Sensor data correction method and device, electronic equipment and storage medium

Abstract

Embodiments of the invention provide a sensor data correction method and apparatus, an electronic device and a storage medium. The method comprises the steps of obtaining to-be-corrected first sensor data; the first sensor data are converted according to a preset conversion model, second sensor data corresponding to the first sensor data are obtained, and the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to a calculation result; according to the embodiment of the invention, a large amount of different types of sensor sample data are calculated, the regression equation is fitted according to the calculation result to obtain the conversion model, the regression equation at least comprises the least square regression equation or the axial regression equation, the first sensor data is converted through the conversion model, and the first sensor data is converted into the second sensor data. According to the method, the influence of data noise is removed in the conversion process, and the quality of the converted data is further improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method, device, electronic device and storage medium for correcting sensor data. Background Art

[0002] Vegetation is an important component of terrestrial ecosystems and a sensitive indicator of the impact of climate and human activities on the environment. The long-term Normalized Difference Vegetation Index (NDVI) remote sensing product is a basic data set for analyzing global climate and vegetation changes. At present, the global long-term NDVI products are basically obtained by AVHRR and MODIS sensors carried on polar-orbiting satellites. The domestic commercial polar-orbiting satellite FY-3D is equipped with the MERSI-Ⅱ sensor (similar to MODIS). Affected by many factors such as sensor spectral response, transit time, product algorithm, etc., there are obvious inconsistencies between domestic and foreign NDVI products. Directly merging them together cannot accurately capture climate change signals (that is, the changes are not caused by climate but by different sensors). Only after the deviations of the two products are removed can they be merged together for climate change analysis.

[0003] Therefore, it is necessary to make consistency corrections for NDVI products of different types of sensors. For example, a linear regression method can be used. Specifically, several pixel points are selected from the data of the overlapping observation period of two sensors, and the conversion relationship of the pixel points based on different sensors is established. In this way, the number of samples is small and easily affected by noise. The regression equation fitted in this way is inaccurate, and the conversion efficiency of different types of sensor data is poor. How to improve the conversion efficiency of different types of sensor data is an urgent problem that needs to be solved. Summary of the invention

[0004] Some embodiments of the present application aim to provide a method, device, electronic device and storage medium for correcting sensor data. Through the technical solutions of the embodiments of the present application, first sensor data to be corrected is obtained; the first sensor data is converted according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a final regression equation. In the embodiments of the present application, a large number of different types of sensor sample data are calculated and the regression equation is fitted according to the calculated results to obtain a conversion model, and the regression equation at least includes a least squares regression equation or a final regression equation. The first sensor data is converted by the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is removed during the conversion process and further improving the quality of the converted data.

[0005] In a first aspect, some embodiments of the present application provide a method for correcting sensor data, including:

[0006] Acquiring first sensor data to be corrected;

[0007] The first sensor data is converted according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a finale regression equation.

[0008] Some embodiments of the present application calculate a large number of different types of sensor sample data, and fit a regression equation according to the calculated results to obtain a conversion model. The regression equation includes at least a least squares regression equation or a finale regression equation. The first sensor data is converted through the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is eliminated during the conversion process, thereby further improving the quality of the converted data.

[0009] Optionally, the preset conversion model is obtained in the following manner:

[0010] Acquire first sensor sample data and second sensor sample data;

[0011] Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data;

[0012] Performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data;

[0013] calculating a mean and a standard deviation of the difference between the fifth sensor sample data and the sixth sensor sample data;

[0014] The sensor sample data are judged according to the mean and the standard deviation to obtain seventh sensor sample data and eighth sensor sample data that meet preset conditions;

[0015] Calculating a correlation coefficient between the seventh sensor sample data and the eighth sensor sample data to obtain a correlation coefficient calculation result;

[0016] According to the correlation coefficient calculation result, a regression equation is fitted to obtain the preset conversion model.

[0017] Some embodiments of the present application collect data within a preset window range of the overlapping period of different types of original sample data, and calculate the correction coefficients pixel by pixel and month by month in order to increase the sample size, reduce noise interference, increase the spatial correlation of the regression coefficients, and improve the accuracy of model training.

[0018] Optionally, performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data, includes:

[0019] Calculating a first change amplitude of the first sensor sample data;

[0020] Calculating a second variation amplitude of the second sensor sample data;

[0021] Determining a target amplitude value according to the first change amplitude and the second change amplitude;

[0022] When the target amplitude value satisfies a preset condition, the first sensor sample data corresponding to the target amplitude value is determined as the third sensor sample data, and the second sensor sample data corresponding to the target amplitude value is determined as the fourth sensor sample data.

[0023] Some embodiments of the present application perform amplitude value judgment on sensor sample data and screen the sensor sample data to improve the accuracy of sensor sample data processing.

[0024] Optionally, performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data, comprises:

[0025] If the third sensor sample data conforms to the normal distribution rule, determining the third sensor sample data as the fifth sensor sample data;

[0026] If the fourth sensor sample data conforms to the normal distribution rule, the fourth sensor sample data is determined as the sixth sensor sample data.

[0027] In some embodiments of the present application, sensor sample data that do not meet the requirements are deleted by performing normal distribution judgment on the sensor sample data, thereby avoiding the problem of different underlying surfaces being mixed in the sample data, and making the established regression equation more targeted.

