A data cleaning method and device for remote sensing images and missing attribute information
By performing banding, local anomalies and band anomalies detection on satellite remote sensing images, and combining multi-time phase image information repair and association rules to fill attribute information, the problem of data incompleteness of remote sensing images is solved, and the accuracy and production of data analysis are improved. efficiency.
Patent Information
- Application Number
- CN202411178341.8
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-08-27
AI Technical Summary
Data incompleteness caused by the acquisition channel and modeling differences in satellite remote sensing image data affects the accuracy of data extraction mode and the accuracy of analysis results. The single structured missing processing method of the prior art is not suitable for unstructured remote sensing images. data.
The data cleaning method for the missing remote sensing images and attribute information is adopted, and missing data is repaired and supplemented through band detection, local anomaly detection and band abnormality detection, combined with the repair algorithm of multi-time phase image information and the attribute information filling method of association rules.
It effectively solves the abnormal problems caused by the lack of image and attribute information in business production, improves the accuracy of data analysis and image production efficiency, and ensures the integrity and use value of data.
Smart Images

Figure CN118689867B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of remote sensing big data cleaning, and in particular to a data cleaning method and device for remote sensing images and missing attribute information. Background Art
[0002] In practical applications, due to differences in satellite remote sensing image data acquisition channels and modeling, the remote sensing image data obtained by satellite remote sensing image data centers and various user units, such as the remote sensing images themselves and their supporting attribute files, contain incomplete data, which will affect the correctness of the data extraction model and the accuracy of the export rules, thereby producing inaccurate analysis results and decisions.
[0003] Missing data is a common type of dirty data. Its existence is a major factor in destroying data integrity, so it is necessary to deal with it appropriately. Traditional data missing generally refers to structured data missing. Remote sensing images belong to unstructured data. Due to their own particularity, their data missing phenomenon can be divided into two categories: one is the missing of remote sensing image entity data, that is, there is data missing in the image itself, and its features cannot be identified through recording. It needs to be eliminated or processed by relevant algorithms before image production. Based on the current phenomenon of missing image information, it mainly includes four situations: horizontal stripes, vertical stripes, local pixel information missing and image band anomalies. The other is the missing of key attribute record information, which mainly occurs in the attribute database files supporting remote sensing images (such as XML data, supporting database tables, etc.). For example, key field information such as latitude and longitude, payload, imaging mode, and receiving station in the scene catalog data will be missing, resulting in low query efficiency for scene imaging data, low image utilization, and inability to issue tasks. Business problems, such as this type of missing data can be directly discovered through detection algorithms.
[0004] Therefore, due to the complexity of remote sensing image data, the traditional single structured missing data and unstructured missing data processing methods are not suitable for remote sensing image data missing processing. It is necessary to adopt a data cleaning strategy for remote sensing images and their missing attribute information. By filtering out images that cannot meet the needs, improving the image attribute information records, improving the image production efficiency, data analysis accuracy and use value, it also provides technical support for large-scale remote sensing image data cleaning. Summary of the invention
[0005] In view of this, the embodiments of the present disclosure provide a method and device for cleaning data of remote sensing images and missing attribute information, which at least partially solve the problems existing in the prior art.
[0006] In a first aspect, the present disclosure provides a data cleaning method for remote sensing images and missing attribute information, including:
[0007] Acquire a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data;
[0008] Performing a missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data strip detection, remote sensing image data local anomaly detection, and band anomaly detection; and
[0009] Process the missing remote sensing image data and attribute information.
[0010] In a specific implementation, remote sensing image data strip detection includes row direction strip detection and column direction strip detection.
[0011] The row direction stripe detection includes:
[0012] Get the row mean parameters of all bands;
[0013] Calculate the row mean gradient between two adjacent rows; and
[0014] When the row mean gradient is greater than a first threshold, determining that the remote sensing image data is missing row information;
[0015] The column direction strip detection includes:
[0016] Get the column mean parameters of all bands;
[0017] Calculate the row mean gradient between two adjacent columns; and
[0018] When the column mean gradient is greater than a second threshold, it is determined that the remote sensing image data is missing column information.
