Landsat image prediction synthesis method for coastal zone remote sensing observation

Abstract

The invention relates to a coastal zone remote sensing observation-oriented Landsat image prediction synthesis method, which comprises the following steps of: acquiring a real image of an adjacent year of a to-be-predicted synthesis image, preprocessing the real image, and constructing time sequence observation; screening time sequence observation according to seasons and cloud cover; removing abnormal observation values for time sequence observation pixel by pixel; dividing an image water body pixel and a land pixel; respectively fitting available observations of the water body pixels and the land pixels through a functional equation, and respectively constructing surface reflectance time sequence observation curves of the water body pixels and the land pixels; solving an undetermined coefficient of the functional equation; and predicting and synthesizing a Landsat image at the corresponding time according to the fitted time sequence observation curve. The method has the beneficial effects that the Landsat images with high confidence in different time periods in the growth period can be synthesized, so that the requirements of large-range, long-time and high-precision coastal zone monitoring caused by rapid change of land utilization can be met.

Description

Technical Field

[0001] The invention relates to the technical field of remote sensing image prediction and synthesis, and more specifically, to a Landsat image prediction and synthesis method for coastal zone remote sensing observation. Background Art

[0002] Long-term, dynamic earth observation helps humans understand the laws of earth change and better adapt to the earth's environment. Since 1972, the Landsat program has provided medium-resolution earth observations and is the longest-term and most important source of earth observation remote sensing data available to the public. At the same time, Google Earth Engine (GEE), as a cloud-based big data storage and computing platform, incorporates Landsat images, laying the foundation for the global application of Landsat data. However, the data loss caused by the failure of the Landsat-7 scan line corrector (SLC) and the lack of effective data acquisition caused by weather such as clouds and cloud shadows in coastal areas around the world have seriously restricted the promotion and application of Landsat data.

[0003] Missing data interpolation and median synthesis methods are often used to deal with such problems. Missing data interpolation is mainly divided into space-based, time-based and hybrid methods, which mainly use existing observations that are close in space and have similar temporal changes to supplement missing data. The median synthesis method mainly selects the best observations within a certain time interval to generate images, and different best observation selection criteria are often used for remote sensing image data of different resolutions. However, whether it is missing data interpolation or median synthesis, when applied to data missing caused by frequent cloud cover, they show great limitations in time and space dimensions. To address the challenges brought by cloudy weather, spatiotemporal data fusion and model fitting methods have been used to synthesize Landsat images in recent years. Data fusion methods combine Landsat and MODIS data to generate images with high temporal resolution and high spatial resolution. At present, many types of spatiotemporal fusion methods with good results have been developed, including weight function-based, solution-based, learning-based, Bayesian-based and hybrid methods. Spatiotemporal data fusion methods still have certain limitations, such as excessive computational complexity when the landscape is changeable and high dependence on MODIS data. In contrast, the Landsat image synthesis method does not rely on other data through model fitting. Recent studies have shown that the harmonic model using the superposition of sine and cosine functions is more robust than the high-order polynomial fitting model.

[0004] Coastal areas are transitional zones between land and sea, and are susceptible to both natural disasters and human activities. Land cover changes in coastal areas are very rapid, and regular dynamic monitoring by remote sensing is of great significance. However, cloudy and rainy weather frequently occurs in coastal areas, and due to the influence of clouds and cloud shadows, it is very difficult to obtain optical remote sensing images, which is particularly prominent in low-latitude areas in summer. Therefore, for coastal areas, the available Landsat observation data are usually sparse and irregular, which can easily cause poor model fitting and thus distort the synthetic Landsat images. The above issues have received little attention at present, and there is an urgent need to develop a model method to predict and synthesize Landsat images in coastal areas. Summary of the invention

[0005] The purpose of the present invention is to address the deficiencies of the prior art and to propose a Landsat image prediction and synthesis method for coastal zone remote sensing observation.

[0006] Firstly, a Landsat image prediction and synthesis method for coastal remote sensing observation is provided, including:

[0007] Step 1: Obtain the real images of the adjacent years of the synthetic image to be predicted, pre-process the real images of the adjacent years, and construct time series observations by sorting them in order of imaging time;

[0008] Step 2: Filter the time series observations constructed in step 1 by season and cloud cover;

[0009] Step 3: Use dynamic threshold segmentation and fixed threshold segmentation methods to remove abnormal observations pixel by pixel for time series observations;

[0010] Step 4: Divide the image water pixels and land pixels according to the improved normalized difference water index and the frequency of water coverage;

[0011] Step 5: Fit the available observations of water pixels and land pixels respectively through functional equations, and construct the surface reflectance time series observation curves of water pixels and land pixels respectively;

[0012] Step 6: Use ridge regression or least square method to solve the unknown coefficients of the function equation;

[0013] Step 7: Predict and synthesize the Landsat image at the corresponding time based on the fitted time series observation curve.