[0028] Optionally, performing regression equation fitting according to the correlation coefficient calculation result to obtain the preset conversion model includes:

[0029] If the correlation coefficient calculation result is a significant positive correlation, then based on the first sensor sample data and the second sensor sample data, determine whether the correlation is improved after adding the zero-crossing sample data. If it is improved, determine to add the zero-crossing sample data, and perform least square regression equation fitting on the first sensor sample data, the second sensor sample data, and the zero-crossing sample data; if it is not improved, do not add the zero-crossing point, and perform least square regression equation fitting on the first sensor sample data and the second sensor sample data. The slope and intercept are obtained by fitting, and the mean and standard deviation of the first sensor sample data are recorded (the statistics of the mean and standard deviation do not consider the zero-crossing point).

[0030] Some embodiments of the present application can reduce the absolute value of the final intercept of the least squares linear regression by adding zero-crossing sample data, because the fitting error of the zero-crossing point is included in the total error; on the other hand, the slope of the regression is slightly larger than before, because the correlation coefficient between the independent variable and the dependent variable increases, and the slope of the regression is slightly larger than before, which helps to maintain the variability of the independent variable data. Optionally, the regression equation is fitted according to the correlation coefficient calculation result to obtain the preset conversion model, including:

[0031] If the correlation coefficient calculation result is a positive correlation but not significant, a regression equation is fitted using a final regression algorithm based on the first sensor sample data and the second sensor sample data to obtain a slope and an intercept, and the mean and standard deviation of the first sensor sample data are recorded.

[0032] Some embodiments of the present application use a final regression algorithm to fit the first sensor sample data and the second sensor sample data to obtain a larger slope and a smaller intercept, which helps to maintain the variability of the data.

[0033] Optionally, performing regression equation fitting according to the correlation coefficient calculation result to obtain the preset conversion model includes:

[0034] If the calculated result of the correlation coefficient is a negative value, the slope is directly set to 1, the intercept is the mean of the second sensor sample data minus the mean of the first sensor sample data, and the mean and standard deviation of the first sensor sample data are recorded.

[0035] Some embodiments of the present application use a conversion method with a slope of 1 and an intercept of the mean difference, which helps to maintain data variability and change direction and eliminate statistical deviations between the first sensor sample and the second sensor sample. Optionally, the first sensor sample data and the second sensor sample data are obtained by extracting data within a preset window range of an overlapping period of different types of original sample data.

[0036] Some embodiments of the present application set a window range for overlapping period data, extract data based on the window range, expand the data in time and space, and based on the law of geographic autocorrelation (i.e., the principle of proximity in geography), adjacent time and space should have more similar conversion methods.

[0037] In a second aspect, some embodiments of the present application provide a sensor data correction device, including:

[0038] An acquisition module, used for acquiring first sensor data to be corrected;

[0039] A correction module is used to convert the first sensor data according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a finale regression equation.

[0040] Some embodiments of the present application calculate a large number of different types of sensor sample data, and fit a regression equation according to the calculated results to obtain a conversion model. The regression equation includes at least a least squares regression equation or a finale regression equation. The first sensor data is converted through the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is eliminated during the conversion process, thereby further improving the quality of the converted data.

[0041] Optionally, the device further comprises a setting module, wherein the setting module is configured to:

[0042] Acquire first sensor sample data and second sensor sample data;

[0043] Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data;

[0044] Performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data;

[0045] calculating a mean and a standard deviation of the difference between the fifth sensor sample data and the sixth sensor sample data;

[0046] The sensor sample data are judged according to the mean and the standard deviation to obtain seventh sensor sample data and eighth sensor sample data that meet preset conditions;

[0047] Calculating a correlation coefficient between the seventh sensor sample data and the eighth sensor sample data to obtain a correlation coefficient calculation result;

[0048] According to the correlation coefficient calculation result, a regression equation is fitted to obtain the preset conversion model.

[0049] Some embodiments of the present application collect data within a preset window range of the overlapping period of different types of original sample data, and calculate the correction coefficients pixel by pixel and month by month in order to increase the sample size, reduce noise interference, increase the spatial correlation of the regression coefficients, and improve the accuracy of model training.

[0050] Optionally, the setting module is used to:

[0051] Calculating a first change amplitude of the first sensor sample data;

[0052] Calculating a second variation amplitude of the second sensor sample data;

[0053] Determining a target amplitude value according to the first change amplitude and the second change amplitude;

[0054] When the target amplitude value satisfies a preset condition, the first sensor sample data corresponding to the target amplitude value is determined as the third sensor sample data, and the second sensor sample data corresponding to the target amplitude value is determined as the fourth sensor sample data.

[0055] Some embodiments of the present application perform amplitude value judgment on sensor sample data and screen the sensor sample data to improve the accuracy of sensor sample data processing.

[0056] Optionally, the setting module is used to:

[0057] If the third sensor sample data conforms to the normal distribution rule, determining the third sensor sample data as the fifth sensor sample data;

[0058] If the fourth sensor sample data conforms to the normal distribution rule, the fourth sensor sample data is determined as the sixth sensor sample data.

[0059] In some embodiments of the present application, sensor sample data that do not meet the requirements are deleted by performing normal distribution judgment on the sensor sample data, thereby avoiding the problem of different underlying surfaces being mixed in the sample data, and making the established regression equation more targeted.

[0060] Optionally, the setting module is used to:

[0061] If the correlation coefficient calculation result is a significant positive correlation, then based on the first sensor sample data and the second sensor sample data, determine whether the correlation is improved after adding the zero-crossing point sample data. If it is improved, determine to add the zero-crossing point, and perform least square regression equation fitting on the first sensor sample data, the second sensor sample data, and the zero-crossing point sample data; if it is not improved, do not add the zero-crossing point, and perform least square regression equation fitting on the first sensor sample data and the second sensor sample data. The slope and intercept are obtained by fitting, and the mean and standard deviation of the first sensor sample data are recorded (the statistics of the mean and standard deviation do not consider the zero-crossing point).

