A time series index data fusion method and device based on multi-source images

Abstract

The application provides a time sequence index data fusion method and device based on multi-source images, which comprises the following steps: obtaining two kinds of basic remote sensing image data sets and a target remote sensing image data set from a target area; based on preset time interval requirements, corresponding standard time sequences and cloud-free time sequence data sets, further constructing corresponding regression models and fusion data sets; performing reconstruction processing on the fusion data sets to obtain a first basic remote sensing result data set; constructing a second basic remote sensing result data set by using the first basic remote sensing result data set; and determining a result data set by using the second basic remote sensing result data set. The method provided by the application is to select a standard time sequence constructed based on a self-defined requirement, and to respectively construct time sequence data sets according to the time sequence, so that the time interval is considered, the representative data in the time interval is reconstructed by using the weight in the calculation time, the requirement of the self-defined time resolution is met, and the difference in subsequent time comparison is reduced.

Description

Technical Field

[0001] This invention relates to the field of remote sensing data fusion technology, and in particular to a method and apparatus for temporal index data fusion based on multi-source images. Background Technology

[0002] Remote sensing index data plays a crucial role in image recognition and remote sensing monitoring. Vegetation indices and water indices, for example, can simply and effectively reflect the relationship between the data and ground features. Currently, widely used high temporal resolution MODIS imagery, medium spatial resolution LANDSAT imagery, and high spatial resolution SENTINEL2 imagery are all affected by surface meteorological factors, making it impossible for remote sensing monitoring to meet the requirements of continuous dynamic tracking. To meet the demand for dynamic remote sensing monitoring of surface information that simultaneously requires remote sensing data with high spatial and temporal resolution characteristics, some researchers have proposed a technique that integrates the spatial resolution characteristics of high spatial resolution remote sensing data with the temporal resolution characteristics of high temporal resolution remote sensing data—a technique known as remote sensing data spatiotemporal fusion technology.

[0003] Existing technologies typically suffer from at least one of the following problems: for example, low reconstruction quality in areas with continuous data loss; unsatisfactory data fusion results when spatial resolutions differ significantly; too limited support for specific data types, failing to meet the needs of other data types besides vegetation indices; or inability to maximize the representation of the actual situation by the data. Summary of the Invention

[0004] The technical problem this invention aims to solve is how to alleviate the technical limitations of existing technologies in achieving long-term, high-frequency, and high-precision remote sensing monitoring. In view of this, this invention provides a method and apparatus for temporal exponential data fusion based on multi-source imagery.

[0005] The technical solution adopted in this invention is the time-series exponential data fusion method based on multi-source images, characterized in that it includes:

[0006] Step S1: Obtain the first basic remote sensing image dataset, the second basic remote sensing image dataset, and the target remote sensing image dataset from the target area. Based on the preset time interval requirement, determine the standard time series corresponding to the time interval requirement: the first cloudless time series dataset, the second cloudless time series dataset, and the third cloudless time series dataset.

[0007] Step S2: Construct a first regression model for the first cloudless time series dataset and the second cloudless time series dataset; construct a second regression model for the second cloudless time series dataset and the third cloudless time series dataset; construct a first fused dataset based on the first regression model; and determine the second fused dataset using the first fused dataset and the first cloudless time series dataset.

[0008] Step S3: Reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset;

[0009] Step S4: Based on the standard time series, determine the value of each basic result cell in the first basic remote sensing result dataset to construct the second basic remote sensing result dataset;

[0010] Step S5: Resample the second basic remote sensing result dataset, use the second regression model to construct the regression result dataset of the resampled second basic remote sensing result dataset, and then use the regression result dataset and the third cloudless time series dataset to determine the result dataset.

[0011] In one embodiment, step S1 includes:

[0012] Step S101: Based on the preset number of days required for the time interval, construct a standard time series for each year with the time interval as one period. If the number of days at the end of the year is less than the time interval, padding is added until the time interval is reached.

[0013] Step S102: Using the first basic remote sensing image dataset, construct an intermediate dataset based on cloud masking to remove clouds, and synthesize the MODIS daily NDVI dataset according to the time interval. Then, combine the intermediate datasets into a first cloudless time series dataset.

[0014] Step S103: Using the second basic remote sensing image dataset, construct an intermediate dataset based on FMASK that synthesizes LANDSAT8 and LANDSAT9 data according to the time interval, and then combine the intermediate data into a second cloudless time series dataset.