[0019] In a specific implementation, remote sensing image data local anomaly detection includes determining whether there is a local anomaly in the remote sensing image data by the following formula:
[0020]
[0021] In the formula, , The pixel values in row and column directions are continuous x The number of pixels, Length is the actual distance on the ground, where the threshold is taken N , GSD is the image resolution, in units of m , and when the above conditions are met, it is determined that there is a local anomaly in the remote sensing image data.
[0022] In a specific implementation, when one of the following equations is satisfied, it is determined that there is a band anomaly:
[0023]
[0024] In the formula, For the The pixel value of a single pixel in the band, For the The pixel value of a single pixel in the band, and Represent different bands. is the maximum value of a certain band pixel value. It is the minimum value of the corresponding band pixel value.
[0025] In a specific implementation, processing the missing remote sensing image data includes:
[0026] When there is replacement data for the missing remote sensing image data, the missing remote sensing image data is directly deleted.
[0027] In a specific implementation, processing the missing remote sensing image data includes repairing the missing portion of the remote sensing image by a repair algorithm based on multi-temporal image information, and includes:
[0028] Obtain multi-source remote sensing image data sets and determine whether there are any missing images in the remote sensing image data;
[0029] Construct a missing sample dataset of multi-source remote sensing image data;
[0030] Perform a global traversal of the missing remote sensing image data, calculate the similarity measure between each phase block and the missing part of the remote sensing image, select the phase block with the highest similarity as the phase complementary information, and reduce the phase difference through linear regression transformation;
[0031] The missing parts of the remote sensing image and the selected time blocks with the highest similarity are subjected to feature extraction using multiple scale convolution filters, and the extracted results are fused;
[0032] Build an improved GAN remote sensing image missing repair model, train the model, and determine the model parameter settings; and
[0033] The improved GAN remote sensing image missing repair model is used to repair the missing parts of the remote sensing image.
[0034] In a specific implementation, labels are also attached to the missing remote sensing image data to record the missing row and column positions or the abnormal band positions in the database.
[0035] In a specific implementation, the method further includes performing missing detection on the attribute information, including:
[0036] Check the attribute information corresponding to the remote sensing image data to determine whether the attribute value that is not allowed to be empty is empty;
[0037] Establish missing identification rules between metadata XML and attribute database tables, and check the content in the attribute database to determine whether data is missing; and
[0038] Metadata XML information is mined to fill in missing data in the attribute database.
[0039] In a specific implementation, mining the metadata XML information to fill in missing data in the attribute database includes:
[0040] Metadata information extraction, including database table metadata information extraction and metadata XML file information extraction;
[0041] Metadata attribute information is stored in the database, and the extracted metadata attribute information is imported into the data warehouse table;
[0042] Metadata content verification: formulate quality verification rules for metadata attribute information after entering the table;
[0043] Use the general modeling and governance tools of the big data application platform to develop image data attribute missing verification rule components;
[0044] Establish verification and identification missing rules through association rules and other algorithms to detect missing data; and
[0045] By associating metadata XML attribute information, repair and backfill missing data fields.
[0046] In a second aspect, the present disclosure provides a data cleaning device for missing remote sensing images and attribute information, including:
[0047] A remote sensing image data set acquisition device, used to acquire a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data;
[0048] A missing check device is used to perform missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data strip detection, remote sensing image data local anomaly detection and band anomaly detection; and
[0049] A processing device is used to process missing remote sensing image data and attribute information.
[0050] The data cleaning method for missing remote sensing images and attribute information in the embodiment of the present disclosure includes: obtaining a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data; performing a missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data stripe detection, remote sensing image data local anomaly detection and band anomaly detection; and processing the missing remote sensing image data and attribute information. Through the scheme of the present disclosure, business problems such as unnecessary production, processing, access anomalies, etc. caused by missing images and attribute information in the actual business production process of remote sensing images can be effectively solved. At the same time, this technology can also be used in large-scale remote sensing image business production processing, cleaning management, data service and other systems to support large-scale image cleaning processing or image quality inspection. BRIEF DESCRIPTION OF THE DRAWINGS
[0051] In order to more clearly illustrate the technical solutions of the embodiments of the present disclosure, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.