[0062] Some embodiments of the present application can reduce the absolute value of the final intercept of the least squares linear regression by adding zero-crossing sample data, because the fitting error of the zero-crossing point is included in the total error; on the other hand, the slope of the regression will be slightly larger than before, because the correlation coefficient between the independent variable and the dependent variable increases, and a slight increase in the slope is more helpful in maintaining the variability of the independent variable data.

[0063] Optionally, the setting module is used to:

[0064] If the correlation coefficient calculation result is an insignificant positive correlation, a regression equation is fitted using a final regression algorithm based on the first sensor sample data and the second sensor sample data to obtain the slope, intercept, mean and standard deviation of the first sensor sample data.

[0065] Some embodiments of the present application use a final regression algorithm to fit the first sensor sample data and the second sensor sample data to obtain a larger slope and a smaller intercept, which helps to maintain the variability of the data.

[0066] Optionally, the setting module is used to:

[0067] If the calculated result of the correlation coefficient is a negative value, the slope is set to 1, the intercept is the mean of the second sensor sample data minus the mean of the first sensor sample data, and the mean and standard deviation of the first sensor sample data are recorded.

[0068] Some embodiments of the present application use a conversion method with a slope of 1 and an intercept of the mean difference, which helps to maintain data variability and change direction and eliminate statistical deviations between the first sensor sample and the second sensor sample.

[0069] Optionally, the first sensor sample data and the second sensor sample data are obtained by extracting data within a preset window range of an overlapping period of original sample data of different types.

[0070] Some embodiments of the present application set a window range for overlapping period data, extract data based on the window range, expand the data in time and space, and based on the law of geographic autocorrelation (i.e., the principle of proximity in geography), adjacent time and space should have more similar conversion methods.

[0071] In a third aspect, some embodiments of the present application provide an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for correcting sensor data as described in any embodiment of the first aspect can be implemented.

[0072] In a fourth aspect, some embodiments of the present application provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, can implement the sensor data correction method as described in any embodiment of the first aspect.

[0073] In a fifth aspect, some embodiments of the present application provide a computer program product, comprising a computer program, wherein the computer program, when executed by a processor, can implement the sensor data correction method as described in any embodiment of the first aspect. BRIEF DESCRIPTION OF THE DRAWINGS

[0074] In order to more clearly illustrate the technical solutions of some embodiments of the present application, the drawings required for use in some embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0075] Figure 1 A schematic diagram of a flow chart of a sensor data correction method provided in an embodiment of the present application;

[0076] Figure 2 A schematic diagram of the structure of a sensor data correction device provided in an embodiment of the present application;

[0077] Figure 3 A schematic diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0078] The technical solutions in some embodiments of the present application will be described below in conjunction with the drawings in some embodiments of the present application.

[0079] It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in subsequent drawings. At the same time, in the description of this application, the terms "first", "second", etc. are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0080] Vegetation is an important component of terrestrial ecosystems and a sensitive indicator of the impact of climate and human activities on the environment. The long-term Normalized Difference Vegetation Index (NDVI) remote sensing product is a basic data set for analyzing global climate and vegetation changes. At present, the global long-term NDVI products are basically obtained by AVHRR and MODIS sensors on polar-orbiting satellites. The AVHRR / NDVI series began in 1981, and the MODIS / NDVI series began in 2000, and both have accumulated more than 20 years of data. The domestic operational polar-orbiting satellite FY3D is equipped with the MERSI-Ⅱ sensor (similar to MODIS). MERSI-Ⅱ began to produce NDVI products in 2019, and currently has accumulated less than 5 years of data. The short data time series seriously limits the application of domestic NDVI products in climate change analysis. If we wait for domestic data to accumulate until it meets the needs of climate change analysis, it will have a high time cost. The only shortcut is to combine short-term domestic data with long-term foreign data. However, due to the influence of many factors such as sensor spectral response, transit time, product algorithm, etc., there are obvious inconsistencies between domestic NDVI and foreign NDVI products. Directly merging them together cannot accurately capture the climate change signal (that is, the change is not affected by climate but by sensors). Only after removing the deviation of the two products can they be merged together for climate change analysis.

[0081] Therefore, it is necessary to make consistency corrections for NDVI products of different types of sensors. For example, a linear regression method can be used. Specifically, several pixel points are selected from the data of the overlapping observation period of two sensors, and a conversion relationship of the pixel points based on different sensors is established. In this way, the number of samples is small and easily affected by noise. The regression equation fitted in this way is accurate. The conversion efficiency of different types of sensor data is poor. How to improve the conversion efficiency of different types of sensor data is an urgent problem to be solved. In view of this, some embodiments of the present application provide a sensor data correction method, which includes obtaining first sensor data to be corrected; correcting the first sensor data according to a preset conversion model; and correcting the first sensor data according to a preset conversion model. The data is converted to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a final regression equation. In the embodiment of the present application, a large number of different types of sensor sample data are calculated, and the regression equation is fitted according to the calculated results to obtain a conversion model, and the regression equation at least includes a least squares regression equation or a final regression equation. The first sensor data is converted by the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is removed during the conversion process, and further improving the quality of the converted data.

[0082] like Figure 1 As shown, an embodiment of the present application provides a method for correcting sensor data, the method comprising:

[0083] S101, obtaining first sensor data to be corrected;

[0084] Specifically, the terminal device acquires first sensor data, which is NDVI (Normalized Difference Vegetation Index) data with a short time series.

[0085] S102. Convert the first sensor data according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a final regression equation.