[0015] Step S104: Using the target remote sensing image dataset, construct an intermediate dataset based on s2cloudless of SENTINEL2 data according to the time interval, and then combine the intermediate data into a third cloudless time series dataset.

[0016] In one implementation, step S2 includes:

[0017] Step S201: Using at least one bicubic convolutional interpolation, the first cloudless time series dataset is resampled into a first fused dataset with the same resolution as the second cloudless time series dataset, and a first regression model is constructed based on the first time series dataset and the second cloudless time series dataset.

[0018] Step S202: Reconstruct the first cloudless time series dataset based on the first regression model, and perform MOSAIC mosaic fusion with the second cloudless time series dataset to generate a second fused dataset;

[0019] Step S203: Construct a second regression model based on the second cloudless time series dataset and the third cloudless time series dataset.

[0020] In one implementation, step S3 includes:

[0021] Step S301: Construct a more complete interpolated dataset from the second fused dataset using linear interpolation;

[0022] Step S302: Using the interpolated dataset, reconstruct the complete NDVI time series based on the weighted SG filtering algorithm to obtain the complete first basic remote sensing result dataset.

[0023] In one implementation, step S4 includes:

[0024] Step S401: Based on the standard time series, in the first basic remote sensing result dataset, find the corresponding MODIS and LANDSAT NDVI image pairs with each time point as the reference, obtain the LANDSAT similar pixel set corresponding to the MODIS pixel by setting the window size, calculate the distance weight between the MODIS pixel and the LANDSAT pixel, and obtain the value of the basic result pixel by weighted summation of the distance weights.

[0025] Step S402: If no similar base pixel is found, the value of the base result pixel is obtained by using the average reference sequence of neighboring pixels to construct the second base remote sensing result dataset.

[0026] In one implementation, step S5 includes:

[0027] Step S501: Using at least one bicubic convolution interpolation, the second basic remote sensing result dataset is resampled into a time-series NDVI dataset with the same resolution as the SENTINEL2 data, and a regression result dataset of the second basic remote sensing result dataset is constructed based on the second regression model.

[0028] Step S502: Based on the third cloudless time series dataset and the regression result dataset, the final result dataset is generated by fusing and embedding them.

[0029] Another aspect of the present invention provides a temporal index data fusion apparatus based on multi-source images, comprising:

[0030] The multi-source data reconstruction module is configured to: acquire a first basic remote sensing image dataset, a second basic remote sensing image dataset, and a target remote sensing image dataset from the target area; and determine a standard time series corresponding to the preset time interval requirement, namely, a first cloudless time series dataset, a second cloudless time series dataset, and a third cloudless time series dataset, based on the preset time interval requirement.

[0031] The dataset filling module is configured to: construct a first regression model for the first cloudless time series dataset and the second cloudless time series dataset, construct a second regression model for the second cloudless time series dataset and the third cloudless time series dataset, construct a first fused dataset based on the first regression model, and determine a second fused dataset using the first fused dataset and the first cloudless time series dataset.

[0032] The reconstruction module is configured to: reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset;

[0033] The fusion computing module is configured to: determine the value of each basic result cell in the first basic remote sensing result dataset based on the standard time series, so as to construct a second basic remote sensing result dataset;

[0034] The determination module is configured to: resample the second basic remote sensing result dataset, construct a regression result dataset of the resampled second basic remote sensing result dataset using the second regression model, and then determine the result dataset using the regression result dataset and the third cloudless time series dataset.

[0035] Another aspect of the present invention provides an electronic device comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the temporal exponential data fusion method based on multi-source images as described in any of the preceding claims.

[0036] Another aspect of the present invention provides a computer storage medium storing a computer program that, when executed by a processor, implements the steps of the temporal exponential data fusion method based on multi-source images as described in any of the preceding claims.

[0037] By adopting the above technical solution, the temporal exponential data fusion method based on multi-source images of the present invention has at least the following advantages:

[0038] 1) The method provided by this invention selects a standard time series based on custom requirements and constructs time series datasets separately according to the time series. It takes into account the time interval and uses the weight of time calculation to reconstruct representative data within the time interval, which meets the needs of custom time resolution and also reduces the differences in subsequent time comparisons.

[0039] 2) The method provided by this invention can choose to use three different remote sensing datasets for fusion: high temporal resolution data for temporal interpolation, medium resolution data as intermediate dataset, and high spatial resolution dataset as target dataset. This reduces the fusion uncertainty caused by the large gap between low and high resolution datasets. At the same time, the use of medium resolution data for complementation increases data accuracy and improves data authenticity.

