particle image data sets are generated by setting different pixel block size parameters and search range parameters. Then, according to the statistical distribution characteristics of the pixel block histograms in different pre-clustered cloud particle image data sets, different pixel block size parameters and search range parameters are analyzed to obtain the N The corresponding filtered parameters;

[0009] Step 3: Use the data obtained in step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data;

[0010] Step 4: Merge all pixel block image data obtained in step 3, and then extract the cloud particle area.

[0011] Furthermore, the step 1: the original cloud particle image data D collected by the cloud particle detection device all The data set is divided into two parts according to the different temperature layers at the time of data collection: the original cloud particle image data D all Divide into n parts, and obtain a sub-dataset D consisting of n original cloud particle image data N , where N is the number of the original cloud particle image sub-dataset, N∈[1,n], and then for each sub-dataset D N Perform random data selection from each sub-dataset D N Randomly select m original cloud particle image data from the dataset, m∈[1,M], the value of m is random within its value range each time it is selected, and M is the number of image data contained in each sub-dataset, thus forming n random original cloud particle image datasets R containing m original cloud particle image data. N , where N is the random original cloud particle image dataset number.

[0012] Further, step 2: for each random original cloud particle image dataset R obtained in step 1 N All image data in the image pre-clustering method are processed, including:

[0013] Step 2.1: Read R N The original cloud particle image data size is selected and the smaller size is set as S. The small, medium and large pixel block size parameters k are set. i =α i S, where i is an integer, i∈[1,3], α i is the pixel block size weight, and sets the search range parameter f corresponding to different pixel block sizes i =β1·ki -β2, where β1 and β2 are search range weights;

[0014] Step 2.2: Randomly select the three pixel block size parameters and the search range parameters corresponding to different pixel block sizes in step 2.1 within their parameter value ranges, and perform random selection on the random original cloud particle image dataset R N Clustering processing based on three pixel block size parameters and corresponding search range parameters is performed to generate a pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks: small pixel blocks, medium pixel blocks, and large pixel blocks. N,i , where variable N is the random raw cloud particle image dataset number, and variable i represents the pixel block size type;

[0015] Step 2.3: The pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks obtained in step 2.2 N,i , generate the histogram h of all pixel blocks N,i,Q , where N is the random raw cloud particle image dataset number, i represents the pixel block size type, and Q is the dataset RC N,i The number of the pixel block in, Q∈[1,q], the variable q is the data set RC N,i The number of pixels in the block;

[0016] Step 2.4: For all histograms h obtained in step 2.3 N,i,Q , set the threshold gv, count the maximum ordinate value in the area with grayscale value greater than or equal to gv and the maximum ordinate value in the area less than gv in each histogram, and calculate the ratio Rh of the smaller value to the larger value of the two maximum ordinate values N,i,Q ;

[0017] Step 2.5: For the ratio Rh obtained in step 2.4 N,i,Q , grouping according to the rule of dividing the same N,i together, and obtain n·3 groups of ratio data sets, each of which contains q ratios Rh N,i,Q ;

[0018] Step 2.6: Analyze the n·3 ratio data sets obtained in step 2.5, and calculate the q ratios Rh in each data set. N,i,Q Calculate the ratio dispersion Ah N,M,i,Q =(Rh N,i,Q -1) 2 , set the threshold th, count the number of ratio dispersions greater than the threshold th and less than the threshold th in each ratio data set, and calculate the ratio of the smaller value to the larger value of the two numbers RhD N,i ;

[0019] Step 2.7: For the n·3 ratios RhD obtained in step 2.6 N,i , grouping is done according to the rule that the same N but different i are grouped together, and n groups of ratio data sets are obtained, each of which contains 3 ratios RhD N,i ;

[0020] Step 2.8: Analyze the n ratio data sets obtained in step 2.7 and analyze the three ratios RhD in each data set. N Compare and select the smallest value as minRhD N , and record all the parameters corresponding to the ratio;

[0021] Step 2.9: Repeat steps 2.2 to 2.8 to obtain new n minRhD N (d), and record all the parameters corresponding to the ratio. The variable d represents the number of repetitions, d = [0, D], where D is the number of repetitions;

[0022] Step 2.10: For the n·(D+1) minRhD obtained in step 2.9 N (d) Analyze and group the groups according to the rule that groups with the same N but different d are grouped together, and select the minimum minRhD in each group N (d) is denoted as MinRhD N , save all the corresponding variable parameters, and thus get n MinRhD N And all its corresponding variable parameters.