[0014] Preferably, in step 1, the real images of adjacent years are preprocessed and sorted in sequence by imaging time to construct a time series observation, including: using the Fmask algorithm to remove invalid pixels corresponding to fill, cloud, cloud shadow and snow cover in each image, retaining the remaining valid pixels, and sorting in sequence by imaging time to construct a time series observation; the filling includes Landsat-7 data missing strips.

[0015] Preferably, in step 2, if the number of time series observations in a certain season is greater than or equal to the threshold number, the time series observations of the season are sorted by cloud cover, and the time series observations with the most and the least cloud cover are screened out according to the threshold number; if the number of time series observations in a certain season is less than the threshold number, all time series observations of the season are retained.

[0016] Preferably, in step 4, the improved normalized difference water index is calculated as follows:

[0017]

[0018] Among them, MNDWI refers to the modified normalized difference water index, ρ GREEN represents the time series observation value of the green light band of the pixel, ρ SWIR1 represents the time series observation value of the shortwave infrared 1 band of the pixel, ρ SWIR2 Indicates the time series observation value of the shortwave infrared 2 band of the pixel;

[0019] If the improved normalized difference water index is greater than the index threshold, the pixel is considered to be covered by water.

[0020] Preferably, step 5 comprises:

[0021] Step 5.1, using a first-order polynomial function to fit the available observations of water body pixels, constructing a time series observation curve of the surface reflectance of water body pixels;

[0022] Step 5.2: Use the linear harmonic function to fit the available observations of land pixels and construct the surface reflectance time series observation curve of land pixels.

[0023] Preferably, in step 6, whether there is an extreme distribution of time series observations is determined based on whether the blue light band meets the rules; if it meets the rules, ridge regression is used to solve the parameters of the blue light, green light, and red light bands, and ordinary least squares regression is used to solve the parameters of the remaining bands; otherwise, ordinary least squares regression is used to solve the parameters of all input bands.

[0024] In a second aspect, a Landsat image prediction and synthesis system for coastal remote sensing observation is provided, which is used to execute any of the methods described in the first aspect, including:

[0025] The acquisition module is used to obtain the real images of the adjacent years of the synthetic image to be predicted, and pre-process the real images of the adjacent years, and construct time series observations by sorting them in order according to the imaging time;

[0026] The filter module is used to filter the time series observations constructed by the acquisition module by season and cloud cover;

[0027] A removal module is used to remove abnormal observations pixel by pixel for time series observations by using dynamic threshold segmentation and fixed threshold segmentation methods;

[0028] The division module is used to divide the image water pixels and land pixels according to the improved normalized difference water index and the frequency of water cover;

[0029] A fitting module is used to fit the available observations of water pixels and land pixels respectively through functional equations, and to construct the surface reflectance time series observation curves of water pixels and land pixels respectively;

[0030] A solution module is used to solve the unknown coefficients of the function equation by using the ridge regression method or the least square method;

[0031] The synthesis module is used to predict and synthesize the Landsat image of the corresponding time based on the fitted time series observation curve.

[0032] According to a third aspect, a computer storage medium is provided, wherein a computer program is stored in the computer storage medium; when the computer program is executed on a computer, the computer executes any method described in the first aspect.

[0033] In a fourth aspect, an electronic device is provided, including:

[0034] Memory, used to store computer programs;

[0035] A processor is used to execute the computer program to implement any method as described in the first aspect.