[0040] 3) In dealing with the numerical differences caused by different sensors, this invention adopts the processing method of regression model, constructs a model between pairs of datasets, and performs data consistency processing before data fusion, which increases the consistency of the fusion comparison and ensures the accuracy of the fusion result data. Attached Figure Description

[0041] Figure 1 This is a flowchart of a temporal index data fusion method based on multi-source images according to an embodiment of the present invention;

[0042] Figure 2 This is a logic flowchart of a temporal index data fusion method based on multi-source images according to an embodiment of the present invention;

[0043] Figure 3 This is another logical flowchart of the temporal exponential data fusion method based on multi-source images according to an embodiment of the present invention;

[0044] Figure 4 This is a rendering of an application example according to an embodiment of the present invention;

[0045] Figure 5 This is a schematic diagram of the device composition of the temporal exponential data fusion method based on multi-source images according to an embodiment of the present invention;

[0046] Figure 6 This is a schematic diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0047] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the present invention will be described in detail below with reference to the accompanying drawings and preferred embodiments.

[0048] In the accompanying drawings, the thickness, size, and shape of the objects have been slightly exaggerated for ease of illustration. The drawings are for illustrative purposes only and are not drawn to scale.

[0049] It should also be understood that the terms "comprising," "including," "having," "containing," and / or "comprising," when used in this specification, indicate the presence of the stated features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or combinations thereof. Furthermore, when expressions such as "at least one of..." appear after a list of listed features, they modify the entire listed feature, not individual elements in the list. Additionally, when describing embodiments of this application, the word "may" is used to mean "one or more embodiments of this application." And the term "exemplary" is intended to refer to an example or illustration.

[0050] As used herein, the terms “basically,” “approximately,” and similar terms are used as terms of approximation rather than terms of degree, and are intended to describe inherent biases in measured or calculated values ​​that will be recognized by those skilled in the art.

[0051] Unless otherwise specified, all terms used herein (including technical and scientific terms) shall have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms (e.g., those defined in common dictionaries) shall be interpreted as having the meaning consistent with their meaning in the context of the relevant art and shall not be interpreted in an idealized or overly formal sense unless expressly so specified herein.

[0052] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0053] The steps described in the specification and the flowcharts in the accompanying drawings of this invention are not necessarily to be strictly followed according to the step numbers; the execution order of the steps can be changed. Furthermore, certain steps can be omitted, multiple steps can be combined into one step, and / or one step can be broken down into multiple steps.

[0054] To facilitate understanding of this article, some specialized terms or concepts used in this article will be explained below.

[0055] NDVI: Normalized Difference Vegetation Index. The Normalized Difference Vegetation Index (NDVI) quantifies vegetation by measuring the difference between near-infrared light (strong reflection from vegetation) and red light (absorption by vegetation).

[0056] Calculation formula:

[0057] MOSAIC (Mosaic) refers to the process of stitching together multiple adjacent remote sensing images into a large, seamless image under certain mathematical control.

[0058] MODIS: Medium Resolution Imaging Spectroradiometer, referring here to the MCD43A4 dataset. The MCD43A4V6 Nadir Bidirectional Reflectance Distribution Function Adjusted Reflectance (NBAR) product provides 500-meter reflectance data for MODIS "Land" bands 1-7. This data is adjusted using the bidirectional reflectance distribution function to simulate values ​​collected from the nadir perspective. The data was generated daily over a 16-day search period, with images taken on day 9. This product combines data from the Terra and Aqua spacecraft, selecting the best representative pixels from the 16-day period.

[0059] LANDSAT: Land Resources Satellite, here referring to the collective data from Landsat 8 and Landsat 9 satellites, using a dataset that combines the two datasets.

[0060] SENTINEL2: SENTINEL-2 is a European wide-scan, high-resolution, multispectral imaging Earth observation mission consisting of two AB satellites.

[0061] SG (Savitzky–Golay) filtering is a filtering method based on local polynomial least squares fitting in the time domain.

[0062] In the first embodiment of the present invention, a method for temporal exponential data fusion based on multi-source images is provided, such as... Figure 1 As shown, the specific steps include the following:

[0063] Step S1: Obtain the first basic remote sensing image dataset d01, the second basic remote sensing image dataset d02, and the target remote sensing image dataset d03 from the target area. Based on the preset time interval requirement, determine the standard time series corresponding to the time interval requirement: the first cloudless time series dataset d11, the second cloudless time series dataset d12, and the third cloudless time series dataset d13.