[0086] Specifically, a large amount of sensor sample data is pre-acquired on the terminal device, and training samples are extracted from the data within a preset window range of the overlapping period of different types of sensor sample data. The least squares regression equation or the final regression equation is fit according to the training samples to obtain a conversion model. The first sensor data is converted by the conversion model to obtain second sensor data corresponding to the first sensor data, for example, converting short-time series NDVI data into long-time series NDVI data.

[0087] Because the establishment of the regression equation is based on the generation of finite samples, the application of the regression equation is guaranteed within the range of variation defined by the sample. If it is out of this range, the uncertainty of the regression equation will increase significantly. When the regression equation is actually used for conversion, the embodiment of the present application will first compare the data to be converted with the mean and standard deviation of the pixel for that month. If the data to be converted is more than 3 standard deviations smaller than the mean, or more than 8 standard deviations higher than the mean, the data to be converted is very likely to be contaminated by noise, and it is not appropriate to use the conversion equation for operation. This data should be replaced with the mean of the month and then converted. Here, it should be explained that since most of the noise (such as cloud and cloud shadow pollution, snow pollution, etc.) will reduce NDVI, the judgment rules are stricter when the NDVI is small, and the judgment rules are looser when the NDVI is large.

[0088] When fitting the regression, the difference characteristics of the two data are used to determine the credibility of the sample, and unreliable samples are deleted. However, in practice, when using one sensor NDVI product to generate another sensor NDVI product, it is impossible to remove the noise data by calculating the difference. Therefore, the variability characteristics of the independent variable data when the statistical conversion relationship is established in the embodiment of the present application are used as the condition for data quality control to ensure that the influence of data noise is removed during the conversion process, thereby further improving the quality of the converted data. This is also very critical for domestic data that still has a lot of noise and incomplete quality control code information.

[0089] Some embodiments of the present application calculate a large number of different types of sensor sample data, and fit a regression equation according to the calculated results to obtain a conversion model. The regression equation includes at least a least squares regression equation or a finale regression equation. The first sensor data is converted through the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is eliminated during the conversion process, thereby further improving the quality of the converted data.

[0090] Another embodiment of the present application further supplements the method for correcting sensor data provided in the above embodiment.

[0091] Optionally, the preset conversion model is obtained by:

[0092] Acquire first sensor sample data and second sensor sample data;

[0093] Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data;

[0094] Performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data;

[0095] calculating a mean and a standard deviation of the difference between the fifth sensor sample data and the sixth sensor sample data;

[0096] The sensor sample data are judged according to the mean and the standard deviation to obtain the seventh sensor sample data and the eighth sensor sample data that meet the preset conditions;

[0097] Calculating the correlation coefficient of the seventh sensor sample data and the eighth sensor sample data to obtain a correlation coefficient calculation result;

[0098] According to the calculation results of the correlation coefficient, the regression equation is fitted to obtain the preset conversion model.

[0099] Some embodiments of the present application collect data within a preset window range of the overlapping period of different types of original sample data, and calculate the correction coefficients pixel by pixel and month by month in order to increase the sample size, reduce noise interference, increase the spatial correlation of the regression coefficients, and improve the accuracy of model training.

[0100] Optionally, performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data, includes:

[0101] Calculating a first change amplitude of first sensor sample data;

[0102] calculating a second variation amplitude of the second sensor sample data;

[0103] Determining a target amplitude value according to the first change amplitude and the second change amplitude;

[0104] When the target amplitude value satisfies a preset condition, the first sensor sample data corresponding to the target amplitude value is determined as the third sensor sample data, and the second sensor sample data corresponding to the target amplitude value is determined as the fourth sensor sample data.

[0105] Specifically, in the embodiment of the present application, a pixel-by-pixel window data extraction method for an overlapping period (such as two years) is adopted, and preliminary denoising is performed to extract pixel-by-pixel samples to be used to establish the pixel regression equation;

[0106] Step (1) For each pixel, extract the data of all overlapping periods within the 3*3 window range; taking the monthly product of 2 years as an example, the number of samples is 9 pixels * 12 months * 2 years = 216. Since MERSI (short, i.e., the first sensor data) and MODIS (long, i.e., the second sensor data) are relatively similar sensors, the difference between the NDVIs will generally not exceed 0.4 under the condition of no interference from clouds or other factors. If it exceeds, it can be considered to be caused by random factors outside the sensor and cannot be used to establish the regression equation.

[0107] Considering that 0.4 is a relatively high threshold for deserts, bare soil or sparsely vegetated areas (small NDVI values), the change amplitudes of the first sensor data and the second sensor data are calculated for the sample data extracted earlier, and the larger amplitude value (denoted as Amp) is retained. When the difference between the two NDVI products in a sample is greater than min[Amp, 0.4] (Note: min[Amp, 0.4] means the smaller value between Amp and 0.4), the sample is considered unsuitable for establishing a regression equation and will be deleted. Samples that meet the conditions will enter the normal distribution test.

[0108] By expanding the time, space, and feature space of data, the embodiment of the present application realizes the fitting of a more robust regression equation. The spatiotemporal expansion is mainly based on the law of geographic autocorrelation (i.e., the principle of close similarity in geography). Adjacent spatiotemporal should have more similar conversion methods. If the spatiotemporal expansion is not performed, the sample size is too small, which easily increases the randomness of the regression equation, thereby leading to larger spatiotemporal changes. When expanding in time, the embodiment of the present application pays special attention to the normal distribution test of the data to avoid the problem of mixing different underlying surfaces into the sample data, so that the regression equation established is more targeted.

[0109] Some embodiments of the present application perform amplitude value judgment on sensor sample data and screen the sensor sample data to improve the accuracy of sensor sample data processing.