[0036] The beneficial effect of the present invention is that the available Landsat remote sensing images in coastal areas suffer from the influence of cloudy weather, and the time distribution is sparse and irregular. The present invention innovatively proposes a method for fitting and generating synthetic images using a linear harmonic model in combination with Landsat images in coastal areas - a series of operations such as seasonal filtering, threshold segmentation, and ridge regression are used to eliminate the irregular time distribution of Landsat images and the influence of cloudy weather on model fitting. At the same time, different fitting models for water surface and land pixels are used to improve the accuracy of prediction results, and finally a Landsat synthetic image is generated by fitting with a small number of available observation data. This method can synthesize Landsat images with high confidence at different time periods over a long period of time to meet the needs of large-scale, long-term, and high-precision monitoring of coastal areas caused by rapid changes in land use. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of the number and time distribution of Landsat available observation images corresponding to the study area provided in the embodiment of the present invention;

[0038] Figure 2 A schematic diagram of using available observation values ​​of adjacent years to overcome the data sparsity problem provided by an embodiment of the present invention;

[0039] Figure 3 A schematic diagram of overcoming the problem of irregular data distribution by performing seasonal filtering for time series observations provided by an embodiment of the present invention;

[0040] Figure 4 A schematic diagram of eliminating haze and thin clouds in Landsat synthetic images by removing abnormal observation values ​​provided by an embodiment of the present invention;

[0041] Figure 5 A schematic diagram of using ridge regression to eliminate thick clouds in Landsat synthetic images caused by extreme distribution of time series provided by an embodiment of the present invention;

[0042] Figure 6 This is a schematic diagram comparing the effects of Landsat real images and Landsat synthetic images;

[0043] Figure 7 This is a flowchart of the Landsat image prediction and synthesis method for coastal remote sensing observations. DETAILED DESCRIPTION

[0044] The present invention is further described below in conjunction with embodiments. The description of the following embodiments is only used to help understand the present invention. It should be noted that for ordinary persons in the art, without departing from the principle of the present invention, the present invention can also be modified in some ways, and these improvements and modifications also fall within the scope of protection of the claims of the present invention.

[0045] Embodiment 1:

[0046] To solve the problems of the prior art, such as Figure 7 As shown, Embodiment 1 of the present application provides a Landsat image prediction and synthesis method for coastal remote sensing observation, comprising:

[0047] Step 1: Obtain the real images of the adjacent years of the synthetic image to be predicted, preprocess the real images of the adjacent years, and construct a time series observation by sorting them in sequence according to the imaging time.

[0048] Among them, images refer to multi-band image files obtained by direct satellite observations, and time series observations are data sets obtained by arranging the reflectance values ​​of different bands of images in the order of imaging time. The smallest unit of an image is a pixel, and each pixel has a corresponding time series reflectance observation value in different bands.

[0049] In step 1, if the predicted synthetic image corresponds to the year 2012, the real Landsat images of the two consecutive years before and after are collected; if the predicted synthetic image corresponds to other years, the real Landsat images of the one consecutive year before and after are collected to make up for the sparse and irregular distribution of time series observation data.

[0050] The real images of adjacent years are preprocessed and sorted in sequence by imaging time to construct time series observations, including: using the Fmask algorithm to remove invalid pixels corresponding to fill, cloud, cloud shadow and snow cover in each image, retaining the remaining valid pixels, and sorting in sequence by imaging time to construct time series observations; the filling includes Landsat-7 data missing strips.

[0051] Step 2: Filter the time series observations constructed in step 1 by season and cloud cover.

[0052] Specifically, corresponding to each pixel, the time series observations are divided into spring (March-May), summer (June-August), autumn (September-November), and winter (December-February) according to the imaging time.

[0053] If the number of time series observations in a season is greater than or equal to the threshold number, the time series observations of the season are sorted by cloud cover, and the time series observations with the most and least cloud cover are filtered out according to the threshold number; if the number of time series observations in a season is less than the threshold number, all time series observations of the season are retained. For example, if the number of time series observations in a season is greater than or equal to 5, the time series observations of the season are sorted in ascending order by cloud cover, and the observations corresponding to the top 5 cloud cover rankings are retained; if the number of time series observations in a season is less than 5, all observations are retained.

[0054] Step 3: Use dynamic threshold segmentation and fixed threshold segmentation methods to remove abnormal observations pixel by pixel for time series observations.

[0055] Due to the error of the FMask algorithm, there are still a small number of outliers caused by cloud cover in the time series observations. The dynamic threshold segmentation and fixed threshold segmentation methods are further combined to remove the outliers. Among them, the dynamic threshold is defined by the mean and standard deviation of the time series observations. In the blue light band, if an observation value is greater than the mean of the time series observation value plus 1.5 times the standard deviation, it is determined to be an outlier and the observation is removed. The judgment formula is:

[0056] ρ BLUE >μ(ρ BLUE )+1.5σ(ρ BLUE )

[0057] Among them, ρ BLUEis the time series observation value of the blue light band of the pixel, μ and σ refer to the mean and standard deviation of the time series observation of the blue light band of the pixel, respectively. The time series observation value refers to the reflectivity in the time series observation.