[0052] Figure 1 A data cleaning flow chart for missing remote sensing images and their attribute information according to an embodiment of the present disclosure is shown;
[0053] Figure 2 A flowchart of a restoration algorithm based on multi-temporal image information according to an embodiment of the present disclosure is shown;
[0054] Figure 3 A flowchart of a technology for filling missing values of attribute information according to an embodiment of the present disclosure is shown;
[0055] Figure 4 Image row and column direction stripe information missing diagram according to an embodiment of the present disclosure;
[0056] Figure 5 A row mean line graph according to an embodiment of the present disclosure;
[0057] Figure 6 A statistical diagram of the distribution of a pixel histogram of a row where local information of an image is missing according to an embodiment of the present disclosure. DETAILED DESCRIPTION
[0058] The embodiments of the present disclosure are described in detail below with reference to the accompanying drawings.
[0059] The following describes the embodiments of the present disclosure through specific examples, and those skilled in the art can easily understand other advantages and effects of the present disclosure from the contents disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present disclosure, rather than all of the embodiments. The present disclosure can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present disclosure. It should be noted that the following embodiments and features in the embodiments can be combined with each other without conflict. Based on the embodiments in the present disclosure, all other embodiments obtained by ordinary technicians in the field without making creative work are within the scope of protection of the present disclosure.
[0060] It should be noted that various aspects of the embodiments within the scope of the appended claims are described below. It should be apparent that the aspects described herein may be embodied in a wide variety of forms, and any specific structure and / or function described herein is merely illustrative. Based on the present disclosure, it should be understood by those skilled in the art that an aspect described herein may be implemented independently of any other aspect, and two or more of these aspects may be combined in various ways. For example, any number of aspects described herein may be used to implement the device and / or practice the method. In addition, other structures and / or functionalities other than one or more of the aspects described herein may be used to implement this device and / or practice this method.
[0061] It should also be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present disclosure. The drawings only show components related to the present disclosure rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.
[0062] Additionally, in the following description, specific details are provided to facilitate a thorough understanding of the examples. However, it will be understood by those skilled in the art that the aspects described may be practiced without these specific details.
[0063] One of the purposes of the present invention is to provide a data cleaning method for missing remote sensing images and their attribute information. Through satellite remote sensing image entity information missing detection and response strategies, and attribute information missing value filling methods based on association rules, the problem of missing remote sensing image information and attribute library table keyword fields is solved to ensure data integrity, correctness of data extraction mode and accuracy of export rules, and provide effective support for large-scale remote sensing data cleaning and effective image production.
[0064] Next, with reference to the accompanying drawings, the data cleaning method for remote sensing images and missing attribute information of the present invention is described in detail.
[0065] Aiming at the phenomenon of missing entity data and missing attribute record information in remote sensing images, the present invention proposes a satellite remote sensing image entity information missing detection method and a missing value filling method based on association rules, so as to achieve effective cleaning of the satellite remote sensing image itself and its attribute files.
[0066] First, refer to Figure 1 The data cleaning method for remote sensing images and missing attribute information of the present invention includes:
[0067] S101: Acquire a remote sensing image data set, that is, analyze the remote sensing image data to be detected to obtain the image data itself and its attribute information.
[0068] In the present invention, remote sensing image data refers to images and data of the earth's surface obtained by remote sensing technology. Remote sensing image data can be obtained by, but is not limited to, satellite remote sensing, aerial photography, drone photography, etc. In addition, remote sensing image data includes the remote sensing image itself and its supporting attribute files, and the attribute files include, for example, XML data, supporting database tables, etc.