[0110] Optionally, performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data, including:

[0111] If the third sensor sample data conforms to the normal distribution rule, the third sensor sample data is determined as the fifth sensor sample data;

[0112] If the fourth sensor sample data conforms to the normal distribution rule, the fourth sensor sample data is determined as the sixth sensor sample data.

[0113] Specifically, the normal distribution test was performed before the pixel-by-pixel and month-by-month regression equation was fitted.

[0114] Step (2) Normal distribution test before fitting the regression equation pixel by pixel and month by month. For each month, extract the data of the three months before and after as the backup data for establishing the regression equation. First, carry out the normal distribution test of the independent variable. If the data of the three months before and after meet the normal distribution after the test, proceed to step (3); if not, remove the data of the adjacent months and only keep the data of the current month for normal distribution test. If it meets the normal distribution, proceed to step (3); if it does not meet the normal distribution after the test, proceed to step (6). The normal test is carried out to avoid the problem of underlying surface mixing in the sample, because the underlying surface may change significantly during the seasonal transition period (such as the northern high latitude area is snow in April and vegetation in May), and the NDVI of adjacent months is significantly different, resulting in the data showing non-normal distribution. The NDVI of different underlying surfaces has different conversion relationships. At this time, if a regression equation is still established, the effect is not good. In some embodiments of the present application, sensor sample data that do not meet the requirements are deleted by performing normal distribution judgment on the sensor sample data, thereby avoiding the problem of different underlying surfaces being mixed in the sample data, and making the established regression equation more targeted.

[0115] Optionally, a regression equation is fitted according to the correlation coefficient calculation result to obtain a preset conversion model, including:

[0116] If the calculation result of the correlation coefficient is a significant positive correlation, then on the basis of the first sensor sample data and the second sensor sample data, determine whether the correlation coefficient is improved after adding the zero-crossing sample. If it is improved, the correlation can be improved, and the least squares regression equation is fitted for the first sensor sample data, the second sensor sample data and the zero-crossing sample data. If there is no improvement, the least squares regression equation is fitted for the first sensor sample data and the second sensor sample data.

[0117] Specifically, in the embodiment of the present application, after judging the normal distribution of the sensor sample data, denoising is performed before fitting the regression equation pixel by pixel and month by month. For the data that pass the normal test, the mean and standard deviation of the difference between the seventh sensor sample data and the eighth sensor sample data are calculated, that is, the difference between the seventh sensor sample data and the eighth sensor sample data is calculated, and the mean and standard deviation of the difference are calculated.

[0118] Step (3) De-noise before fitting the regression equation pixel by pixel and month by month. For the data that pass the normal test, calculate the mean and standard deviation of the difference between the two products. When the difference of a sample exceeds 1.5 standard deviations from the mean, remove the sample data. If there are still more than 5 (inclusive) samples after the operation is completed, conduct a correlation analysis. If there is a significant positive correlation between the two products (5% significance), proceed to step (4); if there is a positive correlation but not significant (5% significance) between the two products, proceed to step (5); if the correlation coefficient between the two products is negative, proceed to step (6).

[0119] Step (4) Fitting the least squares regression equation pixel by pixel and month by month. Take the NDVI of MERSI as the independent variable and the NDVI of MODIS as the dependent variable. At the same time, determine whether adding a (0, 0) point will strengthen the correlation between the independent variable and the dependent variable. If the correlation can be increased, add a (0, 0) point on the basis of the original sample and calculate the slope and intercept of the least squares regression. If the correlation cannot be increased, calculate the slope and intercept of the least squares regression based on the original sample. Save the slope and intercept results of the pixel for that month, and record the mean and standard deviation of the independent variable (excluding the added 0 point) as the quality control file for that month and the pixel. The reason for adding a (0, 0) point is as follows: the data used to fit the equation comes from observations in a local area and a specific time, and the variability of the data is small, that is, the samples are easily concentrated in a certain area in the feature space, which is not conducive to the establishment of the regression equation, or the regression equation will be greatly different from the results based on the global sample. On the basis of the original sample, after adding the (0,0) point, on the one hand, the absolute value of the final intercept of the least squares linear regression can be reduced, because when x=0, the fitting value (i.e., the intercept) should be as close to 0 as possible to obtain a smaller error. On the other hand, because the correlation coefficient between the independent variable and the dependent variable increases, the slope of the regression will be slightly larger than before, which helps to maintain the variability of the independent variable data. In addition, adding the (0,0) point will increase the similarity of the regression equations of adjacent spatiotemporal regions and avoid the excessive spatiotemporal changes of the regression coefficients calculated pixel by pixel and month by month. The mean and standard deviation of the statistical independent variable (the (0,0) point is not considered at this time) are output together with the regression coefficient results. The intercept and slope are fine-tuned to achieve the global optimum.

[0120] Among them, the number of (0, 0) points is not necessarily limited to 1. The optimal number of zero-crossing points can be determined by reserving test samples. The more zero-crossing points there are, the closer the intercept is to 0, and the higher the estimation error of non-zero-crossing points. The number of zero-crossing points is not specifically limited in this application.

[0121] Some embodiments of the present application can reduce the absolute value of the final intercept of the least squares linear regression by adding zero-crossing sample data, because the fitting error of the zero-crossing point is included in the total error; on the other hand, the slope of the regression will be slightly larger than before, because the correlation coefficient between the independent variable and the dependent variable increases, and a slight increase in the slope is more helpful in maintaining the variability of the independent variable data.