[0064] Step S2: Construct a first regression model lm1112 for the first cloudless time series dataset d11 and the second cloudless time series dataset d12, construct a first regression model lm1213 for the second cloudless time series dataset d12 and the third cloudless time series dataset d13, and construct a first fused dataset based on the first regression model lm1112. Use the first fused dataset and the first cloudless time series dataset d11 to determine the second fused dataset.

[0065] Step S3: Reconstruct the second fused dataset to obtain the first basic remote sensing result dataset d21 corresponding to the second fused dataset;

[0066] Step S4: Based on the standard time series, determine the value of each basic result cell in the first basic remote sensing result dataset d21 to construct the second basic remote sensing result dataset d31;

[0067] Step S5: Resample the second basic remote sensing result dataset d31, construct the regression result dataset of the resampled second basic remote sensing result dataset d31 using the first regression model lm1213, and then determine the result dataset using the regression result dataset and the third cloudless time series dataset d13.

[0068] The following will combine Figures 1 to 3 The method provided in this embodiment will be described in detail.

[0069] Step S1: Obtain the first basic remote sensing image dataset d01, the second basic remote sensing image dataset d02, and the target remote sensing image dataset d03 from the target area. Based on the preset time interval requirement, determine the standard time series corresponding to the time interval requirement: the first cloudless time series dataset, the second cloudless time series dataset, and the third cloudless time series dataset.

[0070] For specific spatiotemporal resolution requirements, a specific time series is first constructed based on the temporal resolution. To ensure the consistency of time differences within the time intervals, the position of the data within that time interval is considered when constructing the specific temporal resolution, and temporal weight values ​​are calculated. Based on the target region, two basic remote sensing image datasets d01 and d02, as well as the target remote sensing image dataset d03, are obtained. Corresponding cloud removal, noise filtering, and fusion processing methods are then used to construct cloud-free temporal series index datasets for their respective specific temporal resolutions.

[0071] In some implementations, step S1 may further include:

[0072] Step S101: Based on the preset number of days required for the time interval (optionally, the time interval can be 8 days), construct a standard time series for each year with the time interval as one period. If the number of days at the end of the year is less than the time interval, padding is added until the time interval is reached.

[0073] Step S102: Using the basic remote sensing image dataset d01, construct an intermediate dataset by synthesizing the MODIS daily NDVI dataset according to time intervals based on cloud masking to remove clouds, and then combine the intermediate datasets into the first cloudless time series dataset d11.

[0074] Step S103: Using the basic remote sensing image dataset d02, construct an intermediate dataset based on FMASK that synthesizes LANDSAT8 and LANDSAT9 data according to time intervals, and then combine the intermediate datasets into a second cloudless time series dataset d12.

[0075] Step S104: Using the target remote sensing image dataset d03, construct an intermediate dataset synthesized from SENTINEL2 data according to time intervals based on s2cloudless, and then combine the intermediate data into a third cloudless time series dataset d13.

[0076] Step S2: Construct a first regression model lm1112 for the first cloudless time series dataset d11 and the second cloudless time series dataset d12, construct a first regression model lm1213 for the second cloudless time series dataset d12 and the third cloudless time series dataset d13, and construct a first fused dataset dt1 based on the first regression model lm1112. Use the first fused dataset dt1 and the first cloudless time series dataset d11 to determine the second fused dataset dt2.

[0077] In some implementations, step S2 may further include:

[0078] Step S201: Using at least one bicubic convolution interpolation (preferably three), the first cloudless time series dataset d11 is resampled into a first fusion dataset dt1 with the same resolution as the second cloudless time series dataset d12. Based on the first time series dataset and the second cloudless time series dataset d12, a first regression model lm1112 is constructed.

[0079] Step S202: Reconstruct the first cloudless time series dataset d11 based on the first regression model lm1112, and perform MOSAIC mosaic fusion with the second cloudless time series dataset d12 to generate the second fused dataset dt2.

[0080] Step S203: Construct the first regression model lm1213 based on the second cloudless time series dataset d12 and the third cloudless time series dataset d13.