[0069] In addition, the process of parsing the remote sensing image data to be detected can, for example, first obtain the raw data from the remote sensing platform, and then perform preprocessing to correct the radiation and geometric errors, including radiation correction to eliminate sensor response differences, atmospheric correction to reduce the impact of atmospheric conditions, and geometric correction to ensure the spatial accuracy of the data. Then, the visual effect of the image is improved by image enhancement technology to make the features of the ground objects more obvious. On this basis, multispectral and hyperspectral analysis are performed to identify the spectral characteristics of the ground objects by combining different bands and selecting features. Then, the classification algorithm is applied to classify the ground objects in the image to distinguish different surface features. At the same time, the attribute information of the image is extracted, such as metadata such as sensor type, acquisition time, and spatial resolution. In addition, data fusion may be required to integrate image data from different sources to obtain richer information. Finally, the classification results are evaluated and verified for accuracy to ensure the reliability of the parsing results, and the parsed data and its attribute information are stored in the database for further analysis and application.
[0070] S102: Remote sensing image information missing detection, that is, performing missing value checks on the remote sensing image and its attribute information respectively.
[0071] Specifically, the steps of checking missing values of remote sensing images and their attribute information include:
[0072] S1-1: Perform stripe detection on remote sensing images, including row-direction stripe detection and column-direction stripe detection. Figure 4 The left image shows the missing row image. Figure 4 The image on the right shows the missing columns.
[0073] S1-1-1: Specifically, for row-wise stripe detection, the typical pattern of missing row-wise stripe information is that there are dark stripes in the row direction of the remote sensing image, the color difference between rows is obvious, and the row mean gradient has a maximum value phenomenon. Figure 5 A line graph of the row mean is shown, where the row mean gradient shows a maximum value phenomenon.
[0074] According to the abnormal characteristics of the row, the row mean parameter of all bands of the image is :
[0075]
[0076] in, For the The mean DN of the row, the mean DN of the row, For the Row, No. The pixel DN value of the column, 、 The DN is the number of rows and columns of the image, respectively. In remote sensing image analysis, DN usually refers to the number representing the pixel value in the remote sensing data.
[0077] Calculate the mean value of each row of pixels in a single band of remote sensing images V i , draw a line graph of the row mean, and combine it with abnormal image analysis to find that the line graph position corresponding to the row with missing image information has obvious fluctuations, such as Figure 5 As shown, there is an obvious fluctuation between lines 1020 and 1026.
[0078] After calculating the mean value of each row of pixels in a single band of the remote sensing image After drawing the line graph of the row mean, a certain number of images with different degrees of information loss of this type are used as experimental samples, and the corresponding line graph is drawn. The pixels at the position of the information missing row are calculated based on the line graph. DN The value gradient is obtained, and the mean gradient in the row direction is at a fixed value (set as b ) above, obvious stripes will appear. The row mean gradient calculation formula is as follows:
[0079]
[0080] In the formula, For the Row and Mean gradient between rows.
[0081] When the row mean gradient satisfies When , the image is judged as missing row information, where Is a fixed value.
[0082] S1-1-2: The calculation of stripe detection in the column direction is similar to that in the row direction. The calculation formula of the mean value in the column direction is as follows:
[0083]
[0084] in, For the The mean DN of the column, For the Row, No. The pixel DN value of the column, 、 is the number of rows and columns of the image.
[0085] Calculate the mean value of each column of pixels in a single band of the remote sensing image , draw a line graph of the row mean, and combine it with the abnormal image analysis, it can be found that the line fluctuates significantly at the position of the line graph corresponding to the column where image information is missing.
[0086] Take a certain number of images with different degrees of information loss of this type as experimental samples, draw the corresponding line graph, and calculate the pixels at the position of the information loss column based on the line graph. DN The value gradient is obtained, and the mean gradient in the column direction is at a fixed value (set as d ) above, obvious stripes will appear. The column mean gradient calculation formula is as follows:
[0087]
[0088] When the column mean satisfies The image is judged to be missing column information.
