Temporal and spatial filtering method and device for dynamic residual threshold adaptive selection, equipment and storage medium

Abstract

The application discloses a kind of dynamic residual threshold adaptive selection space-time filtering method, device, equipment and storage medium, the method includes obtaining the remote sensing time series image data of target area;By model training and predicting the possible change trend of remote sensing time series data obtains the prediction result of remote sensing time series data.Furthermore, the absolute residual between image predicted value and image true value is calculated, and the residual threshold determined based on the absolute residual median, quartile range, skewness and kurtosis is selected as the threshold of filtering method selection, specifically, for high residual area, spatiotemporal weighted filtering strategy combined with Euclidean distance is used;For low residual area, the filtering reconstruction method in time domain is used, so as to ensure the continuity of image while suppressing non-systematic noise.Compared with prior art, it has higher degree of automation and adaptability, and is especially suitable for remote sensing time series data processing in high-altitude, frequent cloud cover or complex terrain area.

Description

Technical Field

[0001] This invention relates to the field of digital image processing technology, and in particular to a spatiotemporal filtering method, apparatus, device, and storage medium with dynamic residual threshold adaptive selection. Background Technology

[0002] Since the Landsat archives were made freely available in 2008, the way global remote sensing data is acquired has undergone a fundamental transformation. As an important data source for Earth science research, the optical sensors carried by the Landsat series satellites continuously provide high-quality remote sensing images covering the entire globe, recording land cover changes, vegetation dynamics, and land use processes on a decades-long scale, forming a rich resource of remote sensing time-series data.

[0003] However, due to the superposition of various factors such as the variety of sensors, atmospheric interference, and surface disturbance, Landsat time series data inevitably contains a large amount of noise and outliers, which seriously affects the accuracy and stability of subsequent analysis tasks such as change detection and information extraction, especially in high-altitude, cloudy, and complex terrain areas.

[0004] Existing remote sensing image denoising methods mainly fall into three categories: time-domain filtering methods, such as LandTrendr, Savitzky-Golay (SG), and BISE algorithms, which rely on time-series trend fitting to achieve data smoothing; spatial-domain filtering methods, such as median filtering and Gaussian filtering, which reduce local noise interference based on the pixel spatial neighborhood structure; and frequency-domain methods, such as wavelet filtering and Fourier transform, which identify and remove high-frequency noise through spectrum analysis.

[0005] Although the above methods are effective under certain conditions, they generally have the following limitations: it is difficult to simultaneously ensure temporal continuity and spatial structure fidelity; there is a lack of accurate identification mechanism for outliers, which can easily lead to the misfiltering of actual change information; the degree of automation is low, the parameter dependence is strong, and it is difficult to adapt to the needs of large-scale regional processing; the accuracy of the filtered results is still not ideal in high-altitude areas.

[0006] To address the aforementioned problems, there is an urgent need for a novel spatiotemporal filtering method that can integrate temporal and spatial correlations, automatically identify anomalous regions, and adaptively select the optimal filtering strategy based on dynamic residual threshold selection. This invention proposes an adaptive spatiotemporal filtering method that combines spatiotemporal correlations. It introduces a residual-driven region identification mechanism, effectively identifying anomalous regions without relying on manually set parameters. Based on this, it performs spatiotemporal joint filtering and a dual-channel temporal filtering path, significantly improving the intelligence and adaptability of the filtering process. This method is particularly suitable for high-precision reconstruction and noise suppression of remote sensing time series data in high-altitude areas.

[0007] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0008] The main objective of this invention is to provide a spatiotemporal filtering method, apparatus, device, and storage medium based on residual adaptive selection, aiming to solve the noise pollution problem in long-term image series of high-altitude, frequently clouded, and complex terrain areas.

[0009] To achieve the above objectives, this invention provides a spatiotemporal filtering method with dynamic residual threshold adaptive selection, comprising the following steps:

[0010] Acquire remote sensing time-series image data of the target area;

[0011] A prediction algorithm based on time series modeling is used to perform trend modeling and prediction on remote sensing image data, obtain the predicted value at the corresponding time node, and calculate the pixel-by-pixel absolute residual between the predicted value and the original image value.

