One-button automatic processing method for Landsat 7 remote sensing image

Abstract

The invention provides a one-button automatic processing method for a Landsat 7 remote sensing image. According to the method provided by the embodiment of the invention, full-process automatic processing of the remote sensing image from radiometric calibration to quality enhancement can be realized, the processing efficiency and the quality stability are remarkably improved, and the user operation threshold is reduced.

Description

Technical Field

[0001] This invention relates to the field of remote sensing image processing technology, and in particular to a one-click automated processing method for Landsat 7 remote sensing images. Background Technology

[0002] Remote sensing image processing technology, as a crucial supporting tool in the field of Earth observation, is widely used in resource monitoring, environmental assessment, and disaster emergency response. Among related technologies, a complete technical system from data preprocessing to quality enhancement has been constructed through the collaborative operation of multi-source data fusion, cloud detection algorithms, and radiometric correction models. Specifically, this technology covers the entire process from sensor data acquisition, atmospheric correction, cloud mask generation to image enhancement, including key steps such as radiometric calibration, cloud classification (based on multispectral features such as NDSI and NDVI), and spatiotemporal fusion and infilling. With the exponential growth of remote sensing data, traditional processing methods are insufficient to meet the demands for efficient collaboration of multi-source heterogeneous data, especially in the field of Landsat 7 satellite SLC-off stripe restoration, where existing technologies have significant limitations.

[0003] However, existing remote sensing image processing methods directly adopt a modular, independent processing mode without establishing an end-to-end collaborative operation mechanism. This may lead to problems such as data flow fragmentation and inconsistent processing results. Specifically, existing technologies typically use a single cloud detection algorithm (such as confidence extraction based on QA bands), but have limitations such as insufficient multispectral feature fusion and low accuracy in cloud type identification. In the field of data incompleteness, traditional methods rely on fixed time windows and lack the ability to dynamically adapt to multi-source data. As a result, when dealing with complex cloud interference (thin clouds, cirrus clouds, thick clouds, and mixed shadow scenes), existing technology systems often require manual intervention to adjust parameters (such as cloud detection thresholds and incompleteness strategy selection), resulting in limited processing efficiency (2-4 hours for a single image) and insufficient quality stability (radiative characteristics of the incomplete area fluctuate by ±15%). This fragmented processing mode not only increases operational complexity but also makes it difficult to meet the urgent needs of non-professional users for standardized and rapid processing results. Summary of the Invention

[0004] The main objective of this invention is to provide a one-click automated processing method for Landsat 7 remote sensing images based on multi-source data fusion.

[0005] To achieve the above objectives, a first aspect of the present invention proposes a one-click automated processing method for Landsat 7 remote sensing images based on multi-source data fusion, comprising: S1, acquire the target image and region of interest data, and determine whether the image is an image acquired after the SLC-off event based on the acquisition time of the target image; S2, Perform radiometric calibration and multi-feature cloud detection on the target image to generate cloud detection results including cloud mask, cloud confidence and spectral features; S3, if the image is after the SLC-off event, local stripe filling is performed based on the Landsat 5 image, sensor differences are eliminated through radiometric consistency adjustment, and a preliminary filled image is generated; S4, Based on the spatiotemporal fusion algorithm of MODIS image and historical Landsat 7 image, the remaining stripe area in the preliminary filled image is filled over a large area to generate a multi-source fusion filled image; S5, perform multi-scale cloud restoration processing on the multi-source fusion-filled image, and dynamically select local spatial interpolation or global historical image median restoration strategy according to the cloud scale to ensure the spatial continuity of the filled image.

[0006] In one embodiment of the present invention, S2 further includes: S21, extract cloud confidence information using the QA_PIXEL bitmask, and use the formula... and Perform bitwise operation analysis; S22, calculate the NDSI and NDVI spectral indices, where NDSI is calculated using the formula... NDVI is generated using the formula generate.

[0007] In one embodiment of the present invention, S3 further includes: S31, a 45-day time window is used when screening Landsat 5 images, using the formula... Limit the data range; S32, when performing radiation consistency adjustment, through the formula Perform linear regression compensation.

[0008] In one embodiment of the present invention, S4 further includes: S41, a 15-day time window is used when screening MODIS images, using the formula... Limit the data range; S42, establish band mapping relationships, including pairing MODIS's sur_refl_b01 with Landsat 7's SR_B1, sur_refl_b04 with SR_B2, etc.

[0009] In one embodiment of the present invention, S5 further includes: S51, through formula Determine cloud size; S52, using a 30-meter spatial resolution for unified processing, through the formula... Generate a global reference image.

[0010] In one embodiment of the present invention, it further includes: S6, perform image enhancement processing on the filled image, using the formula... Perform percentile stretching and set the band range to 0.0 to 1.0.

[0011] The method of this invention realizes end-to-end automation of the remote sensing image processing workflow, significantly improving processing efficiency and quality stability, and reducing the user's operating threshold. Attached Figure Description

[0012] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 The flowchart illustrates a one-click automated processing method for Landsat 7 remote sensing images based on multi-source data fusion, as provided in an embodiment of the present invention. Detailed Implementation

[0013] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0014] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0015] The following describes, with reference to the accompanying drawings, a one-click automated processing method for Landsat 7 remote sensing images based on multi-source data fusion, according to an embodiment of the present invention.