[0058] The fixed threshold is defined by a specific value. In the blue light band, if an observation value is greater than 0.15, it is determined to be an outlier and the observation is removed. The judgment formula is:

[0059] ρ BLUE >0.15

[0060] Among them, ρ BLUE is the time series observation value of the blue light band of the pixel.

[0061] Step 4: According to the improved normalized difference water index and the frequency of water cover, the image water pixels and land pixels are divided.

[0062] In step 4, for each pixel, the modified normalized difference water index (MNDWI) for each observation is calculated by the following formula:

[0063]

[0064] Among them, MNDWI refers to the modified normalized difference water index, ρ GREEN represents the time series observation value of the green light band of the pixel, ρ SWIR1 represents the time series observation value of the shortwave infrared 1 band of the pixel, ρ SWIR2 Represents the time series observation value of the shortwave infrared 2 band of the pixel.

[0065] If the MNDWI value of an observation exceeds -0.1, it is considered to be covered by water, expressed as:

[0066] inundation=(MNDWI≥-0.1)

[0067] For each pixel, the water coverage frequency (the ratio of the number of water coverage observations to the number of time series observations) is calculated. If the frequency is greater than 0.66, it is determined to be a water pixel, expressed as:

[0068]

[0069] Otherwise, it is a land pixel, expressed as:

[0070]

[0071] Among them, N inundation is the number of times the pixel is flooded, N time-seres is the number of observations in the pixel time series.

[0072] Step 5: Fit the available observations of water pixels and land pixels respectively through functional equations, and construct the surface reflectance time series observation curves of water pixels and land pixels respectively.

[0073] Step 5 includes;

[0074] Step 5.1: Use a first-order polynomial function to fit the available observations of water pixels and construct a time series observation curve of the surface reflectance of water pixels.

[0075] Specifically, a first-order polynomial (i.e., linear) function is used to fit the changing trend of the surface reflectance of water pixels. The input bands include blue light, green light, red light, near infrared, short-wave infrared 1, and short-wave infrared 2. The expression formula is:

[0076]

[0077] in, is the predicted surface reflectance of the water pixel band i, t represents the Julian day, a 0,i and a 1,i are the intercept and slope coefficients of the fitted model.

[0078] Step 5.2: Use the linear harmonic function to fit the available observations of land pixels and construct the surface reflectance time series observation curve of land pixels.

[0079] Specifically, a linear harmonic function is used to fit the changing trend of the surface reflectance of land pixels. The input bands include blue light, green light, red light, near infrared, shortwave infrared 1 and shortwave infrared 2. The expression formula is:

[0080]

[0081] in, represents the predicted surface reflectance of band i of the land pixel; w represents the frequency, which is a constant (2π / 365.25); a 0,i and a 1,i are the intercept and slope coefficients of the fitting model, respectively, which are used to characterize the overall trend change of the surface reflectivity predicted by the fitting model; a 2j,i and a 2j+1,i It is the mth (m=1, 2)th order harmonic function coefficient, which is used to characterize the periodic annual variation of the surface reflectivity predicted by the fitting model.

[0082] In addition, different orders of harmonic functions are selected according to the number of time series observations. If the number of time series observations exceeds 18, the second-order harmonic function is used, otherwise the first-order harmonic function is used.

[0083] Step 6: Use ridge regression or least squares method to solve the unknown coefficients of the functional equation.

[0084] In step 6, whether the blue light band meets the rules is judged whether there is an extreme distribution of time series observations; if it meets the rules, the ridge regression method is used to solve the parameters of the blue light, green light, and red light bands, and the ordinary least squares regression method is used to solve the parameters of the remaining bands; otherwise, the ordinary least squares regression method is used to solve the parameters of all input bands.

[0085] Specifically, the blue light band is used to define the rules and identify the target pixels that meet the rules. The formula is:

[0086]

[0087] in, represents the predicted value of the surface reflectance of the pixel in the blue light band on July 1; μ(ρ BLUE ) refers to the mean value of the surface reflectance observed in the blue light band time series for that pixel.