[0012] Based on the calculation results of the absolute residuals, the residual threshold determined by the median, interquartile range, skewness and kurtosis of the absolute residuals is selected as the division threshold to divide the time series images into normal time series images that conform to the actual change pattern and abnormal time series images that do not conform to the actual change pattern.

[0013] Based on the time-series image segmentation results, for abnormal time-series image parts that do not conform to the actual change pattern, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance to perform spatiotemporal joint filtering; for normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction; finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

[0014] Optionally, after acquiring the remote sensing time-series image data of the target area, the method may also include:

[0015] Data preprocessing operations are performed on the remote sensing time-series image data; wherein, the data preprocessing operations include radiometric correction, atmospheric correction, and coordination between different sensors;

[0016] Spatial interpolation is performed on the preprocessed remote sensing time series image data; wherein, the spatial interpolation operation includes calculating the mean of the neighborhood of the missing data area using a preset range and filling in the missing values.

[0017] Optionally, a time-series modeling-based prediction algorithm is used to perform trend modeling and prediction on the remote sensing image data, obtaining the predicted values ​​at corresponding time nodes, and calculating the pixel-by-pixel absolute residual between the predicted values ​​and the original image values, specifically including:

[0018] After extracting pixel-based time series data and normalizing the data, a long short-term memory network model is used to make predictions and obtain the prediction results. The prediction results are then denormalized, and the residuals between the prediction results and the true values ​​are calculated. The absolute value of the residuals is then taken.

[0019] Optionally, based on the calculation results of the absolute residuals, a residual threshold determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals is selected as the partitioning threshold to divide the time series images into normal time series images that conform to the actual change patterns and abnormal time series images that do not conform to the actual change patterns. Specifically, this includes:

[0020] The absolute residuals are arranged in order of magnitude. The residual thresholds determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals are used as the division thresholds to divide the time series images into time series images that conform to the actual changes and time series images that do not conform to the actual changes.

[0021] Time series images that conform to actual changes are considered normal time series images, while time series images that do not conform to actual changes are considered abnormal time series images.

[0022] Optionally, the expression for the residual threshold determined by the absolute residual median, interquartile range, skewness, and kurtosis is as follows:

[0023] ,

[0024] ,

[0025] ,

[0026] ,

[0027] In the formula, T is the residual threshold, M is the residual median, IQR is the interquartile range, Skewness is the skewness, Kurtosis is the kurtosis, α is the adjustment coefficient, β, γ, and δ are empirical coefficients, and k1 and k2 are slope parameters. Membership function representing skewness. The membership function representing kurtosis.

[0028] Optionally, based on the temporal image segmentation results, for abnormal temporal image portions that do not conform to actual change patterns, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance for spatiotemporal joint filtering; for normal temporal image regions that conform to actual change patterns, a lightweight reconstruction is performed using a temporal domain filtering algorithm, specifically including:

[0029] For the abnormal temporal image part, the current pixel to be filtered is excluded in the spatiotemporal neighborhood window processing. The weight of each pixel in the filtering operation is determined by calculating the Euclidean distance between other pixels in the neighborhood and the center pixel.

[0030] Calculate the new pixel value in the time domain and the new pixel value in the spatial domain based on the weight of the pixel in the filtering operation;

[0031] The expression for the new pixel value in the time domain is:

[0032] ,

[0033] In the formula: Let i be the new pixel value in the time domain, where i is the year of the pixel and x is the pixel value corresponding to different years.

[0034] The expression for the new pixel value in the spatial domain is:

[0035] ,

[0036] In the formula: For the new pixel value in the spatial domain, It refers to the neighboring pixels in the spatial neighborhood, excluding the center point.

[0037] Optionally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image, including:

[0038] The expression for a new pixel that does not conform to the actual change region is:

[0039] ,

[0040] In the formula: y represents the new pixel point that does not conform to the actual change region, and W represents the proportion of the pixel that participates in the spatiotemporal filtering operation in the time domain and spatial domain.

[0041] For pixels that do not conform to the actual change area, the SG algorithm is used to filter and obtain new pixels.