[0016] This embodiment provides a one-click automated processing method for Landsat 7 remote sensing imagery based on multi-source data fusion. For example... Figure 1 As shown, the method includes the following steps: S1, acquire the target image and region of interest data, and determine whether the image is an image acquired after the SLC-off event based on the acquisition time of the target image.

[0017] Specifically, in this step, the system first acquires the target image and its corresponding region of interest (ROI) data, and determines whether the image was acquired after the SLC-off event based on the acquisition time of the target image. This step is a key entry point in the entire image processing workflow, providing a data foundation and time condition judgment basis for subsequent operations such as radiometric calibration, cloud detection, and restoration.

[0018] ROI data is specified using a user-defined `table` variable and overlaid on the map with a transparent red border using the `Map.addLayer` function for easy visualization and analysis. Target imagery data originates from the Landsat 7 `LANDSAT / LE07 / C02 / T1_L2` dataset, with a time range limited to March 2012. The image acquisition time is extracted using the `image.date()` function and compared with the known occurrence time of the SLC-off event (May 31, 2003). The specific logic is as follows: if the target image's acquisition time is later than the SLC-off event time, it is determined to be an SLC-off image, triggering the subsequent multi-source data incomplete process.

[0019] The definition of ROI must match the spatial extent with the image resolution, typically cropped in 30-meter increments. Time determination uses the `difference` method of the `ee.Date` object, comparing data on a daily basis to ensure the temporal accuracy meets the granularity required for image processing. Furthermore, the system uses `dataset.size().getInfo()` to count the number of original images to verify the dataset's validity, ensuring at least one image is available for processing.

[0020] This step is widely used in remote sensing image preprocessing, especially when processing Landsat 7 imagery. Due to SLC-off events causing stripe loss in some images, it's necessary to determine whether to activate the multi-source infill algorithm based on time-based data analysis. For example, in the study area from March 2012, if the image acquisition time is later than May 31, 2003, the system will automatically call the `fillWithLandsat5` and `fillWithMODISFusion` functions to infill the image using Landsat 5 and MODIS data.

[0021] The technical advantage of this step lies in its ability to effectively identify SLC-off images through time-based condition judgment, thereby avoiding unnecessary infilling operations on complete images, improving processing efficiency, and ensuring data consistency. Simultaneously, the explicit delineation of the ROI provides spatial constraints for subsequent regional statistics and image cropping, enhancing the relevance and reliability of the processing results.

[0022] S2, perform radiometric calibration and multi-feature cloud detection on the target image to generate cloud detection results including cloud mask, cloud confidence and spectral features.

[0023] Specifically, this step involves performing radiometric calibration and multi-feature cloud detection on the target image to generate cloud detection results that include cloud masks, cloud confidence levels, and spectral features. This step is a crucial component of remote sensing image preprocessing, aiming to improve the physical consistency of image data and the accuracy of cloud detection, providing high-quality input for subsequent image fusion, restoration, and analysis.

[0024] Radiometric calibration converts the raw digitally quantized (DN) values ​​of an image into physically meaningful surface reflectance or brightness temperature. Specifically, for the optical bands of Landsat 7 (SR_B1 to SR_B7), a linear transformation formula is used. Convert the DN value to surface reflectivity. For the thermal infrared band (ST_B6), use... This data is then converted to luminance temperature. This calibration method conforms to the standard processing workflow of Landsat Surface Reflectance Products (SR), ensuring the comparability of image data across different time points and sensors.

[0025] In multi-feature cloud detection, information such as cloud bit, cloud confidence, and cirrus is extracted by analyzing the image's quality assessment band (QA_PIXEL). Further, spectral features, such as NDSI (Normalized Differential Snow Index) and NDVI (Normalized Differential Vegetation Index), are combined with thresholds (e.g., NDSI < 0.4, NDVI < 0.8) for auxiliary judgment. Simultaneously, brightness index is calculated as a supplementary feature for cloud detection. This method integrates bitmask parsing and spectral threshold analysis, significantly improving the robustness and accuracy of cloud detection.

[0026] Cloud confidence is extracted through shift operations, such as... Where 3 represents the confidence level and 5 represents the shift bit. In the calculation of spectral characteristics, the formulas for NDSI and NDVI are as follows: and These indicators are widely used in remote sensing cloud and snow detection and land cover classification.

[0027] This step is applicable to multi-source remote sensing image fusion and restoration scenarios, especially when processing Landsat 7 SLC-off data. Cloud detection results can effectively identify cloud and shadow areas in the image, providing accurate masking information for subsequent steps such as stripe filling and MODIS fusion. Its technical value lies in two aspects: firstly, standardized radiometric calibration ensures the physical consistency of the image data; secondly, the multi-feature fusion cloud detection method improves the accuracy and confidence of cloud identification, laying a solid foundation for subsequent image quality improvement.

[0028] Furthermore, S2 includes: S21, extract cloud confidence information using the QA_PIXEL bitmask, and use the formula... and Perform bitwise operation analysis.

[0029] Specifically, in some implementations, extracting cloud confidence information via the QA_PIXEL bitmask is a crucial step in remote sensing image preprocessing, and its technical implementation is based on bit manipulation and mask parsing mechanisms. Specifically, QA_PIXEL is a bit field in Landsat 7 imagery used to store pixel quality information, where each bit represents a different quality indicator, such as cloud, shadow, cirrus cloud, or snow. In this step, extracting cloud confidence information via the bitmask mainly relies on the extraction and analysis of specific locations within the QA_PIXEL.