[0122] Step (5) Fit the reduced major axis equation pixel by pixel and month by month. Take the NDVI of MERSI as the independent variable and the NDVI of MODIS as the dependent variable, and use the reduced major axis to obtain the slope and intercept. Save the slope and intercept results of the pixel for that month, and record the mean and standard deviation of the independent variable at the same time, and write it as the quality control file for that month and pixel. When the correlation is small, the reduced major axis regression method can obtain a larger slope and a smaller intercept than the least squares regression, which helps to maintain the variability of the data.

[0123] Optionally, a regression equation is fitted according to the correlation coefficient calculation result to obtain a preset conversion model, including:

[0124] If the correlation coefficient calculation result is an insignificant positive correlation, the regression equation is fitted using the final regression algorithm based on the first sensor sample data and the second sensor sample data to obtain the slope and intercept, and the mean and standard deviation of the first sensor sample data are recorded.

[0125] Specifically, the pixel-by-pixel and month-by-month finale regression equation fitting in the embodiment of the present application includes taking the NDVI (short time series sensor) of MERSI as the independent variable and the NDVI (long time series sensor) of MODIS as the dependent variable, using the finale regression algorithm (reduced major axis) to obtain the slope and intercept, and recording the mean and standard deviation of the independent variable at the same time, and outputting them together with the regression coefficient (slope and intercept). Finale regression can obtain a larger slope and a smaller intercept than least squares regression, which helps to maintain the variability of the data (least squares regression tends to give an estimated result similar to the mean of the dependent variable when the correlation between the independent variable and the dependent variable is weak, resulting in a significant decrease in variability).

[0126] Step (6) Pixel-by-pixel and month-by-month deviation correction. Set the slope to 1, calculate the average deviation as the intercept, save the slope and intercept results for the pixel for that month, and record the mean and standard deviation of the independent variable at the same time, and write it as the quality control file for that month and the pixel;

[0127] If the correlation coefficient calculation result is a negative value, no regression calculation is performed, and the slope is directly set to 1, and the intercept is the mean of the second sensor sample data minus the mean of the first sensor sample data. That is, if the correlation coefficient calculation result is a negative value, the slope is set to 1, the intercept is the mean of the second sensor sample data minus the mean of the first sensor sample data, and the mean and standard deviation of the first sensor sample data are recorded.

[0128] Some embodiments of the present application use a conversion method with a slope of 1 and an intercept of the mean difference, which helps to maintain data variability and direction of change, and eliminates the statistical deviation between the first sensor sample and the second sensor sample. Adding 1 (0, 0) in the regression can improve the similarity of the data space and the rationality of the regression equation. Because the NDVI products of two similar sensors are generally close, adding 1 (0, 0) point (when the addition conditions are met) can significantly reduce the absolute value of the intercept and maintain the variability of the data. Although the RMSE of training samples other than (0, 0) points will increase slightly, its impact is not large. Previous studies have also shown that overfitting training samples may cause the model to have low accuracy on the test samples, and the rationality of the model is more important. On the other hand, the regression equation is used pixel by pixel and month by month. The data of a single pixel in a single month is equivalent to a subset of the training sample (the data of 3 consecutive months in a 3*3 window). The regression equation after adding the point (0, 0) may be beneficial to this subset, or have a small impact. However, since the slope and intercept of each regression equation are closer to the reasonable range, the final spatiotemporal variation is also smaller, which meets the principle of similarity in geography.

[0129] Some embodiments of the present application use a final regression algorithm to fit the first sensor sample data and the second sensor sample data to obtain a larger slope and a smaller intercept, which helps to maintain the variability of the data.

[0130] Optionally, the first sensor sample data and the second sensor sample data are obtained by extracting data within a preset window range of an overlapping period of original sample data of different types.

[0131] The embodiment of the present application performs pixel-by-pixel and month-by-month correction coefficient calculations for data with an overlapping observation period of at least 2 years (inclusive). In order to increase the sample size, reduce noise interference, and increase the spatial correlation of the regression coefficient, when fitting the regression equation for each pixel and each month, a 3×3 pixel window around the pixel and data for a total of 3 months including the month and the two months before and after are used. For the overlapping period of 2 years, the regression equation for each pixel and each month will have 9 pixels×3 months×2 years=54 samples.

[0132] The spatial expansion window does not necessarily have to be 3×3 and can be modified according to the specific situation. For example, in areas where the underlying surface is very mean, a 5×5 window can be used to increase the sample size and reduce the impact of random noise.

[0133] The time extension window does not necessarily have to be 3 months, and can also be modified according to specific circumstances, such as using the growing season (May-September) as the time window.

[0134] Some embodiments of the present application set a window range for overlapping period data, extract data based on the window range, expand the data in time and space, and based on the law of geographic autocorrelation (i.e., the principle of proximity in geography), adjacent time and space should have more similar conversion methods.

[0135] Step (7) Apply the regression coefficients pixel by pixel and month by month. Determine whether the data to be converted has noise, and then determine the conversion scheme. Because the conversion relationship is established based on a finite sample, the application effect of the conversion relationship is guaranteed within the range of variation defined by the sample. Beyond this range, the uncertainty of the conversion relationship will increase significantly. Compare the data to be converted with the mean and standard deviation of the corresponding month and pixel position. If the data to be converted is more than 3 standard deviations smaller than the mean, or more than 8 standard deviations higher than the mean, the data to be converted is very likely to be contaminated by noise. The data to be converted needs to be replaced with the mean of the month, and then converted using the slope and intercept; if the data to be converted is not contaminated by noise (i.e., it does not meet the above judgment criteria), it is converted according to the slope and intercept. Here, it should be explained that since most of the noise (such as cloud and cloud shadow pollution, snow pollution, etc.) will reduce NDVI, the judgment rules are stricter when the NDVI is small, and the judgment rules are looser when the NDVI is large.