[0089] S1-2: Detect local anomalies in remote sensing images. Local anomalies in satellite remote sensing images are mainly manifested as missing local information of remote sensing image pixels. V Affected by the missing width and pixels of remote sensing images DN The mean gradient of local information missing in the row direction of remote sensing images is:
[0090]
[0091] In the formula, width is the width of the local missing image, DN The pixels of the missing part DN value, is the image row mean, is the number of pixels in the row direction.
[0092] When the width of the row and column missing of the remote sensing image is small and the row and column of the image itself is large, the fluctuation range of the row and column mean can be almost ignored. Therefore, when the local information missing of the remote sensing image is small, the conventional row and column direction detection strategy cannot be used. It is necessary to statistically analyze the distribution of the row and column pixel histogram of the remote sensing image (see for details). Figure 6 ) found that there is a problem, which is manifested in that the row and column pixel histogram of normal remote sensing images changes evenly, while a certain pixel value of the histogram of images with local information missing will fluctuate abnormally. Therefore, for local information missing, we can set a threshold for the number of abnormal pixels to detect abnormal local information missing in remote sensing images.
[0093] In the present invention, the following formula is used for determination:
[0094]
[0095] In the formula, , The pixel values in row and column directions are continuous x The number of pixels, Length is the actual distance on the ground, where the threshold is taken N (Set according to actual situation), GSD is the image resolution, in units of m .when When the above conditions are met, the image is judged to be missing information. Generally speaking, by statistically analyzing the histogram distribution of image row / column pixels, it can be analyzed that: in an image without local outlier noise, the histogram of row / column pixels changes evenly; in an image with local outlier noise, the histogram of row / column pixels will fluctuate violently, that is, abnormal fluctuations will occur at a certain pixel value x (that is, abnormal pixel value), and the number of pixels with pixel value x appearing continuously can be obtained by statistics. , Based on this, we can set a threshold for the number of pixels corresponding to the abnormal pixel value x (the threshold here is set based on production experience and can be modified during the production process) to detect whether there are local abnormalities in the image.
[0096] S1-3: Band anomaly detection. Remote sensing image band anomalies include abnormalities such as the overall low or high pixel values of a certain band of the remote sensing image, or the unchanged pixel values. This type of remote sensing image information is missing abnormalities.
[0097] In the present invention, the following formula is used for determination:
[0098]
[0099] In the formula, For the The pixel value of a single pixel in the band, For the The pixel value of a single pixel in the band, and Represent different bands. is the maximum value of a certain band pixel value. is the minimum value of the corresponding band pixel value. When the image band meets any of the above three conditions, the image is judged to be a band anomaly, and in the present invention, the symbol " " means much less than, the symbol " " means much greater than, and the degree of size can be set according to the actual situation, such as 10 times or 100 times.
[0100] For example, a certain image data has a band anomaly, that is, the image color is abnormal, which is mainly manifested in that the DN value of a certain band is overall low or high in the color abnormal area, and its maximum value is much lower or higher than the DN mean of a certain band (or the DN value of a certain band at the same pixel position).
[0101] S103: For the missing stripe information in the row and column direction, the missing local pixel information and the missing image band anomaly in the image data, the missing image phenomenon is detected using stripe detection, local anomaly detection and band anomaly detection algorithms. There are three main processing methods for images with missing data:
[0102] ① Direct elimination: Images with missing information are usually eliminated directly if there is replacement data.
[0103] ② Data labeling. When images with missing information are automatically filtered, the problematic images are labeled, and the specific row and column missing or band abnormality positions of the image are recorded in the database, and the images are effectively labeled, which can play a role in tracking and evaluation in the data storage and management process.
[0104] ③ Data restoration. Based on data annotation, for images with missing information, we conduct correlation analysis based on data shooting time, region, resolution, etc. If there is replacement data, we can directly remove it, or use a restoration algorithm based on multi-temporal image information (see the specific process for details). Figure 2 ) Repair the missing parts of remote sensing images.