[0081] Step S3: Reconstruct the second fused dataset dt2 to obtain the first basic remote sensing result dataset d21 corresponding to the second fused dataset dt2;

[0082] The dataset obtained after step S2 is the complement result that retains the true data to the greatest extent. However, this result has noise in the data and missing data after cloud removal. Next, linear interpolation can be used to fill in the missing regions, and a weighted SG filtering algorithm can be used to complete the reconstruction of the time series exponent with high quality.

[0083] Specifically, in some implementations, step S3 may further include:

[0084] Step S301: Construct a more complete interpolated dataset from the second fused dataset dt2 using linear interpolation;

[0085] Step S302: Using the interpolated dataset, reconstruct the complete NDVI time series based on the weighted SG (Savitzky–Golay) filtering algorithm to obtain the complete first basic remote sensing result dataset d21.

[0086] Step S4: Based on the standard time series, determine the value of each basic result cell in the first basic remote sensing result dataset d21 to construct the second basic remote sensing result dataset d31.

[0087] Based on the standard time series in step S1, image pairs corresponding to the underlying data are found for each time point. Since the reconstructed dataset of the time series has been pre-built, there is no need to consider the deviation in time correspondence. Similar pixel sets for the underlying data pairs are found by traversing the pixels. The underlying image set is based on the filtered and smoothed data after mosaicking and fusion of two image pairs. This satisfies most cases in the step of finding similar pixels. For cases that do not meet the requirements, the set is obtained by averaging the reference sequence of adjacent pixels.

[0088] Specifically, in some implementations, step S4 may further include:

[0089] Step S401: Based on the standard time series, in the first basic remote sensing result dataset d21, find the corresponding MODIS and LANDSAT NDVI image pairs for each time point as the reference, obtain the LANDSAT similar pixel set corresponding to the MODIS pixel by setting the window size, calculate the distance weight between the MODIS pixel and the LANDSAT pixel, and obtain the value of the basic result pixel by weighted summation of the distance weights; specifically, find the corresponding MODIS and LANDSAT NDVI image pairs for each time point as the reference, obtain the LANDSAT similar pixel set corresponding to the MODIS pixel by setting the window size, calculate the distance weight between the MODIS pixel and the LANDSAT pixel, and obtain the value of the basic result pixel by weighted summation of the distance weights.

[0090] In step S402, if no similar base pixel is found, the value of the base result pixel is obtained by using the average reference sequence of neighboring pixels through an implementation method similar to step S401, so as to construct the second base remote sensing result dataset d31.

[0091] Step S5: Resample the second basic remote sensing result dataset d31, construct the regression result dataset of the resampled second basic remote sensing result dataset d31 using the first regression model lm1213, and then determine the result dataset d41 using the regression result dataset dt4 and the third cloudless time series dataset d13.

[0092] Specifically, to ensure the requirements of specific spatial resolution and the synergistic unity of fused data, a specific dataset, namely the target remote sensing image dataset d03, can be introduced.

[0093] In some implementations, step S5 may further include:

[0094] Step S501: Using at least one (preferably three) bicubic convolution interpolation, the second basic remote sensing result dataset d31 is resampled into a time-series NDVI dataset dt3 with the same resolution as the SENTINEL2 data. Based on the first regression model lm1213, the regression result dataset dt4 of the second basic remote sensing result dataset d31 is constructed.

[0095] Step S502: Based on the third cloudless time series dataset d13 and the regression result dataset dt4, the final result dataset d41 is generated by fusing and mosaicking.

[0096] In summary, compared to existing technologies, combining Figure 4 The temporal exponential data fusion method based on multi-source images provided in this embodiment has at least the following advantages:

[0097] 1) The method provided in this embodiment selects a standard time series based on custom requirements and constructs time series datasets separately according to the time series. It takes into account the time interval and uses the weight of time to reconstruct representative data within the time interval, which meets the needs of custom time resolution and also reduces the differences in subsequent time comparisons.

[0098] 2) The method provided in this embodiment can choose to use three different remote sensing datasets for fusion. It uses high temporal resolution data for temporal interpolation, medium resolution data as an intermediate dataset, and high spatial resolution dataset as the target dataset. This reduces the fusion uncertainty caused by the large gap between low and high resolution datasets. At the same time, the use of medium resolution data for complementation increases data accuracy and improves data authenticity.

[0099] 3) In this embodiment, the problem of numerical differences caused by different sensors is addressed by using a regression model to construct a model between pairs of datasets. Data consistency processing is performed before data fusion, which increases the consistency of the fusion comparison and ensures the accuracy of the fusion result data.