[0030] The extraction of cloud confidence information employs bitwise operations. Specifically, it uses the formula... and Perform bitmasking on QA_PIXEL. Specifically, This means shifting the binary number 1 left by 3 bits, corresponding to the 3rd bit of the cloud mask flag (cloud_bit) in QA_PIXEL, used to determine whether the pixel is marked as a cloud. This means shifting the binary number 3 (i.e., binary 11) left by 5 bits, corresponding to the 5th and 6th bits of the cloud confidence flag in QA_PIXEL, used to quantify the confidence level of the cloud. The value of the corresponding bit is extracted through a bitwise AND operation, and then it is restored to its numerical form through a right shift operation, thus obtaining the cloud confidence level information.

[0031] Cloud confidence indicators are typically represented by integer values ​​from 0 to 3, where 0 indicates no clouds, 1 indicates low-confidence clouds, 2 indicates medium-confidence clouds, and 3 indicates high-confidence clouds. In this step, through... The extracted cloud_bit is a boolean mask (0 or 1), while through The extracted cloud_confidence is an integer mask with a value range of 0-3. These parameters are used for subsequent cloud classification and mask construction. For example, in the advancedMultiFeatureCloudDetection function, cloudConfidence.eq(1).or(cloudConfidence.eq(2)) is used to identify thin clouds with low to medium confidence.

[0032] This step is widely used in multi-source remote sensing data fusion, image inpainting, and quality control. For example, in Landsat 7 SLC-off image processing, cloud mask information is used to identify stripe regions that need to be filled, and combined with Landsat 5 or MODIS data for multi-scale filling and restoration. Furthermore, in the multiScaleTemporalRecovery function, cloud confidence information is used to construct a joint mask of clouds and shadows, thereby guiding time-series-based cloud recovery strategies.

[0033] This step extracts cloud confidence information through precise bitwise operations, providing a reliable data foundation for subsequent cloud detection, cloud incompleteness, and quality assessment. Its innovation lies in combining multi-level masking information with physical features (such as NDSI, NDVI, and brightness) for multi-feature cloud detection, improving the accuracy and robustness of cloud identification. Furthermore, this method exhibits good computational efficiency when processing large-scale remote sensing data, making it suitable for automated processing workflows on cloud platforms such as Google Earth Engine.

[0034] S22, calculate the NDSI and NDVI spectral indices, where NDSI is calculated using the formula... NDVI is generated using the formula generate.

[0035] Specifically, this step involves calculating two key spectral indices, NDSI (Normalized Difference Snow Index) and NDVI (Normalized Difference Vegetation Index), in remote sensing images. The technology is based on multi-band calculations using the surface reflectance bands (SR_B1 to SR_B7) of Landsat 7 to extract information on surface snow cover and vegetation cover.

[0036] The formula for calculating NDSI is as follows: Where green corresponds to the SR_B2 band (560–655 nm) and swir1 corresponds to the SR_B5 band (1550–1750 nm). The formula for calculating NDVI is as follows: The values ​​are defined as follows: nir represents the SR_B4 band (760–900 nm), and red represents the SR_B3 band (630–690 nm). Both indices are obtained by differential and normalized processing of the surface reflectance bands of remote sensing images to enhance the distinguishability of snow cover and vegetation in multispectral space. In implementation, image processing functions of the Google Earth Engine (GEE) platform, such as `select()`, `subtract()`, and `divide()`, are used to perform pixel-by-pixel calculations on the specified bands, and the results are added to the original image as new bands for subsequent analysis and visualization.

[0037] Both NDSI and NDVI have values ​​ranging from -1 to 1. NDSI is typically used for snow cover detection, with values ​​above 0.4 indicating snow cover areas; NDVI is used for vegetation cover assessment, with values ​​above 0.2 indicating good vegetation cover. In this step, band selection strictly follows the Landsat 7 surface reflectance band naming convention (SR_B1 to SR_B7), and calculations are performed using a 30-meter spatial resolution. Furthermore, to ensure calculation accuracy, all bands have undergone radiometric calibration before calculation, converting the raw digital values ​​(DN) to surface reflectance to eliminate sensor response differences and atmospheric effects.

[0038] This step is widely used in the preprocessing and surface feature extraction of remote sensing images, especially in snow cover monitoring, vegetation cover analysis, and surface classification tasks. In this technical solution, the calculation of NDSI and NDVI provides crucial auxiliary information for subsequent cloud detection, multi-source data imputation, and image enhancement. For example, in the `advancedMultiFeatureCloudDetection` function, NDVI and NDSI are used to construct a multi-feature cloud detection model to distinguish between cloud, snow, and vegetation areas, thereby improving the accuracy of cloud masking.

[0039] By calculating NDSI and NDVI, the spectral characteristics of snow cover and vegetation in multispectral images can be effectively enhanced, providing a reliable basis for subsequent cloud detection and land cover restoration. This step plays an important role in multi-source data fusion and image quality improvement. Its calculation results can serve as the basic input for image classification, change detection, and land cover modeling, significantly improving the usability and analytical accuracy of remote sensing data.