[0105] Specifically, since the missing remote sensing image data cannot be directly used for image production, the present invention adopts the following processing strategy:
[0106] S2-1: Images with missing information are usually directly eliminated if there is replacement data; the data replacement principle is to screen and replace the replacement data according to the imaging range, imaging season and time, imaging angle and resolution, etc.
[0107] S2-1-1: Imaging range screening. Remote sensing images can be screened according to the image data range.
[0108] S2-1-2: Screening by imaging angle and resolution. Remote sensing images can be screened according to the imaging angle and resolution of the actual image.
[0109] S2-1-3: Imaging season and time screening. Based on the data range screening, data can be screened according to the four seasons of spring, summer, autumn and winter or a specific time range, or remote sensing images can be screened according to a certain season or a certain time range.
[0110] S2-2: When automatically filtering images with missing information, the problematic images are labeled and the specific row and column missing or band abnormality locations of the image are recorded in the database. This will facilitate tracking and evaluation during the data storage and management process.
[0111] S2-3: Based on data annotation, for images with missing information, correlation analysis is performed in combination with data shooting time, region, resolution, etc. If there is replacement data, they can be directly eliminated, or the missing parts of the remote sensing image can be repaired through a repair algorithm based on multi-temporal image information. The specific processing flow is as follows.
[0112] S2-3-1: Obtain a multi-source remote sensing image dataset and use the satellite remote sensing image information missing detection algorithm to determine whether there is image missing in the remote sensing image.
[0113] S2-3-2: Construct a multi-source remote sensing image missing sample dataset. As the number of training model samples is too small, data enhancement operations such as rotation and flipping are performed on the remote sensing image dataset to prevent overfitting, and then the dataset is divided into training set and test set.
[0114] S2-3-3: Perform a global traversal of the missing remote sensing images, calculate the similarity measure between each phase block and the missing part of the remote sensing image, select the phase block with the highest similarity as the phase complementary information, and reduce the phase difference through linear regression transformation.
[0115] S2-3-4: Use multiple scale convolution filters to extract features from the missing parts of the remote sensing image and the maximum similarity measure phase block selected in S2-3-3, and fuse the extracted results.
[0116] S2-3-5: Construct an improved GAN remote sensing image missing repair model (the overall model structure is as follows Figure 3 ), train the model and determine the model parameter settings. The phase block with the maximum similarity measure is selected by calculating the similarity measure between the missing information and the phase block of the remote sensing image, and then the multi-scale feature fusion module is used to extract and fuse the features of the target block with missing information of the remote sensing image and the phase block with the maximum similarity measure, and the feature map of the fused phase information and spatial information is input into the improved GAN generator. The model generates reconstruction results through the GAM-DUnet generator, and the multi-scale discriminator determines the source of the reconstruction results through the activation function to promote the reconstruction results to be more globally consistent and locally consistent, thereby completing the entire remote sensing image missing information repair process.
[0117] S2-3-6: Use the improved GAN remote sensing image missing repair model to repair the missing parts of the remote sensing image.
[0118] S104: Finally, for the missing values in the database table, the remote sensing image data attribute missing verification rules are established by using association rules and other methods, and the missing attribute data is found. By mining the metadata XML attribute information, the relevant attribute information is repaired and backfilled into the missing attribute field or the relevant repository table.
[0119] Specifically, the attribute data of remote sensing images or database tables may be missing key attributes, resulting in low query efficiency for scene imaging data, low image utilization, and inability to issue tasks. If these missing data records are deleted, the accuracy of data analysis will be affected and the use value of the data will be reduced. Therefore, the better way to clean missing data is to use existing technologies or means to fill in the incomplete data records. The specific processing flow is as follows: Figure 3 shown.
[0120] S3-1: Missing data detection. Check the attribute records corresponding to the remote sensing image to see if the attribute values that are not allowed to be empty are empty, "Null" or "N / A".
[0121] S3-2: Establish missing identification rules between metadata XML and attribute database tables, and check the content in the database to determine whether the data is missing.