[0023] Further, step 3: use the result from step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data:

[0024] Step 3.1: D obtained in step 1 N and the N MinRhD obtained in step 2 N And all its corresponding variable parameters are matched, and the matching rule is the same data set D of N N With MinRhD N And all its corresponding variable parameters are associated and matched;

[0025] Step 3.2: The pixel block size parameter k in the variable parameter obtained by the association matching in step 3.1 is i =α i S, search range parameter f i =β1·k i -β2 is applied to the corresponding dataset D N All image data FN,j On the data F N,j The image data result SF after processing a specific pixel block N,j , where variable N is the original cloud particle image sub-dataset number, and j represents the image data number in the same sub-dataset.

[0026] The present invention solves the following technical problems:

[0027] 1. The original cloud particle image data was randomly screened and pre-clustered on the random samples, which reduced the amount of calculation and improved the accuracy of cloud particle area extraction.

[0028] 2. The randomly selected cloud particle data were clustered based on a variety of pixel block size parameters and a variety of search range parameters. Different clustering processes were performed on the original cloud particles in different acquisition time periods, which improved the robustness of cloud particle area extraction.

[0029] 3. An analysis of different clustering results of cloud particle data was conducted based on ratio dispersion, and the effectiveness of different clustering results was compared. A more appropriate clustering scheme was selected to improve the accuracy and reliability of cloud particle area extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a cloud particle data processing method based on optimized clustering; DETAILED DESCRIPTION

[0031] The following will provide a clear and complete description of the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings in the embodiments of the present invention. The method flow chart is as follows: Figure 1 As shown, the following steps are included:

[0032] A cloud particle data processing method based on optimized clustering mainly includes the following steps:

[0033] Step 1: Divide the original cloud particle image data collected by the cloud particle detection equipment into a data set to obtain a sub-data set D composed of the original cloud particle image data. N and random raw cloud particle image dataset R N , where N is the data set number;

[0034] Step 2: For each random raw cloud particle image dataset R obtained in step 1 NAll image data in the image pre-clustering method are processed, and different pre-clustered cloud particle image data sets are generated by setting different pixel block size parameters and search range parameters. Then, according to the statistical distribution characteristics of the pixel block histograms in different pre-clustered cloud particle image data sets, different pixel block size parameters and search range parameters are analyzed to obtain the N The corresponding filtered parameters;

[0035] Step 3: Use the data obtained in step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data;

[0036] Step 4: Merge all pixel block image data obtained in step 3, and then extract the cloud particle area.

[0037] Furthermore, the step 1: the original cloud particle image data D collected by the cloud particle detection device all The data set is divided into two parts according to the different temperature layers at the time of data collection: the original cloud particle image data D all Divide into n parts, and obtain a sub-dataset D consisting of n original cloud particle image data N , where N is the number of the original cloud particle image sub-dataset, N∈[1,n], n∈[1,24], and then for each sub-dataset D N Perform random data selection from each sub-dataset D N Randomly select m original cloud particle image data from the dataset, m∈[1,M], the value of m is random within its value range each time it is selected, and M is the number of image data contained in each sub-dataset, thus forming n random original cloud particle image datasets R containing m original cloud particle image data. N , where N is the random original cloud particle image dataset number.

[0038] Further, step 2: for each random original cloud particle image dataset R obtained in step 1 N All image data in the image pre-clustering method are processed, including:

[0039] Step 2.1: Read R N The original cloud particle image data size is selected and the smaller size is set as S. The small, medium and large pixel block size parameters k are set. i =α i S, where i is an integer, i∈[1,3], α iis the pixel block size weight, α1∈[0.008, 0.012], α2∈[0.02, 0.03], α3∈[0.045, 0.055], and sets the search range parameter f corresponding to different pixel block sizes. i =β1·k i -β2, where β1 and β2 are search range weights, β1∈[0.1,0.5], β2∈[1,5];

[0040] Step 2.2: Randomly select the three pixel block size parameters and the search range parameters corresponding to different pixel block sizes in step 2.1 within their parameter value ranges, and perform random selection on the random original cloud particle image dataset R N Clustering processing based on three pixel block size parameters and corresponding search range parameters is performed to generate a pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks: small pixel blocks, medium pixel blocks, and large pixel blocks. N,i , where variable N is the random raw cloud particle image dataset number, and variable i represents the pixel block size type;

[0041] Step 2.3: The pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks obtained in step 2.2 N,i , generate the histogram h of all pixel blocks N,i,Q , where N is the random raw cloud particle image dataset number, i represents the pixel block size type, and Q is the dataset RC N,i The number of the pixel block in, Q∈[1,q], the variable q is the data set RC N,i The number of pixels in the block;

[0042] Step 2.4: For all histograms h obtained in step 2.3 N,i,Q , set the threshold gv, count the maximum ordinate value in the area with grayscale value greater than or equal to gv and the maximum ordinate value in the area less than gv in each histogram, and calculate the ratio Rh of the smaller value to the larger value of the two maximum ordinate values N,i,Q , where the threshold gv∈[100,120];