[0042] Furthermore, to achieve the above objectives, the present invention also provides a spatiotemporal filtering device with dynamic residual threshold adaptive selection, the device comprising:

[0043] The acquisition module is used to acquire remote sensing time-series image data of the target area;

[0044] The prediction module is used to perform trend modeling and prediction on remote sensing image data using a prediction algorithm based on time series modeling, to obtain the predicted value at the corresponding time node, and to calculate the pixel-wise absolute residual between the predicted value and the original image value.

[0045] The classification module is used to select a residual threshold determined based on the absolute residual median, interquartile range, skewness and kurtosis as the division threshold based on the calculation results of the absolute residual, and divide the time series images into normal time series images that conform to the actual change pattern and abnormal time series images that do not conform to the actual change pattern.

[0046] An adaptive filtering module is used to construct a filtering window based on spatial and temporal neighborhoods for abnormal time-series image parts that do not conform to the actual change pattern, and to determine the weight distribution according to Euclidean distance for spatiotemporal joint filtering based on the time-series image segmentation results. For normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction. Finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

[0047] Furthermore, to achieve the above objectives, the present invention also provides a spatiotemporal filtering device with dynamic residual threshold adaptive selection. The spatiotemporal filtering device with dynamic residual threshold adaptive selection includes: a memory, a processor, and a spatiotemporal filtering program with dynamic residual threshold adaptive selection stored in the memory and executable on the processor. When the spatiotemporal filtering program with dynamic residual threshold adaptive selection is executed by the processor, it implements the steps of the spatiotemporal filtering method with dynamic residual threshold adaptive selection as described above.

[0048] In addition, to achieve the above objectives, the present invention also provides a storage medium storing a spatiotemporal filtering program for adaptive selection of dynamic residual threshold, wherein the spatiotemporal filtering program for adaptive selection of dynamic residual threshold implements the steps of the above-described spatiotemporal filtering method for adaptive selection of dynamic residual threshold when executed by a processor.

[0049] The beneficial effects of this invention are as follows: It proposes a spatiotemporal filtering method, apparatus, device, and storage medium with dynamic residual threshold adaptive selection, applicable to noise suppression and reconstruction of remote sensing time-series images. The method includes acquiring remote sensing time-series image data of the target area; obtaining prediction results of the remote sensing time-series data by training a model and predicting possible changing trends of the remote sensing time-series data; then calculating the absolute residual between the predicted image value and the actual image value; and selecting a residual threshold determined based on the median, interquartile range, skewness, and kurtosis of the absolute residual as the threshold for selecting the filtering method. Specifically, for high residual areas, a spatiotemporally weighted filtering strategy combining Euclidean distance is adopted; for low residual areas, a time-domain filtering and reconstruction method is adopted, thereby suppressing unsystematic noise while ensuring image continuity. This invention adaptively completes the filtering and reconstruction of remote sensing time-series images by combining the spatiotemporal correlation of remote sensing images through a series of calculations. Compared with existing technologies, this method has a higher degree of automation and adaptability, and is particularly suitable for processing remote sensing time-series data in high-altitude, frequently clouded, or complex terrain areas. It solves the technical problems of noise pollution, low automation level, and low image accuracy in time-series images of complex mountainous areas. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention;

[0052] Figure 2 This is a flowchart illustrating the spatiotemporal filtering method for adaptive selection of dynamic residual threshold in an embodiment of the present invention.

[0053] Figure 3 This is a schematic diagram illustrating the principle of the spatiotemporal filtering method with dynamic residual threshold adaptive selection in an embodiment of the present invention.

[0054] Figure 4 This is a diagram showing the spatiotemporal filtering result after adaptive selection of the dynamic residual threshold in an embodiment of the present invention.

[0055] Figure 5 This is a schematic diagram showing the original pixel value positions involved in the spatiotemporal filtering operation in an embodiment of the present invention;

[0056] Figure 6 The peak signal-to-noise ratio result for evaluating the accuracy of spatiotemporally filtered images in this embodiment of the invention.

[0057] Figure 7 Structural similarity results diagram related to the accuracy evaluation of spatiotemporally filtered images in this embodiment of the invention.

[0058] Figure 8 This is a structural block diagram of a spatiotemporal filtering device for dynamic residual threshold adaptive selection in an embodiment of the present invention. Detailed Implementation

[0059] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. 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 are within the scope of protection of the present invention.