[0136] It should be noted that each implementable method in this embodiment may be implemented separately, or may be implemented in combination in any manner without conflict, and this application is not limited thereto.

[0137] Another embodiment of the present application provides a sensor data correction device, which is used to execute the sensor data correction method provided by the above embodiment.

[0138] like Figure 2 , which is a schematic diagram of the structure of the sensor data correction device provided in an embodiment of the present application. The sensor data correction device includes an acquisition module 201 and a correction module 202, wherein:

[0139] The acquisition module 201 is used to acquire the first sensor data to be corrected;

[0140] The correction module 202 is used to convert the first sensor data according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a final regression equation.

[0141] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0142] Some embodiments of the present application calculate a large number of different types of sensor sample data, and fit a regression equation according to the calculated results to obtain a conversion model. The regression equation includes at least a least squares regression equation or a finale regression equation. The first sensor data is converted through the conversion model to obtain the second sensor data, thereby ensuring that the influence of data noise is eliminated during the conversion process, thereby further improving the quality of the converted data.

[0143] Another embodiment of the present application further supplements the sensor data correction device provided in the above embodiment.

[0144] Optionally, the device further comprises a setting module, and the setting module is used to:

[0145] Acquire first sensor sample data and second sensor sample data;

[0146] Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data;

[0147] Performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data;

[0148] calculating a mean and a standard deviation of the difference between the fifth sensor sample data and the sixth sensor sample data;

[0149] The sensor sample data are judged according to the mean and the standard deviation to obtain the seventh sensor sample data and the eighth sensor sample data that meet the preset conditions;

[0150] Calculating the correlation coefficient of the seventh sensor sample data and the eighth sensor sample data to obtain a correlation coefficient calculation result;

[0151] According to the calculation results of the correlation coefficient, the regression equation is fitted to obtain the preset conversion model.

[0152] Some embodiments of the present application collect data within a preset window range of the overlapping period of different types of original sample data, and calculate the correction coefficients pixel by pixel and month by month in order to increase the sample size, reduce noise interference, increase the spatial correlation of the regression coefficients, and improve the accuracy of model training.

[0153] Optionally, the module is set to:

[0154] Calculating a first change amplitude of first sensor sample data;

[0155] calculating a second variation amplitude of the second sensor sample data;

[0156] Determining a target amplitude value according to the first change amplitude and the second change amplitude;

[0157] When the target amplitude value satisfies a preset condition, the first sensor sample data corresponding to the target amplitude value is determined as the third sensor sample data, and the second sensor sample data corresponding to the target amplitude value is determined as the fourth sensor sample data.

[0158] Some embodiments of the present application perform amplitude value judgment on sensor sample data and screen the sensor sample data to improve the accuracy of sensor sample data processing.

[0159] Optionally, the module is set to:

[0160] If the third sensor sample data conforms to the normal distribution rule, the third sensor sample data is determined as the fifth sensor sample data;

[0161] If the fourth sensor sample data conforms to the normal distribution rule, the fourth sensor sample data is determined as the sixth sensor sample data.

[0162] In some embodiments of the present application, sensor sample data that do not meet the requirements are deleted by performing normal distribution judgment on the sensor sample data, thereby avoiding the problem of different underlying surfaces being mixed in the sample data, and making the established regression equation more targeted.

[0163] Optionally, the module is set to:

[0164] If the result of the correlation coefficient calculation is a significant positive correlation, then add zero-crossing sample data on the basis of the first sensor sample data and the second sensor sample data. If adding zero-crossing samples can improve the correlation coefficient, the first sensor sample data, the second sensor sample data and the zero-crossing sample data are fitted with a least squares regression equation. If the correlation coefficient cannot be improved, the first sensor sample data and the second sensor sample data are fitted with a least squares regression equation. After the fitting is completed, the mean and variance of the first sensor sample data are recorded.

[0165] Some embodiments of the present application can reduce the absolute value of the final intercept of the least squares linear regression by adding zero-crossing sample data. Because the correlation coefficient between the independent variable and the dependent variable increases, the slope of the regression will be slightly larger than before, which helps to maintain the variability of the independent variable data.

[0166] Optionally, the module is set to:

[0167] If the correlation coefficient calculation result is an insignificant positive correlation, the regression equation is fitted using the final regression algorithm based on the first sensor sample data and the second sensor sample data to obtain the slope and intercept, and the mean and standard deviation of the first sensor sample data are recorded.

[0168] Some embodiments of the present application use a final regression algorithm to fit the first sensor sample data and the second sensor sample data to obtain a larger slope and a smaller intercept, which helps to maintain the variability of the data.

[0169] Optionally, the module is set to:

[0170] If the calculated result of the correlation coefficient is a negative value, the slope is directly set to 1, the intercept is the mean of the second sensor sample data minus the mean of the first sensor sample data, and the mean and standard deviation of the first sensor sample data are recorded.

[0171] Some embodiments of the present application use a conversion method with a slope of 1 and an intercept of the mean difference, which helps to maintain data variability and change direction and eliminate statistical deviations between the first sensor sample and the second sensor sample.

[0172] Optionally, the first sensor sample data and the second sensor sample data are obtained by extracting data within a preset window range of an overlapping period of original sample data of different types.

[0173] Some embodiments of the present application set a window range for overlapping period data, extract data based on the window range, expand the data in time and space, and based on the law of geographic autocorrelation (i.e., the principle of proximity in geography), adjacent time and space should have more similar conversion methods.