[0043] Step 2.5: For the ratio Rh obtained in step 2.4 N,i,Q , grouping according to the rule of dividing the same N,i together, and obtain n·3 groups of ratio data sets, each of which contains q ratios Rh N,i,Q ;

[0044] Step 2.6: Analyze the n·3 ratio data sets obtained in step 2.5, and calculate the q ratios Rh in each data set. N,i,Q Calculate the ratio dispersion Ah N,M,i,Q =(Rh N,i,Q -1)2 , set the threshold th, count the number of ratio dispersions greater than the threshold th and less than the threshold th in each ratio data set, and calculate the ratio of the smaller value to the larger value of the two numbers RhD N,i , threshold th∈[0.5,0.6];

[0045] Step 2.7: For the n·3 ratios RhD obtained in step 2.6 N,i , grouping is done according to the rule that the same N but different i are grouped together, and n groups of ratio data sets are obtained, each of which contains 3 ratios RhD N,i ;

[0046] Step 2.8: Analyze the n ratio data sets obtained in step 2.7 and analyze the three ratios RhD in each data set. N Compare and select the smallest value as minRhD N , and record all the parameters corresponding to the ratio;

[0047] Step 2.9: Repeat steps 2.2 to 2.8 to obtain new n minRhD N (d) and record all the parameters corresponding to the ratio. The variable d represents the number of repetitions, d = [0, D], D is the number of repetitions, where D ∈ [1, 20];

[0048] Step 2.10: For the n·(D+1) minRhD obtained in step 2.9 N (d) Analyze and group the groups according to the rule that groups with the same N but different d are grouped together, and select the minimum minRhD in each group N (d) is denoted as MinRhD N , save all the corresponding variable parameters, and thus get n MinRhD N And all its corresponding variable parameters.

[0049] Further, step 3: use the result from step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data:

[0050] Step 3.1: D obtained in step 1 N and the N MinRhD obtained in step 2 N And all its corresponding variable parameters are matched, and the matching rule is the same data set D of N N With MinRhD N And all its corresponding variable parameters are associated and matched;

[0051] Step 3.2: The pixel block size parameter k in the variable parameter obtained by the association matching in step 3.1 is i =α i S, search range parameter f i =β1·k i -β2 is applied to the corresponding dataset D N All image data F N,j On the data F N,j The image data result SF after processing a specific pixel block N,j , where variable N is the original cloud particle image sub-dataset number, and j represents the image data number in the same sub-dataset.

[0052] Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, other embodiments obtained by ordinary technicians in this field without creative work are all within the scope of protection of the present invention.