[0060] like Figure 1 As shown, Figure 1 This is a schematic diagram of the device structure of the hardware operating environment involved in the embodiments of the present invention.

[0061] like Figure 1 As shown, the device may include: a processor 1001, such as a CPU; a communication bus 1002; a user interface 1003; a network interface 1004; and a memory 1005. The communication bus 1002 is used to enable communication between these components. The user interface 1003 may include a display screen or an input unit such as a keyboard; optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a Wi-Fi interface). The memory 1005 may be high-speed RAM or non-volatile memory, such as a disk drive. Optionally, the memory 1005 may also be a storage device independent of the aforementioned processor 1001.

[0062] Those skilled in the art will understand that Figure 1 The structure of the device shown does not constitute a limitation on the device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0063] like Figure 1 As shown, the memory 1005, which serves as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a spatiotemporal filtering program that adaptively selects dynamic residual thresholds.

[0064] exist Figure 1In the terminal shown, network interface 1004 is mainly used to connect to the backend server and communicate with it; user interface 1003 is mainly used to connect to the client (user terminal) and communicate with it; while processor 1001 can be used to call the spatiotemporal filtering program for dynamic residual threshold adaptive selection stored in memory 1005 and perform the following operations:

[0065] Acquire remote sensing time-series image data of the target area;

[0066] A prediction algorithm based on time series modeling is used to perform trend modeling and prediction on remote sensing image data, obtain the predicted value at the corresponding time node, and calculate the pixel-by-pixel absolute residual between the predicted value and the original image value.

[0067] Based on the calculation results of the absolute residuals, the residual threshold determined by the median, interquartile range, skewness and kurtosis of the absolute residuals is selected as the division threshold to divide the time series images into normal time series images that conform to the actual change pattern and abnormal time series images that do not conform to the actual change pattern.

[0068] Based on the time-series image segmentation results, for abnormal time-series image parts that do not conform to the actual change pattern, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance to perform spatiotemporal joint filtering; for normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction; finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

[0069] The specific embodiments of the present invention applied to the device are basically the same as the embodiments of the spatiotemporal filtering method using dynamic residual threshold adaptive selection described below, and will not be repeated here.

[0070] This invention provides a spatiotemporal filtering method with dynamic residual threshold adaptive selection, referring to... Figure 2 , Figure 2 This is a flowchart illustrating an embodiment of the spatiotemporal filtering method for adaptive selection of dynamic residual thresholds according to the present invention.

[0071] In this embodiment, the spatiotemporal filtering method with adaptive selection of dynamic residual threshold includes the following steps:

[0072] Step S100: Obtain remote sensing time series image data of the target area.

[0073] Specifically, the first step is to acquire time-series optical remote sensing imagery and digital elevation data for the target area. Time-series data, as the name suggests, consists of long-term imagery sequences spaced at regular intervals. The time-series optical remote sensing imagery undergoes preprocessing, including radiometric calibration, atmospheric correction, and coordination between different sensors.

[0074] Following this, spatial interpolation is performed on the preprocessed optical remote sensing time-series image data; wherein, the spatial interpolation operation includes calculating the mean value of a certain range of neighborhoods for missing data regions and filling in the missing values.

[0075] In practical applications, taking the spatiotemporal filtering of dynamic residual threshold adaptive selection in Shangri-La City as an example, when acquiring remote sensing data, the Google Earth engine can be used to filter Landsat remote sensing image data of high quality for each year from 1987 to 2017 using the strip number, and the filtered images can be composited using median. The images are concentrated from January to March, with cloud coverage of less than 10% and a spatial resolution of 30 meters.

[0076] The usage of Landsat remote sensing image data for the target area is shown in the table below:

[0077]

[0078] In the preprocessing of remote sensing images, GEE was first used to coordinate different types of sensors from different years to address the varying spectral, spatial, and radiometric resolutions among the Landsat series sensors. Then, cloud removal was performed on the images. Since the study area involved three images, high-quality image sets selected from each year were combined using a median algorithm to create a comprehensive image covering Shangri-La City. Finally, the images from each year were merged to form a time-series dataset with yearly intervals.

[0079] In this embodiment, after acquiring the remote sensing time series images, it is also necessary to perform interpolation operations on the remote sensing time series images.