[0088] For the blue, green, and red bands of the target pixels that meet the rules, the ridge regression method is used to solve the unknown coefficients of the function to suppress the influence of the extreme distribution of time series observations on the function fitting. The formula is:

[0089]

[0090] in, represents the unknown coefficient of the function solved by ridge regression method, ρ k and They represent the true and fitted model predicted surface reflectance at the kth observation, respectively.

[0091] For the near infrared, shortwave infrared 1 and shortwave infrared 2 bands of the target pixel, as well as all bands of the non-target pixel, the least squares method is used to solve the unknown coefficients of the function. The formula is:

[0092]

[0093] in, represents the unknown coefficients of the function to be solved by the least squares method, ρ k and They represent the true and fitted model predicted surface reflectance at the kth observation, respectively.

[0094] Step 7: Predict and synthesize the Landsat image at the corresponding time based on the fitted time series observation curve.

[0095] Specifically, for pixels with the unknown coefficients of the functions of each band, the coefficients are substituted into the fitting model corresponding to the pixel (such as the corresponding formula for water pixels and the corresponding formula for land pixels), and the predicted surface reflectance of the corresponding band can be obtained by given the target time. The predicted surface reflectance of each band is combined pixel by pixel on the Google Earth Engine platform to generate a Landsat synthetic image of the target time.

[0096] Embodiment 2:

[0097] Based on Example 1, Example 2 of the present application provides a more specific Landsat image prediction and synthesis method for coastal remote sensing observation.

[0098] Four coastal areas of China were selected as study areas, located in different Landsat paths (Path / Row: 120 / 32, 121 / 34, 118 / 39, 122 / 44), with different climatic conditions and land cover types. In order to make the content and advantages of the present invention clearer, the following four study areas are used as cases to illustrate the specific implementation process of the invention in combination with the accompanying drawings, and the detailed steps are as follows:

[0099] Step 1: Obtain the real images of the adjacent years of the synthetic image to be predicted, preprocess the real images of the adjacent years, and construct a time series observation by sorting them in sequence according to the imaging time.

[0100] Specifically, a total of 3307 Landsat Collection 2 Tier 1 surface reflectance images from 1985 to 2020 were collected from the four study areas through the Google Earth Engine (GEE) platform, and the Fmask algorithm was used to remove invalid pixels corresponding to fill (usually missing strips of Landsat-7 data), cloud, cloud shadow and snow cover in each image, and the remaining valid pixels were retained. The distribution of available Landsat observation data in the four selected study areas is as follows: Figure 1 As shown, the observations are obviously sparse and irregularly distributed. Figure 1 a1-d1 in the figure show the number of Landsat observation images available for the study area. Figure 1 a2-d2 show the time distribution.

[0101] In addition, the available observation data of adjacent years in the four study areas are merged. If the year corresponding to the predicted synthetic image is 2012, the valid observations of the two consecutive years before and after are collected to construct the time series observations; if the predicted synthetic image corresponds to other years, the valid observations of the pixels of the one consecutive year before and after are collected to construct the time series observations to make up for the sparse and irregular distribution of the observation data. The merging of available observations in adjacent years is significantly helpful for the accuracy of model fitting. Figure 2 In the figure, a1 is the linear harmonic fitting curve synthesized using only the available observations in 1994 (a2), 1998 (a3), 2000 (a4) and 2002 (a5), b1 is the linear harmonic fitting curve synthesized using the available observations in the adjacent years of 1994 (b2), 1998 (b3), 2000 (b4) and 2002 (b5), and c shows the impact of using or not using the available observations in the adjacent years on the model fitting.

[0102] Step 2: Filter the time series observations constructed in step 1 by season and cloud cover.

[0103] Specifically, seasonal filtering is performed on the time series observations after the available observations of adjacent years are merged. For each pixel, the time series observations are divided into spring (March-May), summer (June-August), autumn (September-November) and winter (December-February) according to the imaging time. The filtering rules are as follows: If the number of available observations in the season is less than 5, all observations are used for model fitting; otherwise, the 5 observations with the least corresponding scene cloud cover are selected for model fitting. Seasonal filtering helps to deal with the problem of irregular observation distribution and further improve the accuracy of model fitting. Figure 3 In the figure, a1 is the linear harmonic fitting curve synthesized by using the time series observations of the adjacent years of 1994 (a2), 1998 (a3), 2000 (a4) and 2002 (a5), and b1 is the linear harmonic fitting curve synthesized by seasonal filtering of the time series observations of the adjacent years of 1994 (b2), 1998 (b3), 2000 (b4) and 2002 (b5). Figure 3 c shows the effect of seasonal filtering of time series observations on model fitting.