[0122] S3-3: Missing value filling. The idea of filling missing values in remote sensing big data is mainly to mine metadata XML information and then fill in the missing data. The specific steps are as follows:
[0123] S3-3-1: Metadata information extraction. Metadata information extraction mainly includes database table metadata information extraction and metadata XML file information extraction. Among them, the metadata information extraction of the database table is automatically collected by the tool module; the metadata XML file needs to obtain the element attribute information in the metadata XML file according to the metadata standard structure through a fixed program.
[0124] S3-3-2: Metadata attribute information is stored in the database. The extracted metadata attribute information is imported into the data warehouse table to facilitate the verification of metadata information.
[0125] S3-3-3: Metadata content verification. Formulate quality verification rules for metadata attribute information after it is entered into the table, including completeness, consistency, etc., and generate a quality report.
[0126] S3-3-4: Use the general modeling and governance tools of the big data application platform to develop image data attribute missing verification rule components;
[0127] S3-3-5: Establish verification and identification missing rules through association rules and other algorithms to detect missing data;
[0128] S3-3-6: Repair and fill in missing data fields by associating metadata XML attribute information.
[0129] The data cleaning method proposed in the present invention for the missing remote sensing images and their attribute information uses strip detection, local anomaly detection and band anomaly detection algorithms to detect the missing phenomenon of images, and realizes the automatic processing of missing image information through methods such as data elimination, data annotation and data repair; the attribute information missing value filling method based on association rules solves the business problems such as low query efficiency of scene imaging data, low image utilization rate and inability to issue tasks caused by the missing key fields of image attributes. Through the research of this method, it is possible to effectively solve the business problems such as unnecessary production, processing and access anomalies of remote sensing images caused by the missing images and attribute information in the actual business production process. At the same time, this technology can also be used in large-scale remote sensing image business production processing, cleaning management, data service and other systems to support large-scale image cleaning processing or image quality inspection.
[0130] In addition, the present invention also provides a data cleaning device for missing remote sensing images and attribute information, comprising:
[0131] A remote sensing image data set acquisition device, used to acquire a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data;
[0132] A missing check device is used to perform missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data strip detection, remote sensing image data local anomaly detection and band anomaly detection; and
[0133] A processing device is used to process missing remote sensing image data and attribute information.
[0134] The functions of the various components of the data cleaning device for remote sensing images and attribute information missing have been referenced above. Figure 1-Figure 6 A detailed description has been given and will not be repeated here.
[0135] The above is only a specific implementation of the present disclosure, but the protection scope of the present disclosure is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present disclosure should be included in the protection scope of the present disclosure. Therefore, the protection scope of the present disclosure should be based on the protection scope of the claims.
Claims
1. A data cleaning method for remote sensing images and missing attribute information, characterized in that: include: Acquire a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data; Performing a missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data strip detection, remote sensing image data local anomaly detection, and band anomaly detection; and Process the missing remote sensing image data and attribute information; The local anomaly detection of remote sensing image data includes judging whether there is a local anomaly in the remote sensing image data through the following formula: ; In the formula, The pixel values in row and column directions are continuous x The number of pixels, Length is the actual distance on the ground, where the threshold is taken N , GSD is the image resolution, in units of m , and when the above conditions are met, it is determined that there is a local anomaly in the remote sensing image data; When one of the following equations is met, it is judged that there is a band anomaly: ; In the formula, For the The pixel value of a single pixel in the band, For the The pixel value of a single pixel in the band, and Represent different bands. is the maximum value of a certain band pixel value. is the minimum value of the corresponding band pixel value; The method further includes performing missing detection on the attribute information, including: Check the attribute information corresponding to the remote sensing image data to determine whether the attribute value that is not allowed to be empty is empty; Establish missing identification rules between metadata XML and attribute database tables, and check the content in the attribute database to determine whether data is missing; and Metadata XML information is mined to fill in missing data in the attribute database.