[0080] Specifically, to address the issue of missing pixel values ​​in the data, spatial interpolation was performed using information from surrounding pixels. In practical applications, by spatially interpolating Landsat time-series data from 1987 to 2017 and using the mean value of pixels within a 21×21 neighborhood window to fill in the missing data, the completeness of remote sensing time-series data can be significantly improved.

[0081] Step S200: A prediction algorithm based on time series modeling is used to perform trend modeling and prediction on remote sensing image data to obtain the predicted value at the corresponding time node, and the pixel-wise absolute residual between the predicted value and the original image value is calculated.

[0082] Specifically, the Long Short-Term Memory (LSTM) network model is used to capture the hidden nonlinear and long-term dependency features in remote sensing time series data and generate possible reference trends for each time series containing 31 years of observation data.

[0083] In practical applications, a sliding time window method is used to construct the data for model predictions, with five consecutive years of data as input features. A long short-term memory network model is then trained to predict pixel values ​​for the following year.

[0084] Step S300: Based on the calculation results of the absolute residuals, select a residual threshold determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals as the division threshold, and divide the time series images into normal time series images that conform to the actual change pattern and abnormal time series images that do not conform to the actual change pattern.

[0085] Specifically, in this embodiment, the step of dividing the time series image is based on a residual threshold determined by the absolute residual median, interquartile range, skewness, and kurtosis as the dividing threshold.

[0086] The expression for the residual threshold determined by the absolute residual median, interquartile range, skewness, and kurtosis is as follows:

[0087] ,

[0088] ,

[0089] ,

[0090] ,

[0091] In the formula, T is the residual threshold, M is the residual median, IQR is the interquartile range, Skewness is the skewness, Kurtosis is the kurtosis, α is the adjustment coefficient, β, γ, and δ are empirical coefficients, and k1 and k2 are slope parameters. Membership function representing skewness. The membership function representing kurtosis.

[0092] Among them, when the skewness is > 0.5, it is a right-skewed distribution, and when the kurtosis is > 3, it is a leptokurtic distribution.

[0093] It should be noted that for the adjustment coefficient α, IQR is used to reflect the dispersion of the residual data. The larger the IQR, the wider the range of data fluctuation. M represents the central tendency of the residual data and is used to measure the overall level of the data. β controls the adjustment of IQR on α and needs to be set according to the data characteristics (usually positive).

[0094] Therefore, this invention quantifies the relative dispersion of residuals by using the ratio of IQR to M (IQR / M). When IQR is large, it indicates that the residual distribution is more dispersed, and α will increase with the increase of this ratio, thereby raising the overall threshold and avoiding misjudgments caused by strong dispersion (such as misjudging reasonable fluctuations as abnormal), achieving the goal of dynamically adapting to data dispersion. The median M is not sensitive to extreme values, while IQR can supplement the reflection of the data's fluctuation range. By combining the two, α allows the threshold calculation to take into account both central tendency and dispersion characteristics, improving adaptability to complex data distributions and achieving the goal of balancing the limitations of the median.

[0095] Among them, membership degree and It varies continuously between 0 and 1. When Skewness = 0.5 or Kurtosis = 3, and The threshold is set to 0.5, and the adjustment term gradually changes with the distribution characteristics to avoid sudden threshold changes when conditions are triggered. Therefore, this invention employs a fuzzy weighted model based on the absolute residual median, interquartile range, skewness, and kurtosis. By simultaneously considering features such as skewness (distribution symmetry), kurtosis (sharpness), and IQR (dispersion), the threshold can be dynamically adjusted according to the actual data distribution. Furthermore, considering that mountain remote sensing data is affected by terrain and weather, the residual distribution may exhibit both skewness and dispersion characteristics. The fuzzy weighted model can automatically adjust the weights by learning from historical data; for example, increasing the weights of IQR and kurtosis in cloudy and foggy mountainous areas improves the filter's adaptability to extreme weather noise.

[0096] In practical applications, the image quality of high-altitude and complex mountainous areas is poor and the frequency of noise is high. Based on the noise resistance characteristics of the median, the residual threshold determined by the absolute residual median, interquartile range, skewness and kurtosis is selected as the segmentation threshold for image segmentation.