[0040] S3, if it is an image after an SLC-off event, then local stripe filling is performed based on the Landsat 5 image, sensor differences are eliminated through radiometric consistency adjustment, and a preliminary filled image is generated.

[0041] Specifically, in the processing of Landsat 7 images following an SLC-off event, if local stripe loss is detected, a local stripe filling method based on Landsat 5 images is employed. This method achieves high-quality restoration of the missing regions through spatiotemporal matching and radiometric consistency adjustment.

[0042] In practice, Landsat 5 images are first selected within a 90-day (±45-day) period before and after the target image date, requiring cloud cover to be less than 20%, and then radiometrically calibrated using the `enhancedScaleFactorsL5` function. Subsequently, the study area is defined using `filterBounds(roi)` to ensure spatial consistency. The Landsat 5 image with the smallest temporal difference from the target image is selected as the source image for infilling, and radiometric consistency is adjusted using the `adjustRadiometricConsistency` function. This function, based on a linear regression model, calculates the slope and intercept of the source and target images in the overlapping region, using the following formula:

[0043] If the regression calculation fails, the unit slope (1) and zero intercept (0) are used by default to adjust the values ​​to eliminate the differences in radiation response between sensors.

[0044] In radiometric calibration, the optical bands (SR_B1 to SR_B7) are multiplied by 0.0000275 and then -0.2 is added to convert the original digital quantization (DN) values ​​to surface reflectance (SR). Linear regression calculations are performed at a spatial resolution of 30 meters, with a maximum pixel count limit of [missing value]. This ensures computational efficiency and accuracy. During the filling process, only the unfilled areas (gapmasks) in the target image are operated on to avoid repeated coverage.

[0045] This method is applicable to the restoration of localized defects in Landsat 7 images caused by SLC-off events, and is particularly effective in image processing after May 31, 2003. By introducing high-quality Landsat 5 data, stripe filling can be achieved in areas without MODIS data support, improving image spatial continuity and usability.

[0046] This step effectively solves the problem of missing stripes in SLC-off images caused by sensor failure. By adjusting the radiometric consistency, it ensures that the filled image and the target image are highly matched in spectral characteristics, providing high-quality input for subsequent cloud detection, enhancement and fusion, and significantly improving the spatiotemporal consistency and analysis accuracy of remote sensing images.

[0047] Furthermore, S3 includes: S31, a 45-day time window is used when screening Landsat 5 images, using the formula... Limit the data range.

[0048] Specifically, in some implementations, a 45-day time window is used when filtering Landsat 5 images, using a formula... By limiting the data scope, the technology is implemented based on time-series matching and data availability optimization strategies for remote sensing imagery. The core objective of this step is to provide high-quality, temporally close alternative data sources for subsequent stripe filling and cloud detection, thereby improving the accuracy and reliability of image fusion and restoration.

[0049] This method first defines the acquisition time of the target image (`targetDate`), and then extends it by 45 days before and after this time point to form a 90-day time window. Within this window, the system loads the Landsat 5 surface reflectance dataset (`LANDSAT / LT05 / C02 / T1_L2`) through the Earth Engine platform and performs spatial filtering in conjunction with the study area (`roi`). Simultaneously, the cloud cover ratio is limited to no more than 20% using `filter(ee.Filter.lt('CLOUD_COVER', 20))` to ensure high data quality of the selected images. Subsequently, radiometric calibration (`enhancedScaleFactorsL5`) and multi-feature cloud detection (`advancedMultiFeatureCloudDetection`) are performed on the filtered image set to further remove low-quality pixels.

[0050] The 45-day time window is an empirically optimized parameter designed to balance temporal resolution and data availability. The cloud cover threshold is set to 20%, conforming to the Landsat data quality standard's definition of "high-quality imagery." The image spatial resolution is 30 meters, consistent with the Landsat 7 SR band, facilitating subsequent spatiotemporal fusion and radiometric consistency adjustments.

[0051] This step is widely used in image restoration workflows following the Landsat 7 SLC-off event, particularly in image processing after May 31, 2003. By introducing Landsat 5 images that are close in time, the striped areas caused by SLC failure can be effectively filled in, providing a more complete data foundation for subsequent MODIS fusion and cloud recovery.

[0052] The technical advantage of this step lies in significantly improving the spatiotemporal continuity and usability of image data. By limiting the time window, it ensures that the selected image has a high temporal correlation with the target image, thereby obtaining more accurate regression coefficients (such as slope and intercept) in radiometric consistency adjustment. Furthermore, this method plays a crucial bridging role in multi-source data fusion, providing reliable data support for achieving high-precision image restoration and enhancement.

[0053] S32, when performing radiation consistency adjustment, through the formula Perform linear regression compensation.

[0054] Specifically, in the radiation consistency adjustment step, the formula is used... This involves achieving linear regression compensation for multi-source remote sensing imagery. The core objective of this step is to eliminate systematic differences in radiometric response between images acquired from different sensors or at different times, thereby improving the accuracy and consistency of image fusion and infilling.

[0055] This step, based on a linear regression model, statistically fits the pixels of the source and target images within the overlapping area. Specifically, it uses the `ee.Reducer.linearFit()` function provided by Earth Engine to perform regional statistics on the image bands within the ROI (Region of Interest) to calculate the regression coefficients (slope) and intercept. If the regression calculation fails (e.g., due to lack of valid data), the default values ​​are used. , This ensures the robustness of the image processing workflow. The adjusted bands are achieved through pixel-by-pixel multiply-accumulate operations, i.e. This completes the radiation consistency correction.