[0174] Regarding the device in this embodiment, the specific manner in which each module performs operations has been described in detail in the embodiment of the method, and will not be elaborated here.

[0175] It should be noted that each implementable method in this embodiment may be implemented separately, or may be implemented in combination in any manner without conflict, and this application is not limited thereto.

[0176] The embodiments of the present application further provide a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the operation of the method corresponding to any embodiment of the sensor data correction method provided in the above embodiments can be implemented.

[0177] An embodiment of the present application further provides a computer program product, wherein the computer program product includes a computer program, wherein when the computer program is executed by a processor, it can implement the operations corresponding to the method of any embodiment of the sensor data correction method provided in the above embodiments.

[0178] like Figure 3 As shown, some embodiments of the present application provide an electronic device 300, which includes: a memory 310, a processor 320, and a computer program stored in the memory 310 and executable on the processor 320, wherein the processor 320 can implement any embodiment of the method for correcting sensor data as described above when reading the program from the memory 310 through a bus 330 and executing the program.

[0179] Processor 320 can process digital signals and can include various computing structures, such as complex instruction set computer structure, reduced instruction set computer structure, or a structure that implements a combination of multiple instruction sets. In some examples, processor 320 can be a microprocessor.

[0180] The memory 310 may be used to store instructions executed by the processor 320 or data related to the execution of instructions. These instructions and / or data may include codes for implementing some or all functions of one or more modules described in the embodiments of the present application. The processor 320 of the disclosed embodiment may be used to execute instructions in the memory 310 to implement the method shown above. The memory 310 includes a dynamic random access memory, a static random access memory, a flash memory, an optical memory, or other memory known to those skilled in the art.

[0181] The above are only embodiments of the present application and are not intended to limit the scope of protection of the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application. It should be noted that similar reference numerals and letters represent similar items in the following drawings, so once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0182] 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.

[0183] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

Claims

1. A method for correcting sensor data, characterized in that: The method comprises: Acquiring first sensor data to be corrected; The first sensor data is converted according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a finale regression equation.

2. The sensor data correction method according to claim 1, characterized in that: The preset conversion model is obtained in the following way: Acquire first sensor sample data and second sensor sample data; Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data; Performing a normal distribution test on the third sensor sample data to obtain tested fifth sensor sample data, and performing a normal distribution test on the fourth sensor sample data to obtain tested sixth sensor sample data; Calculating a mean and a standard deviation of the difference between the fifth sensor sample data and the sixth sensor sample data; The sensor sample data are judged according to the mean and the standard deviation to obtain seventh sensor sample data and eighth sensor sample data that meet preset conditions; Calculating a correlation coefficient between the seventh sensor sample data and the eighth sensor sample data to obtain a correlation coefficient calculation result; According to the correlation coefficient calculation result, a regression equation is fitted to obtain the preset conversion model.

3. The sensor data correction method according to claim 2, characterized in that: Performing difference processing on the first sensor sample data to obtain processed third sensor sample data, and performing difference processing on the second sensor sample data to obtain processed fourth sensor sample data, including: Calculating a first change amplitude of the first sensor sample data; Calculating a second variation amplitude of the second sensor sample data; Determining a target amplitude value according to the first change amplitude and the second change amplitude; When the target amplitude value satisfies a preset condition, the first sensor sample data corresponding to the target amplitude value is determined as the third sensor sample data, and the second sensor sample data corresponding to the target amplitude value is determined as the fourth sensor sample data.

4. The sensor data correction method according to claim 2, characterized in that: The performing a normal distribution test on the third sensor sample data to obtain the fifth sensor sample data after the test, and performing a normal distribution test on the fourth sensor sample data to obtain the sixth sensor sample data after the test, comprises: If the third sensor sample data conforms to the normal distribution rule, determining the third sensor sample data as the fifth sensor sample data; If the fourth sensor sample data conforms to the normal distribution rule, the fourth sensor sample data is determined as the sixth sensor sample data.

5. The method for correcting sensor data according to claim 2, characterized in that: The step of fitting a regression equation according to the correlation coefficient calculation result to obtain the preset conversion model includes: If the calculation result of the correlation coefficient is a significant positive correlation, then based on the first sensor sample data and the second sensor sample data, determine whether it is necessary to add zero-crossing sample data. If zero-crossing sample data is added, a least squares regression equation is fitted for the first sensor sample data, the second sensor sample data and the zero-crossing sample data, and the mean and standard deviation of the first sensor sample data are recorded.

6. The sensor data correction method according to claim 2, characterized in that: The step of fitting a regression equation according to the correlation coefficient calculation result to obtain the preset conversion model includes: If the correlation coefficient calculation result is a positive correlation but not significant, a regression equation is fitted using a final regression algorithm based on the first sensor sample data and the second sensor sample data to obtain a slope and an intercept, and the mean and standard deviation of the first sensor sample data are recorded.

7. The sensor data correction method according to claim 1, characterized in that: The first sensor sample data and the second sensor sample data are obtained by extracting data within a preset window range of an overlapping period of different types of original sample data.

8. A sensor data correction device, characterized in that: The device comprises: An acquisition module, used for acquiring first sensor data to be corrected; A correction module is used to convert the first sensor data according to a preset conversion model to obtain second sensor data corresponding to the first sensor data, wherein the preset conversion model is obtained by calculating different types of sensor sample data and fitting a regression equation according to the calculated results, and the regression equation at least includes a least squares regression equation or a finale regression equation.

9. An electronic device, characterized in that: The invention comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor can implement the method for correcting sensor data as claimed in any one of claims 1 to 7 when executing the program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, wherein the program, when executed by a processor, can implement the sensor data correction method described in any one of claims 1 to 7.

Images