[0097] Step S400: Based on the time-series image segmentation results, for abnormal time-series image parts that do not conform to the actual change pattern, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance to perform spatiotemporal joint filtering; for normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction; finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

[0098] Specifically, time-series data acquired in early spring at high altitudes often contain numerous sources of noise, which typically exhibits anomalous variations. Adaptive filtering based on absolute residuals, by selecting different filtering methods for different time segments, can selectively suppress noise signals, thereby improving image quality and clarity.

[0099] In this embodiment, for the abnormal temporal image portion, the current pixel to be filtered is excluded during the spatiotemporal neighborhood window processing. The weight of each pixel in the filtering operation is determined by calculating the Euclidean distance between other pixels in the neighborhood and the center pixel. The expression for the new pixel value in the temporal domain is:

[0100] ,

[0101] In the formula: Let i be the new pixel value in the time domain, where i is the year of the pixel and x is the pixel value corresponding to different years.

[0102] The expression for the new pixel value in the spatial domain is:

[0103] ,

[0104] In the formula: For the new pixel value in the spatial domain, These are the neighboring pixels within the spatial neighborhood, excluding the center point. The original pixel positions are as follows: Figure 5 As shown.

[0105] The expression for the new pixel that does not conform to the actual change area is:

[0106] ,

[0107] y is the new pixel value of the filtered pixel, Wi is the weight of the pixel value in the time domain, and Wj is the weight of the pixel value in the spatial domain.

[0108] When the image represents a region that reflects actual changes, the SG algorithm with a smaller window is used for filtering.

[0109] By combining the results of the two filters, adaptive filtering of remote sensing time series images is completed.

[0110] After filtering and reconstructing the remote sensing time series images, the root mean square error, peak signal-to-noise ratio, and structural similarity are used to evaluate the image quality. The expression for the image quality evaluation index is as follows:

[0111] ,

[0112] ,

[0113] ,

[0114] ,

[0115] ,

[0116] ,

[0117] ,

[0118] In the formula: These are the filtered image values. The average value of the image before filtering. is the number of samples. f is the filtered image, o is the unfiltered image, (x,y) is the pixel position, W is the image width, and H is the image height. This represents the average value of the image before and after filtering. This represents the covariance of the images before and after filtering. represents the standard deviation of the image before and after filtering.

[0119] Specifically, the above filtering method was used in Python to filter the time series imagery of Shangri-La from 1987 to 2017, and the accuracy of the filtering results was evaluated. The specific calculation results of the root mean square error are shown in the table below:

[0120] Bands Max Min Mean Blue 8268.11 0.00 95.88 Green 7665.23 0.00 94.76 Red 10479.10 0.00 104.97 NIR 6819.18 0.00 254.63 SWIR1 9336.76 0.00 206.24 SWIR2 4115.19 0.00 146.25

[0121] The specific calculation results of peak signal-to-noise ratio are as follows: Figure 6 As shown, the structural similarity calculation results are as follows: Figure 7 As shown.

[0122] In this embodiment, as Figure 3-4 As shown, a spatiotemporal filtering method with dynamic residual threshold adaptive selection is provided. This method acquires remote sensing time-series image data of the target area; constructs and trains a time-series prediction model; uses the model to predict the acquired image data; and calculates the absolute residual between the predicted result and the true value. Based on the calculated absolute residual, a residual threshold determined by the median, interquartile range, skewness, and kurtosis of the absolute residual is selected as the partitioning threshold, dividing the time-series image into normal time-series images that conform to actual change patterns and abnormal time-series images that do not conform to actual change patterns. Based on the partitioning results, for the abnormal time-series image portion, the pixel weights participating in the filtering operation are determined by calculating the Euclidean distance between pixels within the spatiotemporal neighborhood window, and filtering is performed accordingly. For the normal time-series image portion, a temporal domain algorithm is used for filtering. Finally, the two filtering results are merged to complete the filtering and reconstruction of the remote sensing time-series image. This invention, by combining the spatiotemporal correlation of remote sensing images and through a series of calculations, adaptively completes the filtering and reconstruction of remote sensing time-series images. It solves the technical problems of noise pollution, low automation, and low image accuracy in time-series images under complex mountainous terrain.