[0056] scale: The resolution for linear regression calculations is set to 30 meters, conforming to the Landsat standard resolution; maxPixels: Set to This ensures that calculations will not fail due to pixel limitations when processing large areas; finalSlope and finalIntercept are dynamically determined based on regression results, and default to 1 and 0 if there is no data, ensuring that the image is not distorted; Band selection includes SR_B1 to SR_B7, covering the visible and near-infrared bands, suitable for multi-source data fusion.

[0057] This step is widely used in multi-source remote sensing data fusion scenarios, such as the fusion and incomplete filling of Landsat 7 SLC-off data with Landsat 5 or MODIS data. When there are stripes or cloud cover gaps in the image, radiometric consistency adjustment is used to ensure that the incomplete data is consistent with the target image in terms of radiometric characteristics, thereby improving the spectral continuity and application reliability of the final image.

[0058] This linear regression compensation method effectively eliminates radiation differences between sensors or over time, improving image fusion quality. In subsequent cloud detection, infilling, and enhancement steps, it ensures data comparability in both spatial and temporal dimensions, providing a reliable data foundation for applications such as high-precision surface monitoring and vegetation index calculation.

[0059] S4. Based on the spatiotemporal fusion algorithm of MODIS image and historical Landsat 7 image, the remaining stripe areas in the preliminary filled image are filled over a large area to generate a multi-source fusion filled image.

[0060] Specifically, this step utilizes a spatiotemporal fusion algorithm based on MODIS imagery and historical Landsat 7 imagery to extensively fill in the remaining striped areas in the initially filled image, generating a multi-source fused filled image. This method, through spatiotemporal consistency modeling, combines high temporal resolution MODIS data with high spatial resolution historical Landsat 7 data to further repair striped areas in Landsat 7 imagery caused by sensor malfunctions (such as SLC-off).

[0061] First, MODIS images within 15 days before and after the target date were selected from the MODIS / 006 / MOD09GA dataset, and the study area and time range were defined using `filterBounds` and `filterDate`. Then, a historical Landsat 7 image was selected, with a time range between 365 and 60 days prior to the target date and cloud cover below 10%. This historical image, after radiometric calibration and cloud detection processing, served as reference data for spatiotemporal fusion. During the fusion process, median images were extracted from both the MODIS image and the historical Landsat 7 image within 7 days before and after the target date and the historical date, respectively. These median images were used to construct inter-band change vectors, and based on these vectors, the pixel values ​​of missing areas in the target image were predicted.

[0062] The MODIS imagery uses bands `sur_refl_b01` to `sur_refl_b07`, while the Landsat 7 imagery uses bands `SR_B1` to `SR_B7`. The band pairings are: `sur_refl_b01` with `SR_B1`, `sur_refl_b04` with `SR_B2`, `sur_refl_b03` with `SR_B3`, `sur_refl_b02` with `SR_B4`, `sur_refl_b06` with `SR_B5`, and `sur_refl_b07` with `SR_B7`. Using a linear regression model, the change vectors from the MODIS imagery are overlaid onto historical Landsat 7 imagery to generate a predicted image used to fill in remaining striped areas.

[0063] In application scenarios, this step is suitable for addressing the issue of widespread data loss in Landsat 7 imagery after an SLC-off event, especially when cloud coverage is high or Landsat 5 data is unavailable. It effectively improves image integrity and availability. Leveraging MODIS's high temporal resolution, this method can acquire alternative data in a short time, thus avoiding prolonged waiting for high-quality imagery.

[0064] In terms of technical effects, this step significantly improves the spatial continuity and spectral consistency of the imagery, filling in areas that were not covered by the initial processing. Through spatiotemporal fusion algorithms, not only is the high spatial resolution of Landsat 7 preserved, but also the temporal information of MODIS is introduced, enhancing the reliability and accuracy of the imagery in applications such as remote sensing monitoring and land use classification.

[0065] Furthermore, S4 includes: S41, a 15-day time window is used when screening MODIS images, using the formula... Limit the data range.

[0066] Specifically, in this method, a 15-day time window is used when screening MODIS images, using the formula... The technology involves limiting the data scope and implementing a strategy based on time-series matching and data availability optimization of remote sensing imagery. This step aims to extract imagery data from the MODIS image collection that is closest in time to the target date (`targetDate`) to ensure the continuity and representativeness of the images in the temporal dimension, thereby providing high-quality input data for subsequent spatiotemporal fusion.

[0067] This step utilizes the `filterDate` function of the Google Earth Engine (GEE) platform to filter the time window. Specifically, it extends 15 days forward and 15 days backward from the target date, forming a 30-day time interval. Within this interval, the system filters out all MODIS image sets that meet the time criteria (`ee.ImageCollection('MODIS / 006 / MOD09GA')`) and confines them to the study area (ROI). This operation ensures that the selected MODIS image is highly synchronized with the target Landsat 7 image in time, thereby improving the spatiotemporal consistency of the fusion result.