[0100] The second embodiment of the present invention, corresponding to the first embodiment, introduces a temporal exponential data fusion device based on multi-source images, such as... Figure 5 As shown, it includes the following components:

[0101] The multi-source data reconstruction module is configured to: acquire a first basic remote sensing image dataset, a second basic remote sensing image dataset, and a target remote sensing image dataset from the target area; and determine a standard time series corresponding to the preset time interval requirement, namely, a first cloudless time series dataset, a second cloudless time series dataset, and a third cloudless time series dataset, based on the preset time interval requirement.

[0102] The dataset population module is configured to: construct a first regression model for a first cloudless time series dataset and a second cloudless time series dataset; construct a second regression model for a second cloudless time series dataset and a third cloudless time series dataset; construct a first fused dataset based on the first regression model; and determine a second fused dataset using the first fused dataset and the first cloudless time series dataset.

[0103] The reconstruction module is configured to reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset.

[0104] The fusion computing module is configured to: determine the value of each basic result cell in the first basic remote sensing result dataset based on the standard time series, in order to construct the second basic remote sensing result dataset;

[0105] The determination module is configured to: resample the second basic remote sensing result dataset, construct a regression result dataset of the resampled second basic remote sensing result dataset using the second regression model, and then determine the result dataset using the regression result dataset and the third cloudless time series dataset.

[0106] In some implementations, the multi-source data reconstruction module is further configured as follows:

[0107] Based on the preset number of days required for the time interval, a standard time series is constructed for each year, with each period consisting of a time interval. If the number of days at the end of the year is less than the required time interval, the interval is padded until it is reached.

[0108] Using the first basic remote sensing image dataset, an intermediate dataset was constructed by synthesizing the MODIS daily NDVI dataset according to time intervals based on cloud masking, and then the intermediate dataset was combined into the first cloudless time series dataset.

[0109] Using the second basic remote sensing image dataset, an intermediate dataset was constructed based on FMASK to synthesize LANDSAT8 and LANDSAT9 data according to time intervals. The intermediate datasets were then combined into the second cloudless time series dataset.

[0110] Using the target remote sensing image dataset, an intermediate dataset synthesized from SENTINEL2 data according to time intervals is constructed based on s2cloudless, and then the intermediate dataset is combined into a third cloudless time series dataset.

[0111] In some implementations, the dataset population module is further configured as follows:

[0112] Using at least one bicubic convolution interpolation, the first cloudless time series dataset is resampled into a first fused dataset with the same resolution as the second cloudless time series dataset. A first regression model is constructed based on the first time series dataset and the second cloudless time series dataset.

[0113] The first cloudless time series dataset is reconstructed based on the first regression model, and then MOSAIC mosaicking is performed with the second cloudless time series dataset to generate the second fused dataset.

[0114] A second regression model was constructed based on the second and third cloudless time series datasets.

[0115] In some implementations, the refactoring module is further configured as follows:

[0116] A more complete interpolation dataset is constructed using linear interpolation on the second fused dataset;

[0117] Using the interpolated dataset, the complete NDVI time series was reconstructed based on the weighted SG filtering algorithm to obtain the complete first basic remote sensing results dataset.

[0118] In some implementations, the fusion computing module is further configured as follows:

[0119] Based on the standard time series, in the first basic remote sensing result dataset, the corresponding MODIS and LANDSAT NDVI image pairs are found with each time point as the reference. By setting the window size, the set of LANDSAT similar pixels corresponding to MODIS pixels is obtained. The distance weight between MODIS pixels and LANDSAT pixels is calculated. The value of the basic result pixel is obtained by weighted summation of the distance weights.

[0120] If no similar base pixel is found, the value of the base result pixel is obtained by using the average reference sequence of neighboring pixels to construct a second base remote sensing result dataset.

[0121] In some implementations, the step determination module is further configured to:

[0122] Using at least one bicubic convolution interpolation, the second basic remote sensing result dataset is resampled into a time-series NDVI dataset with the same resolution as the SENTINEL2 data. Based on the second regression model, a regression result dataset of the second basic remote sensing result dataset is constructed.

[0123] The final result dataset is generated by fusing and embedding the third cloudless time series dataset with the regression result dataset.