[0123] Reference Figure 8 , Figure 8This is a structural block diagram of an embodiment of the spatiotemporal filtering device for adaptive selection of dynamic residual threshold according to the present invention.

[0124] like Figure 8 As shown, the spatiotemporal filtering device with dynamic residual threshold adaptive selection proposed in this embodiment of the invention includes:

[0125] The acquisition module 10 is used to acquire remote sensing time-series image data of the target area;

[0126] The prediction module 20 is used to perform trend modeling and prediction on remote sensing image data using a prediction algorithm based on time series modeling, to obtain the predicted value at the corresponding time node, and to calculate the pixel-wise absolute residual between the predicted value and the original image value.

[0127] The classification module 30 is used to select a residual threshold determined based on the absolute residual median, interquartile range, skewness and kurtosis as the division threshold based on the calculation result of the absolute residual, and divide the time series image into normal time series images that conform to the actual change pattern and abnormal time series images that do not conform to the actual change pattern.

[0128] The adaptive filtering module 40 is used to construct a filtering window based on spatial and temporal neighborhoods for abnormal time-series image parts that do not conform to the actual change pattern, based on the time-series image segmentation results, and to determine the weight distribution according to Euclidean distance for spatiotemporal joint filtering; for normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction; finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

[0129] Other embodiments or specific implementations of the spatiotemporal filtering device for adaptive selection of dynamic residual threshold of the present invention can be found in the above-described method embodiments, and will not be repeated here.

[0130] Furthermore, the present invention also proposes a spatiotemporal filtering device for adaptive selection of dynamic residual threshold. The spatiotemporal filtering device for adaptive selection of dynamic residual threshold includes: a memory, a processor, and a spatiotemporal filtering program for adaptive selection of dynamic residual threshold stored in the memory and executable on the processor. When the spatiotemporal filtering program for adaptive selection of dynamic residual threshold is executed by the processor, it implements the steps of the spatiotemporal filtering method for adaptive selection of dynamic residual threshold as described above.

[0131] The specific implementation of the spatiotemporal filtering device with dynamic residual threshold adaptive selection in this application is basically the same as the various embodiments of the spatiotemporal filtering method with dynamic residual threshold adaptive selection described above, and will not be repeated here.

[0132] Furthermore, this invention also proposes a readable storage medium, comprising a computer-readable storage medium storing a spatiotemporal filtering program for adaptive selection of dynamic residual thresholds. The readable storage medium may be... Figure 1 The memory 1005 in the terminal can also be at least one of ROM (Read-Only Memory) / RAM (Random Access Memory), magnetic disk, optical disk, etc. The readable storage medium includes several instructions to cause a spatiotemporal filtering device with a processor to perform the spatiotemporal filtering method of dynamic residual threshold adaptive selection as described in various embodiments of the present invention.

[0133] The specific implementation in the readable storage medium of this application is basically the same as the various embodiments of the spatiotemporal filtering method with adaptive selection of dynamic residual threshold described above, and will not be repeated here.

[0134] It is understood that in the description of this specification, references to terms such as "one embodiment," "another embodiment," "other embodiments," or "first embodiment to Nth embodiment," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the present invention. In this specification, 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.

[0135] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0136] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0137] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, or network device, etc.) to execute the methods described in the various embodiments of the present invention.

[0138] Finally, it should be noted that the above embodiments are only used to illustrate and not limit the technical solutions of the present invention. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the present invention without departing from the spirit and scope of the present invention. Any modifications or partial substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A spatiotemporal filtering method with dynamic residual threshold adaptive selection, characterized in that, Includes the following steps: Acquire remote sensing time-series image data of the target area; A prediction algorithm based on time series modeling is used to perform trend modeling and prediction on remote sensing image data, obtain the predicted value at the corresponding time node, and calculate the pixel-by-pixel absolute residual between the predicted value and the original image value. Based on the calculation results of the absolute residuals, a residual threshold determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals is selected as the partitioning threshold. This divides the time-series images into normal time-series images that conform to actual change patterns and abnormal time-series images that do not conform to actual change patterns. Specifically, this includes: The absolute residuals are arranged in order of magnitude. Based on the residual thresholds determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals, the time series images are divided into time series images that conform to the actual changes and time series images that do not conform to the actual changes. The expression for the residual threshold determined by the absolute residual median, interquartile range, skewness, and kurtosis is as follows: 、 、 、 、 In the formula, T is the residual threshold, M is the residual median, IQR is the interquartile range, Skewness is the skewness, Kurtosis is the kurtosis, α is the adjustment coefficient, β, γ, and δ are empirical coefficients, and k1 and k2 are slope parameters. Membership function representing skewness. Membership function representing kurtosis; Time series images that conform to actual changes are considered normal time series images, while time series images that do not conform to actual changes are considered abnormal time series images. Based on the time-series image segmentation results, for abnormal time-series image parts that do not conform to the actual change pattern, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance to perform spatiotemporal joint filtering; for normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction; finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