[0068] The time window is set at ±15 days, that is... This parameter is based on a combination of the revisit period of MODIS imagery (approximately 1-2 days) and the stripe loss characteristics of Landsat 7 imagery. A 15-day timeframe covers multiple revisit periods of MODIS imagery, increasing the probability of acquiring cloudless or low-cloudy images while avoiding the impact of vegetation or surface condition changes caused by excessively long time spans on fusion accuracy. Furthermore, this step incorporates the geometric constraints of the ROI to ensure that the spatial coverage of the imagery aligns with the study area, avoiding the introduction of invalid data.

[0069] In this application scenario, this step is primarily used for fringe repair during the SLC-off (scan line corrector off) period of Landsat 7 imagery. Because Landsat 7 experienced an SLC failure after May 31, 2003, resulting in significant fringe loss in the images, it was necessary to introduce high temporal resolution images such as MODIS for filling in the gaps. By filtering MODIS images through a 15-day time window, high-quality MODIS data close to the target date can be effectively obtained, providing a reliable data foundation for subsequent spatiotemporal fusion algorithms.

[0070] The technical effects of this step are reflected in two aspects: First, it significantly improves the temporal matching degree between MODIS imagery and the target Landsat 7 imagery, thereby enhancing the spatiotemporal continuity of the fusion results; second, by limiting the time window, it reduces the errors introduced by excessive time differences and improves the infilling accuracy. In the overall processing flow, this step is a key link in realizing the collaborative processing of multi-source data, providing high-quality input data for subsequent steps such as cloud detection, cloud shadow infilling, and radiometric consistency adjustment, and plays an important technical support role.

[0071] S42, establish the band mapping relationship, including MODIS's sur_refl_b01 and Landsat 7's SR_B.

[0072] Specifically, in this invention, establishing a band mapping relationship is a crucial step in realizing the fusion and completion of multi-source remote sensing data. Its core lies in accurately mapping the `sur_refl_b01` band of the MODIS sensor to the `SR_B1` band of the Landsat 7, thereby providing a consistent band structure and physical dimensions for subsequent spatiotemporal fusion algorithms. This step is technically implemented based on a spectral response function (SRF) matching and band renaming mechanism for remote sensing data, ensuring comparability between MODIS and Landsat 7 in terms of spatial resolution, wavelength range, and reflectance units.

[0073] The MODIS `sur_refl_b01` band corresponds to the visible light band (wavelength range approximately 0.62–0.67 μm), while the Landsat 7 `SR_B1` band is the blue band (wavelength range approximately 0.45–0.52 μm), differing in their wavelength center and bandwidth. Therefore, this invention defines a band pairing array `bandPairs` to map MODIS's `sur_refl_b01` to Landsat 7's `SR_B1`, and establishes similar mapping relationships for other bands. This mapping array is called in the `spatialTemporalFusion` function to guide subsequent image fusion operations.

[0074] The reflectance unit in the MODIS band is dimensionless reflectance (unit: 1), while the surface reflectance (SR) band of Landsat 7, after radiometric calibration, typically ranges from 0.0 to 0.3. To ensure data consistency, MODIS data needs to be normalized before fusion, specifically through a linear regression model for radiometric consistency adjustment. The regression model is as follows:

[0075] The `slope` and `intercept` values ​​were obtained by linear fitting of MODIS and Landsat 7 in the overlapping region, ensuring that the MODIS data are consistent with Landsat 7 on the reflectance scale.

[0076] This band mapping relationship is widely used in stripe filling tasks for Landsat 7 SLC-off (Scan Line Corrector failure) imagery. By performing band-level fusion of high temporal resolution data from MODIS with high spatial resolution data from Landsat 7, image missing areas caused by SLC failure can be effectively recovered, especially suitable for areas with high vegetation cover and low cloud cover.

[0077] The technical advantage of this step lies in providing a unified band structure and physical dimensions for the fusion of multi-source remote sensing data, thereby improving the accuracy and reliability of image inpainting. Through precise band mapping, MODIS data can serve as a supplementary data source for Landsat 7 imagery, enabling high-quality spatiotemporal fusion and image restoration, laying a solid foundation for subsequent remote sensing analysis and applications.

[0078] S5, perform multi-scale cloud restoration processing on the multi-source fusion-filled image, and dynamically select local spatial interpolation or global historical image median restoration strategy according to the cloud scale to ensure the spatial continuity of the filled image.

[0079] Specifically, this step involves multi-scale cloud restoration processing on multi-source fused infill images. Its core lies in dynamically selecting local spatial interpolation or global historical image median restoration strategies based on cloud scale to ensure the spatial continuity of the infilled image. This technology is based on the principle of spatiotemporal consistency in remote sensing image processing, combining image masking analysis and statistical regression methods to achieve adaptive restoration of cloud-covered areas.

[0080] First, a cloud mask is constructed using the `multiScaleTemporalRecovery` function, which generates a comprehensive cloud mask `cloudMask` using the `thick_cloud` and `cloud_shadow` bands. Then, the `connectedPixelCount` method is used to calculate the pixel connectivity of the cloud region, with a threshold of 25 pixels to distinguish between localized small-scale clouds and large-scale cloud coverage. For localized cloud regions with 25 or fewer connected pixels, the `focal_mean` method is used to interpolate the mean of a 3×3 circular neighborhood to preserve local ground features. For global cloud regions with more than 25 connected pixels, the median image of a cloudless area from the historical image set is used to replace the local cloud cover, ensuring spectral consistency across large-scale cloud-covered areas.