[0124] A third embodiment of the present invention provides an electronic device, such as... Figure 6 As shown, it can be understood as a physical device, including a processor and a memory storing processor-executable instructions. When the instructions are executed by the processor, the following operations are performed:

[0125] Step S1: Obtain the first basic remote sensing image dataset, the second basic remote sensing image dataset, and the target remote sensing image dataset from the target area. Based on the preset time interval requirement, determine the standard time series corresponding to the time interval requirement: the first cloudless time series dataset, the second cloudless time series dataset, and the third cloudless time series dataset.

[0126] Step S2: Construct a first regression model for the first cloudless time series dataset and the second cloudless time series dataset, construct a second regression model for the second cloudless time series dataset and the third cloudless time series dataset, construct a first fused dataset based on the first regression model, and determine the second fused dataset using the first fused dataset and the first cloudless time series dataset.

[0127] Step S3: Reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset;

[0128] Step S4: Based on the standard time series, determine the value of each basic result cell in the first basic remote sensing result dataset to construct the second basic remote sensing result dataset;

[0129] Step S5: Resample the second basic remote sensing result dataset, use the second regression model to construct the regression result dataset of the resampled second basic remote sensing result dataset, and then use the regression result dataset and the third cloudless time series dataset to determine the result dataset.

[0130] In the fourth embodiment of the present invention, the process of the time-series exponential data fusion method based on multi-source images is the same as that of the first, second, or third embodiments. The difference lies in the engineering implementation: this embodiment can be implemented using software plus necessary general-purpose hardware platforms. While hardware implementation is also possible, the former is often a preferred method. Based on this understanding, the method of the present invention can be embodied in the form of a computer software product stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk), including several instructions to cause a device to execute the method described in the embodiments of the present invention.

[0131] Through the description of specific embodiments, a more in-depth and specific understanding should be gained of the technical means and effects adopted by the present invention to achieve the intended purpose. However, the accompanying drawings are only provided for reference and illustration and are not intended to limit the present invention.

Claims

1. A method for fusion of temporal index data based on multi-source images, characterized in that, include: Step S1: Obtain the first basic remote sensing image dataset, the second basic remote sensing image dataset, and the target remote sensing image dataset from the target area. Based on the preset time interval requirement, determine the standard time series corresponding to the time interval requirement, the first cloudless time series dataset, the second cloudless time series dataset, and the third cloudless time series dataset. The target remote sensing image dataset is a specific dataset introduced for the specific spatial resolution requirement and the collaborative unification of fused data. Step S2: Construct a first regression model for the first cloudless time series dataset and the second cloudless time series dataset, and construct a second regression model for the second cloudless time series dataset and the third cloudless time series dataset. The method further includes: resampling the first cloudless time series dataset into a first fusion dataset with the same resolution as the second cloudless time series dataset, reconstructing the first cloudless time series dataset using the first regression model, and performing mosaic fusion with the second cloudless time series dataset to generate a second fusion dataset. Step S3: Reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset; Step S4: Based on the standard time series, determine the value of each basic result cell in the first basic remote sensing result dataset to construct the second basic remote sensing result dataset; Step S5: Resample the second basic remote sensing result dataset, use the second regression model to construct the regression result dataset of the resampled second basic remote sensing result dataset, and then use the regression result dataset and the third cloudless time series dataset to determine the result dataset. Step S1 includes: Step S101: Based on the preset number of days required for the time interval, construct a standard time series for each year with the time interval as one period. If the number of days at the end of the year is less than the time interval, padding is added until the time interval is reached. Step S102: Using the first basic remote sensing image dataset, construct an intermediate dataset based on cloud masking to remove clouds, and synthesize the MODIS daily NDVI dataset according to the time interval. Then, combine the intermediate datasets into a first cloudless time series dataset. Step S103: Using the second basic remote sensing image dataset, construct an intermediate dataset based on FMASK that synthesizes LANDSAT8 and LANDSAT9 data according to the time interval, and then combine the intermediate data into a second cloudless time series dataset. Step S104: Using the target remote sensing image dataset, construct an intermediate dataset based on s2cloudless of SENTINEL2 data according to the time interval, and then combine the intermediate data into a third cloudless time series dataset.

2. The temporal exponential data fusion method based on multi-source images according to claim 1, characterized in that, Step S2 includes: Step S201: Using at least one bicubic convolutional interpolation, the first cloudless time series dataset is resampled into a first fused dataset with the same resolution as the second cloudless time series dataset, and a first regression model is constructed based on the first fused dataset and the second cloudless time series dataset. Step S202: Reconstruct the first cloudless time series dataset based on the first regression model, and perform MOSAIC mosaic fusion with the second cloudless time series dataset to generate a second fused dataset; Step S203: Construct a second regression model based on the second cloudless time series dataset and the third cloudless time series dataset.