[0104] Step 3: Use dynamic threshold segmentation and fixed threshold segmentation methods to remove abnormal observations pixel by pixel for time series observations.

[0105] Remove abnormal observations pixel by pixel. This step can remove haze and thin clouds caused by abnormal observations in the synthetic image. Figure 4 As shown in the figure, (a1-a2) are examples of haze and thin clouds appearing in the Landsat synthetic image before the removal of abnormal observations; (b1-b2) are examples of the effect of eliminating haze and thin clouds in the Landsat synthetic image after the removal of abnormal observations; (c1-c3) correspond to the changes in the fitting curve during the removal of abnormal observations for the pixels marked with the yellow cross c.

[0106] Step 4: According to the improved normalized difference water index and the frequency of water cover, the image water pixels and land pixels are divided.

[0107] Step 5: Fit the available observations of water pixels and land pixels respectively through functional equations, and construct the surface reflectance time series observation curves of water pixels and land pixels respectively.

[0108] The time series observations of blue, green, red, near infrared, shortwave infrared 1, and shortwave infrared 2 bands are input into the fitting model. Specifically, for water pixels, a linear fitting model is used; for land pixels, a linear harmonic fitting model is used. If the number of time series observations exceeds 18, the second-order harmonic function form is used, otherwise the first-order harmonic function form is used.

[0109] Step 6: Use ridge regression or least squares method to solve the unknown coefficients of the functional equation.

[0110] According to the fitting model, the surface reflectance of the blue light band on July 1st of the target year is predicted pixel by pixel, and the rules are used to judge: if the rules are met, the ridge regression method is used to solve the model parameters of the blue light, green light, and red light bands, and the ordinary least squares regression method is used to solve the parameters of the remaining bands; if the rules are not met, the ordinary least squares regression method is used to solve the parameters of all bands. Figure 5 As shown in Figure 1, this step can remove the thick clouds that may appear in the synthetic image due to the extreme distribution of the observation data. The thick clouds caused by the extreme distribution of the observation data are shown in Figure a; as shown in Figure c1-c4, the above rules are used to determine the time series observation values, and the corresponding band model parameters are solved using ridge regression when they meet the rules; as shown in the yellow cross c-marked pixels corresponding to Figure b1-b2, the thick clouds caused by the extreme distribution of the time series observations are successfully removed.

[0111] Step 7: Predict and synthesize the Landsat image at the corresponding time based on the fitted time series observation curve.

[0112] In the GEE platform, the above steps were used to generate Landsat synthetic images of the target time for the four study areas. Some synthetic images were compared with real Landsat images. Figure 6 shown.

[0113] It should be noted that the parts in this embodiment that are the same or similar to those in Embodiment 1 can be referenced to each other and will not be described in detail in this application.

[0114] Embodiment 3:

[0115] On the basis of embodiments 1 and 2, embodiment 3 of the present application provides a Landsat image prediction and synthesis system, including:

[0116] The acquisition module is used to obtain the real images of the adjacent years of the synthetic image to be predicted, and pre-process the real images of the adjacent years, and construct time series observations by sorting them in order according to the imaging time;

[0117] The filter module is used to filter the time series observations constructed by the acquisition module by season and cloud cover;

[0118] A removal module is used to remove abnormal observations pixel by pixel for time series observations by using dynamic threshold segmentation and fixed threshold segmentation methods;

[0119] The division module is used to divide the image water pixels and land pixels according to the improved normalized difference water index and the frequency of water cover;

[0120] A fitting module is used to fit the available observations of water pixels and land pixels respectively through functional equations, and to construct the surface reflectance time series observation curves of water pixels and land pixels respectively;

[0121] A solution module is used to solve the unknown coefficients of the function equation by using the ridge regression method or the least square method;

[0122] The synthesis module is used to predict and synthesize the Landsat image of the corresponding time according to the fitting curve.

[0123] Specifically, the system provided in this embodiment is a system corresponding to the method provided in Embodiments 1 and 2. Therefore, the parts in this embodiment that are the same or similar to Embodiments 1 and 2 can be referenced to each other and will not be repeated in this application.