[0081] The neighborhood radius of `focal_mean` is 3 pixels, and the neighborhood shape is circular, with units in pixels. The time window for historical imagery is set to 60 days before and after the current image, and the cloud cover threshold is 20% to ensure that the selected reference imagery has high quality. In addition, the image bands selected are `SR_B1` to `SR_B7`, where `SR_B6` is the thermal infrared band and is not involved in cloud recovery processing.

[0082] This step is applicable to the restoration of Landsat 7 SLC-off imagery, especially in image processing after May 31, 2003, where the problem of missing stripes due to SLC failure is particularly prominent. Through multi-scale cloud restoration strategies, the spatial integrity of the imagery can be effectively improved, providing high-quality data support for subsequent applications such as land use classification and vegetation index calculation.

[0083] This step significantly improves the spatial continuity and spectral consistency of the imagery, and reduces the interference of cloud cover on surface feature extraction. Through dynamic strategy selection, it avoids the blurring effects that local interpolation may introduce and overcomes the spectral shift problems that global replacement may cause, thereby improving processing efficiency and applicability while ensuring image quality.

[0084] Furthermore, S5 includes: S51, through formula Determine cloud scale Specifically, in the steps of this method, through the formula Determining cloud size is a crucial step in the multi-source remote sensing data fusion and cloud detection process. It is used to quantify the connectivity characteristics of the cloud-covered area, thereby providing a basis for subsequent cloud mask construction and data incompleteness.

[0085] This step leverages the image processing capabilities of the Google Earth Engine (GEE) platform, utilizing the `connectedPixelCount` function to count connected pixels in the cloud mask image. Specifically, the `connectedPixelCount` function identifies and counts the number of connected pixel blocks within a cloud region by setting a neighborhood connectivity method (e.g., 8-neighborhood or 4-neighborhood) and a maximum pixel count threshold. In this formula, the parameter `200` represents the maximum connected pixel count threshold, used to distinguish between small cloud patches and large cloud coverage; the parameter `false` indicates the use of a 4-neighborhood connectivity method, considering only pixel connectivity in the four directions (up, down, left, and right), rather than the diagonal direction. This function performs region analysis on the cloud mask image, outputting the number of pixels in each cloud region, thus forming a quantitative indicator of cloud size.

[0086] - Maximum pixel count threshold: 200, used to define the minimum size of the cloud area. Cloud spots smaller than this value will be regarded as small clouds or noise and will not be included in subsequent processing.

[0087] Neighborhood connection method: 4-neighborhood (`false`), ensuring stricter boundary identification of cloud areas and avoiding misjudgments caused by diagonal connections.

[0088] Output variable: `cloudSize`, representing the number of pixels in each cloud region, with a spatial resolution of 30 meters (Landsat 7 standard resolution).

[0089] This step is applicable to cloud cover restoration scenarios in Landsat 7 imagery during SLC-off (Scan Line Corrector failure). In remote sensing data processing in March 2012, some image metadata was missing due to the presence of clouds and shadows. The `connectedPixelCount` function effectively identifies large areas of cloud cover, providing a spatial mask for subsequent multi-source data incompleteness (such as Landsat 5 or MODIS data). This method is particularly suitable for remote sensing image preprocessing workflows that require distinguishing between localized cloud patches and large areas of cloud cover.

[0090] This step, by quantifying the connectivity of cloud regions, enables accurate determination of cloud size, providing crucial input for subsequent cloud masking, data imputation, and spatiotemporal fusion. By setting appropriate pixel thresholds and connectivity methods, noisy cloud spots can be effectively filtered, improving the accuracy and robustness of cloud detection. In multi-source data fusion, this step helps determine whether MODIS or Landsat 5 data needs to be introduced for imputation, thereby improving image integrity and usability, demonstrating significant practical value and algorithmic optimization implications.

[0091] S52, using a 30-meter spatial resolution for unified processing, through the formula... Generate a global reference image.

[0092] Specifically, in this method, remote sensing images are uniformly processed using a 30-meter spatial resolution, and the formula is used... Generate a global reference image. This step is a crucial part of the multi-source remote sensing data fusion and cloud detection and restoration process. Its technical implementation is based on the statistical characteristics and spatial consistency principles of remote sensing images, aiming to provide a stable and consistent reference benchmark for subsequent cloud detection, infilling, and enhancement.

[0093] First, the system loads historical Landsat 7 imagery data for the target area using `ee.ImageCollection` and performs spatial filtering based on the Region of Interest (ROI). Then, quality control is applied to the image collection, including cloud cover threshold filtering (CLOUD_COVER < 20%) and multi-band masking to ensure only high-quality, cloudless or low-cloudy images are retained. Based on this, the system updates the masks for all bands (SR_B1 to SR_B7), retaining only pixels from cloudless areas. Finally, the `median` function performs pixel-level statistics on all processed images, generating a global reference image with a spatial resolution of 30 meters. This reference image is spatially representative and reflects the average surface reflectance characteristics of the target area within a specific time window.