3. The temporal exponential data fusion method based on multi-source images according to claim 1, characterized in that, Step S3 includes: Step S301: Construct an interpolated dataset from the second fused dataset using linear interpolation; Step S302: Using the interpolated dataset, reconstruct the complete NDVI time series based on the weighted SG filtering algorithm to obtain the complete first basic remote sensing result dataset.

4. The temporal exponential data fusion method based on multi-source images according to claim 1, characterized in that, Step S4 includes: Step S401: Based on the standard time series, in the first basic remote sensing result dataset, find the corresponding MODIS and LANDSAT NDVI image pairs with each time point as the reference, obtain the LANDSAT similar pixel set corresponding to the MODIS pixel by setting the window size, calculate the distance weight between the MODIS pixel and the LANDSAT pixel, and obtain the value of the basic result pixel by weighted summation of the distance weights. Step S402: If no similar base pixel is found, the value of the base result pixel is obtained by using the average reference sequence of neighboring pixels to construct the base remote sensing result dataset.

5. The temporal index data fusion device based on multi-source images according to claim 1, characterized in that, Step S5 includes: Step S501: Using at least one bicubic convolution interpolation, the second basic remote sensing result dataset is resampled into a time-series NDVI dataset with the same resolution as the SENTINEL2 data, and a regression result dataset of the basic remote sensing result dataset is constructed based on the second regression model. Step S502: Based on the third cloudless time series dataset and the regression result dataset, the final result dataset is generated by fusing and embedding them.

6. A temporal index data fusion device based on multi-source images, characterized in that, include: The multi-source data reconstruction module is configured to: acquire a first basic remote sensing image dataset, a second basic remote sensing image dataset, and a target remote sensing image dataset from the target area; and determine a standard time series corresponding to the time interval requirement based on a preset time interval requirement, namely a first cloudless time series dataset, a second cloudless time series dataset, and a third cloudless time series dataset, wherein the target remote sensing image dataset is a specific dataset introduced for the requirement of specific spatial resolution and the collaborative unification of fused data. The dataset filling module is configured to: construct a first regression model for the first cloudless time series dataset and the second cloudless time series dataset, construct a second regression model for the second cloudless time series dataset and the third cloudless time series dataset, construct a first fused dataset based on the first regression model, and determine a second fused dataset using the first fused dataset and the first cloudless time series dataset. The reconstruction module is configured to: reconstruct the second fused dataset to obtain the first basic remote sensing result dataset corresponding to the second fused dataset; The fusion computing module is configured to: determine the value of each basic result cell in the first basic remote sensing result dataset based on the standard time series, so as to construct a second basic remote sensing result dataset; The determination module is configured to: resample the basic remote sensing result dataset, construct a regression result dataset of the resampled basic remote sensing result dataset using the second regression model, and then determine the result dataset using the regression result dataset and the third cloudless time series dataset; The multi-source data reconstruction module is further configured as follows: Based on the preset number of days required for the time interval, a standard time series is constructed for each year, with the time interval as one period. If the number of days at the end of the year is less than the time interval, it is padded to the end until the time interval is reached. Using the first basic remote sensing image dataset, an intermediate dataset is synthesized from the MODIS daily NDVI dataset according to the time interval based on cloud masking to remove clouds. The intermediate dataset is then combined into the first cloudless time series dataset. Using the second basic remote sensing image dataset, an intermediate dataset synthesized from LANDSAT8 and LANDSAT9 data according to the time interval is constructed based on FMASK, and then the intermediate dataset is combined into a second cloudless time series dataset. Using the target remote sensing image dataset, an intermediate dataset synthesized from SENTINEL2 data according to the time interval is constructed based on s2cloudless, and then the intermediate dataset is combined into a third cloudless time series dataset.

7. An electronic device, characterized in that, The electronic device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program, when executed by the processor, implements the steps of the temporal exponential data fusion method based on multi-source images as described in any one of claims 1 to 5.

8. A computer storage medium, characterized in that, The computer storage medium stores a computer program, which, when executed by a processor, implements the steps of the time-series exponential data fusion method based on multi-source images as described in any one of claims 1 to 5.

Images