Claims

1. A Landsat image prediction and synthesis method for coastal remote sensing observation, characterized in that: include: Step 1: Obtain the real images of the adjacent years of the synthetic image to be predicted, pre-process the real images of the adjacent years, and construct time series observations by sorting them in order of imaging time; Step 2: Filter the time series observations constructed in step 1 by season and cloud cover; Step 3: Use dynamic threshold segmentation and fixed threshold segmentation methods to remove abnormal observations pixel by pixel for time series observations; Step 4: Divide the image water pixels and land pixels according to the improved normalized difference water index and the frequency of water coverage; Step 5: Fit the available observations of water pixels and land pixels respectively through functional equations, and construct the surface reflectance time series observation curves of water pixels and land pixels respectively; Step 6: Use ridge regression or least square method to solve the unknown coefficients of the function equation; Step 7: Predict and synthesize the Landsat image at the corresponding time based on the fitted time series observation curve.

2. The Landsat image prediction and synthesis method for coastal zone remote sensing observation according to claim 1, characterized in that: In step 1, the real images of adjacent years are preprocessed and sorted in sequence by imaging time to construct a time series observation, including: using the Fmask algorithm to remove invalid pixels corresponding to fill, cloud, cloud shadow and snow cover in each image, retaining the remaining valid pixels, and sorting in sequence by imaging time to construct a time series observation; the filling includes Landsat-7 data missing strips.

3. The Landsat image prediction and synthesis method for coastal zone remote sensing observation according to claim 2, characterized in that: In step 2, if the number of time series observations in a certain season is greater than or equal to the threshold number, the time series observations of the season are sorted by cloud cover, and the time series observations with the most and the least cloud cover are filtered out according to the threshold number; if the number of time series observations in a certain season is less than the threshold number, all time series observations of the season are retained.

4. The Landsat image prediction and synthesis method for coastal zone remote sensing observation according to claim 3, characterized in that: In step 4, the improved normalized difference water index is as follows: Among them, MNDWI refers to the modified normalized difference water index, ρ GREEN represents the time series observation value of the green light band of the pixel, ρ SWIR1 represents the time series observation value of the shortwave infrared 1 band of the pixel, ρ SWIR2 Indicates the time series observation value of the shortwave infrared 2 band of the pixel; If the improved normalized difference water index is greater than the index threshold, the pixel is considered to be covered by water.

5. The Landsat image prediction and synthesis method for coastal zone remote sensing observation according to claim 4, characterized in that: Step 5 includes; Step 5.1, using a first-order polynomial function to fit the available observations of water body pixels, constructing a time series observation curve of the surface reflectance of water body pixels; Step 5.2: Use the linear harmonic function to fit the available observations of land pixels and construct the surface reflectance time series observation curve of land pixels.

6. The Landsat image prediction and synthesis method for coastal zone remote sensing observation according to claim 5, characterized in that: In step 6, whether the blue light band meets the rules is judged whether there is an extreme distribution of time series observations; if it meets the rules, the ridge regression method is used to solve the parameters of the blue light, green light, and red light bands, and the ordinary least squares regression method is used to solve the parameters of the remaining bands; otherwise, the ordinary least squares regression method is used to solve the parameters of all input bands.

7. A Landsat image prediction and synthesis system for coastal zone remote sensing observation, characterized in that: Used to perform the method according to any one of claims 1 to 6, comprising: The acquisition module is used to obtain the real images of the adjacent years of the synthetic image to be predicted, and pre-process the real images of the adjacent years, and construct time series observations by sorting them in order according to the imaging time; The filter module is used to filter the time series observations constructed by the acquisition module by season and cloud cover; A removal module is used to remove abnormal observations pixel by pixel for time series observations by using dynamic threshold segmentation and fixed threshold segmentation methods; The division module is used to divide the image water pixels and land pixels according to the improved normalized difference water index and the frequency of water cover; A fitting module is used to fit the available observations of water pixels and land pixels respectively through functional equations, and to construct the surface reflectance time series observation curves of water pixels and land pixels respectively; A solution module is used to solve the unknown coefficients of the function equation by using the ridge regression method or the least square method; The synthesis module is used to predict and synthesize the Landsat image of the corresponding time based on the fitted time series observation curve.

8. A computer storage medium, characterized in that: The computer storage medium stores a computer program; when the computer program is executed on a computer, the computer executes any one of the methods described in claims 1 to 6.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method according to any one of claims 1 to 6.

Images