Claims

1. A cloud particle data processing method based on optimized clustering mainly includes the following steps: Step 1: Divide the original cloud particle image data collected by the cloud particle detection equipment into a data set to obtain a sub-data set D composed of the original cloud particle image data. N and random raw cloud particle image dataset R N , where N is the data set number; Specifically, the original cloud particle image data D collected by the cloud particle detection equipment all The data set is divided into two parts according to the different temperature layers at the time of data collection: the original cloud particle image data D all Divide into n parts, and obtain a sub-dataset D consisting of n original cloud particle image data N , where N is the number of the original cloud particle image sub-dataset, N∈[1,n], and then for each sub-dataset D N Perform random data selection from each sub-dataset D N Randomly select m original cloud particle image data from the dataset, m∈[1,M], the value of m is random within its value range each time it is selected, and M is the number of image data contained in each sub-dataset, thus forming n random original cloud particle image datasets R containing m original cloud particle image data. N , where N is the random original cloud particle image dataset number; Step 2: For each random raw cloud particle image dataset R obtained in step 1 N All image data in the image pre-clustering method are processed, and different pre-clustered cloud particle image data sets are generated by setting different pixel block size parameters and search range parameters. Then, according to the statistical distribution characteristics of the pixel block histograms in different pre-clustered cloud particle image data sets, different pixel block size parameters and search range parameters are analyzed to obtain the N The corresponding filtered parameters; Specifically: Step 2.1: Read R N The original cloud particle image data size is selected and the smaller size is set as S. The small, medium and large pixel block size parameters k are set. i =α i S, where i is an integer, i∈[1,3], α i is the pixel block size weight, and sets the search range parameter f corresponding to different pixel block sizes i =β1·k i -β2, where β1 and β2 are search range weights; Step 2.2: Randomly select the three pixel block size parameters and the search range parameters corresponding to different pixel block sizes in step 2.1 within their parameter value ranges, and perform random selection on the random original cloud particle image dataset R N Clustering processing based on three pixel block size parameters and corresponding search range parameters is performed to generate a pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks: small pixel blocks, medium pixel blocks, and large pixel blocks. N,i , where variable N is the random raw cloud particle image dataset number, and variable i represents the pixel block size type; Step 2.3: The pre-clustered cloud particle image dataset RC containing three sizes of pixel blocks obtained in step 2.2 N,i , generate the histogram h of all pixel blocks N,i,Q , where N is the random raw cloud particle image dataset number, i represents the pixel block size type, and Q is the dataset RC N,i The number of the pixel block in, Q∈[1,q], the variable q is the data set RC N,i The number of pixel blocks in the Step 2.4: For all histograms h obtained in step 2.3 N,i,Q , set the threshold gv, count the maximum ordinate value in the area with grayscale value greater than or equal to gv and the maximum ordinate value in the area less than gv in each histogram, and calculate the ratio Rh of the smaller value to the larger value of the two maximum ordinate values N,i,Q ; Step 2.5: For the ratio Rh obtained in step 2.4 N,i,Q , grouping according to the rule of dividing the same N,i together, and obtain n·3 groups of ratio data sets, each of which contains q ratios Rh N,i,Q ; Step 2.6: Analyze the n·3 ratio data sets obtained in step 2.5, and calculate the q ratios Rh in each data set. N,i,Q Calculate the ratio dispersion Ah N,M,i,Q =(Rh N,i,Q -1) 2 , set the threshold th, count the number of ratio dispersions greater than the threshold th and less than the threshold th in each ratio data set, and calculate the ratio of the smaller value to the larger value of the two numbers RhD N,i ; Step 2.7: For the n·3 ratios RhD obtained in step 2.6 N,i , grouping is done according to the rule that the same N but different i are grouped together, and n groups of ratio data sets are obtained, each of which contains 3 ratios RhD N,i ; Step 2.8: Analyze the n ratio data sets obtained in step 2.7 and analyze the three ratios RhD in each data set. N Compare and select the smallest value as minRhD N , and record all the parameters corresponding to the ratio; Step 2.9: Repeat steps 2.2 to 2.8 to obtain new n minRhD N (d), and record all the parameters corresponding to the ratio. The variable d represents the number of repetitions, d = [0, D], where D is the number of repetitions; Step 2.10: For the n·(D+1) minRhD obtained in step 2.9 N (d) Analyze and group the groups according to the rule that groups with the same N but different d are grouped together, and select the minimum minRhD in each group N (d) is denoted as MinRhD N , save all the corresponding variable parameters, and thus get n MinRhD N And all its corresponding variable parameters; Step 3: Use the data obtained in step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data; Step 4: Merge all pixel block image data obtained in step 3, and then extract the cloud particle area.

2. A cloud particle data processing method based on optimized clustering according to claim 1, characterized in that: Step 3: Use the data obtained in step 2 and each R N The corresponding filtered parameters for the sub-dataset D with the same number N N Perform pixel block processing to obtain pixel block image data: Step 3.1: D obtained in step 1 N and the N MinRhD obtained in step 2 N And all its corresponding variable parameters are matched, and the matching rule is the same data set D of N N With MinRhD N And all its corresponding variable parameters are associated and matched; Step 3.2: The pixel block size parameter k in the variable parameter obtained by the association matching in step 3.1 is i =α i S, search range parameter f i =β1·k i -β2 is applied to the corresponding dataset D N All image data F N,j On the data F N,j The image data result SF after processing a specific pixel block N,j , where variable N is the original cloud particle image sub-dataset number, and j represents the image data number in the same sub-dataset.

3. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: The original cloud particle image data D all Divide into n parts, where n∈[1,24].

4. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: Step 2.1: Set the small, medium, and large pixel block size parameters k i =α i S, where i is an integer, i∈[1,3], α i is the pixel block size weight, where α1∈[0.008, 0.012], α2∈[0.02, 0.03], α3∈[0.045, 0.055].

5. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: Step 2.1: Set the search range parameter f corresponding to different pixel block sizes i =β1·k i -β2, where β1 and β2 are search range weights, where β1∈[0.1,0.5], β2∈[1,5].

6. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: Step 2.4: Set the threshold gv, count the maximum ordinate value in the area with grayscale value greater than or equal to gv and the maximum ordinate value in the area with grayscale value less than gv in each histogram, and calculate the ratio Rh of the smaller value to the larger value of the two maximum ordinate values N,i,Q , where the threshold gv∈[100,120].

7. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: Step 2.6: For each set of q ratios Rh N,i,Q Calculate the ratio dispersion Ah N,M,i,Q =(Rh N,i,Q -1) 2 , set the threshold th, where the threshold th∈[0.5,0.6].

8. The cloud particle data processing method based on optimized clustering according to claim 1 is characterized in that: Step 2.9: The variable d represents the number of repetitions, d = [0, D], D is the number of repetitions, where D ∈ [1, 20].

Images