[0094] The spatial resolution is fixed at 30 meters, conforming to the standard resolution of the Landsat 7 ETM+ sensor. The time window is set to 60 days before and after the image acquisition date to ensure sufficient temporal coverage of the reference image set. The cloud cover threshold is set to 20% to exclude images with excessive cloud cover from interfering with the statistical results of the reference images. In addition, the `median` function uses a non-null value statistical method in pixel-level calculations to avoid overall image distortion due to missing pixels.

[0095] In application scenarios, this step is widely applicable to the preprocessing stage of remote sensing images, especially in areas with frequent cloud cover or time-sensitive remote sensing missions. By generating a global reference image, the system can provide stable background information during subsequent cloud detection and infilling processes, thereby improving the accuracy and consistency of image restoration. For example, in the restoration of SLC-off (scan line corrector failure) images, the reference image can serve as an alternative data source for filling missing stripes, ensuring the spatial and temporal continuity of the image.

[0096] The technical advantage of this step lies in its ability to effectively reduce the heterogeneity of multi-source images during the fusion process by unifying spatial resolution and statistical methods, thereby improving the automation and stability of image processing. The generation of a global reference image provides a reliable data foundation for subsequent steps such as multi-scale cloud restoration and radiometric consistency adjustment, thus enhancing the robustness and practicality of the entire remote sensing processing workflow.

[0097] Also includes: S6, perform image enhancement processing on the filled image, using the formula... Perform percentile stretching and set the band range to 0.0 to 1.0.

[0098] Specifically, in the image enhancement step, the filled remote sensing image undergoes percentile stretching to improve image contrast and visualization. This step, based on the statistical characteristics of image bands, maps the image data to a standardized dynamic range [0.0, 1.0] through linear stretching, thereby enhancing the recognizability of ground features. In practice, firstly, statistical analysis is performed on the target bands (e.g., SR_B1 to SR_B7), calculating their 5th percentile within the region of interest (ROI). ) and the 95th percentile ( The pixel value of ) is used as the minimum value for stretching. ) and maximum value ( The stretching formula is:

[0099] in, This represents the pixel value of the original band. and These are the 5th and 95th percentile values ​​of this band within the ROI, respectively. Using this formula, 90% of the pixel values ​​in the image are compressed to the range of [0.0, 1.0], effectively suppressing the impact of outliers on overall contrast. The stretched image is then used... The function ensures that all pixel values ​​are strictly limited to the target range, avoiding exceeding the effective display interval.

[0100] In practical applications, this step is typically performed after multi-source data infilling (such as Landsat 5 and MODIS fusion infilling) and cloud detection correction, and is suitable for post-processing workflows of Landsat 7 SLC-off imagery. Its operating environment is based on the Google Earth Engine (GEE) platform, utilizing its built-in... and The method performs regional statistics and band updates. This step plays a crucial role in remote sensing image preprocessing, significantly improving the visual appearance of the images and providing high-quality input data for subsequent tasks such as land cover classification and change detection.

[0101] The method of this invention further optimizes the contrast and visual readability of Landsat 7 images by introducing percentile stretching image enhancement processing after stripe filling and cloud detection correction, making the image data in the range of 0.0 to 1.0 more in line with the needs of subsequent analysis, and improving the applicability and accuracy of the processing results.

[0102] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0103] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0104] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A one-click automated processing method for Landsat 7 remote sensing images, characterized in that, include: S1, acquire the target image and region of interest data, and determine whether the image is an image acquired after the SLC-off event based on the acquisition time of the target image; S2, Perform radiometric calibration and multi-feature cloud detection on the target image to generate cloud detection results including cloud mask, cloud confidence and spectral features; S3, if the image is after the SLC-off event, local stripe filling is performed based on the Landsat 5 image, sensor differences are eliminated through radiometric consistency adjustment, and a preliminary filled image is generated; S4, Based on the spatiotemporal fusion algorithm of MODIS image and historical Landsat 7 image, the remaining stripe area in the preliminary filled image is filled over a large area to generate a multi-source fusion filled image; S5, perform multi-scale cloud restoration processing on the multi-source fusion-filled image, and dynamically select local spatial interpolation or global historical image median restoration strategy according to the cloud scale to ensure the spatial continuity of the filled image.

2. The method as described in claim 1, characterized in that, S2 further includes: S21, extract cloud confidence information using the QA_PIXEL bitmask, and use the formula... and Perform bitwise operation analysis; S22, calculate the NDSI and NDVI spectral indices, where NDSI is calculated using the formula... NDVI is generated using the formula generate.

3. The method as described in claim 1, characterized in that, S3 further includes: S31, a 45-day time window is used when screening Landsat 5 images, using the formula... Limit the data range; S32, when performing radiation consistency adjustment, through the formula Perform linear regression compensation.

4. The method as described in claim 1, characterized in that, S4 further includes: S41, a 15-day time window is used when screening MODIS images, using the formula... Limit the data range; S42, establish band mapping relationships, including pairing MODIS's sur_refl_b01 with Landsat 7's SR_B1, sur_refl_b04 with SR_B2, etc.

5. The method as described in claim 1, characterized in that, The S5 also includes: S51, through formula Determine cloud size; S52, using a 30-meter spatial resolution for unified processing, through the formula... Generate a global reference image.

6. The method as described in claim 1, characterized in that, Also includes: S6, perform image enhancement processing on the filled image, using the formula... Perform percentile stretching and set the band range to 0.0 to 1.0.

Images