2. The data cleaning method for remote sensing images and missing attribute information according to claim 1 is characterized in that: Remote sensing image data stripe detection includes row direction stripe detection and column direction stripe detection. The row direction stripe detection includes: Get the row mean parameters of all bands; Calculate the row mean gradient between two adjacent rows; and When the row mean gradient is greater than a first threshold, determining that the remote sensing image data is missing row information; The column direction strip detection includes: Get the column mean parameters of all bands; Calculate the column mean gradient between two adjacent columns; and When the column mean gradient is greater than a second threshold, it is determined that the remote sensing image data is missing column information.
3. The data cleaning method for remote sensing images and missing attribute information according to claim 1 is characterized in that: Processing of missing remote sensing image data includes: When there is replacement data for the missing remote sensing image data, the missing remote sensing image data is directly eliminated.
4. The data cleaning method for remote sensing images and missing attribute information according to claim 1 is characterized in that: Processing the missing remote sensing image data includes repairing the missing parts of the remote sensing image through a repair algorithm based on multi-temporal image information, and includes: Obtain multi-source remote sensing image data sets and determine whether there are any missing images in the remote sensing image data; Construct a missing sample dataset of multi-source remote sensing image data; Perform a global traversal of the missing remote sensing image data, calculate the similarity measure between each phase block and the missing part of the remote sensing image, select the phase block with the highest similarity as the phase complementary information, and reduce the phase difference through linear regression transformation; The missing parts of the remote sensing image and the selected time blocks with the highest similarity are subjected to feature extraction using multiple scale convolution filters, and the extracted results are fused; Build an improved GAN remote sensing image missing repair model, train the model, and determine the model parameter settings; and The improved GAN remote sensing image missing repair model is used to repair the missing parts of the remote sensing image.
5. The data cleaning method for remote sensing images and missing attribute information according to claim 1 is characterized in that: Missing remote sensing image data are also labeled to record the missing row and column locations or band anomaly locations in the database.
6. The data cleaning method for remote sensing images and missing attribute information according to claim 5 is characterized in that: The mining of metadata XML information to fill in missing data in the attribute database includes: Metadata information extraction, including database table metadata information extraction and metadata XML file information extraction; Metadata attribute information is stored in the database, and the extracted metadata attribute information is imported into the data warehouse table; Metadata content verification: formulate quality verification rules for metadata attribute information after entering the table; Use the general modeling and governance tools of the big data application platform to develop image data attribute missing verification rule components; Establish verification and identification missing rules through association rule algorithms to detect missing data; and By associating metadata XML attribute information, repair and backfill missing data fields.
7. A data cleaning device for remote sensing images and missing attribute information, characterized in that: include: A remote sensing image data set acquisition device, used to acquire a remote sensing image data set, wherein the remote sensing image data set includes remote sensing image data and attribute information corresponding to the remote sensing image data; A missing check device is used to perform missing check on the remote sensing image data set, wherein the missing check on the remote sensing image data includes remote sensing image data strip detection, remote sensing image data local anomaly detection and band anomaly detection; and A processing device for processing missing remote sensing image data and attribute information; The remote sensing image data local anomaly detection includes determining whether there is a local anomaly in the remote sensing image data by the following formula: ; In the formula, The pixel values in row and column directions are continuous x The number of pixels, Length is the actual distance on the ground, where the threshold is taken N , GSD is the image resolution, in units of m , and when the above conditions are met, it is determined that there is a local anomaly in the remote sensing image data; When one of the following equations is met, it is judged that there is a band anomaly: ; In the formula, For the The pixel value of a single pixel in the band, For the The pixel value of a single pixel in the band, and Represent different bands. is the maximum value of a certain band pixel value. is the minimum value of the corresponding band pixel value; Performing missing detection on the attribute information includes: Check the attribute information corresponding to the remote sensing image data to determine whether the attribute value that is not allowed to be empty is empty; Establish missing identification rules between metadata XML and attribute database tables, and check the content in the attribute database to determine whether data is missing; and Metadata XML information is mined to fill in missing data in the attribute database.