2. The spatiotemporal filtering method with dynamic residual threshold adaptive selection as described in claim 1, characterized in that, After acquiring remote sensing time-series image data of the target area, the following steps are also included: Data preprocessing operations are performed on the remote sensing time-series image data; wherein, the data preprocessing operations include radiometric correction, atmospheric correction, and coordination between different sensors; Spatial interpolation is performed on the preprocessed remote sensing time series image data; wherein, the spatial interpolation operation includes calculating the mean of the neighborhood of the missing data area using a preset range and filling in the missing values.

3. The spatiotemporal filtering method with dynamic residual threshold adaptive selection as described in claim 1, characterized in that, A time-series modeling-based prediction algorithm is used to perform trend modeling and prediction on remote sensing image data, obtaining predicted values ​​for corresponding time nodes, and calculating the pixel-by-pixel absolute residual between the predicted values ​​and the original image values, specifically including: After extracting pixel-based time series data and normalizing the data, a long short-term memory network model is used to make predictions and obtain the prediction results. The prediction results are then denormalized, and the residuals between the prediction results and the true values ​​are calculated. The absolute value of the residuals is then taken.

4. The spatiotemporal filtering method with dynamic residual threshold adaptive selection as described in claim 1, characterized in that, Based on the temporal image segmentation results, for abnormal temporal image parts that do not conform to the actual change pattern, a filtering window based on spatial and temporal neighborhoods is constructed, and the weight distribution is determined according to Euclidean distance to perform spatiotemporal joint filtering. For normal time-series image regions that conform to actual change patterns, a lightweight reconstruction is performed using a time-domain filtering algorithm, specifically including: For the abnormal temporal image part, the current pixel to be filtered is excluded in the spatiotemporal neighborhood window processing. The weight of each pixel in the filtering operation is determined by calculating the Euclidean distance between other pixels in the neighborhood and the center pixel. Calculate the new pixel value in the time domain and the new pixel value in the spatial domain based on the weight of the pixel in the filtering operation; The expression for the new pixel value in the time domain is: 、 In the formula: Let i be the new pixel value in the time domain, where i is the year of the pixel and x is the pixel value corresponding to different years. The expression for the new pixel value in the spatial domain is: 、 In the formula: For the new pixel value in the spatial domain, It refers to the neighboring pixels in the spatial neighborhood, excluding the center point.

5. The spatiotemporal filtering method with dynamic residual threshold adaptive selection as described in claim 1, characterized in that, By fusing the results of the two types of filtering, the reconstructed time-series remote sensing imagery is obtained, including: The expression for a new pixel that does not conform to the actual change region is: 、 In the formula: y represents the new pixel point that does not conform to the actual change area, and W represents the proportion of the pixel that participates in the spatiotemporal filtering operation in the time domain and spatial domain; For pixels that do not conform to the actual change area, the SG algorithm is used to filter and obtain new pixels.

6. A spatiotemporal filtering device with dynamic residual threshold adaptive selection, characterized in that, The device includes: The acquisition module is used to acquire remote sensing time-series image data of the target area; The prediction module is used to perform trend modeling and prediction on remote sensing image data using a prediction algorithm based on time series modeling, to obtain the predicted value at the corresponding time node, and to calculate the pixel-wise absolute residual between the predicted value and the original image value. The classification module, based on the calculation results of the absolute residuals, selects a residual threshold determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals as a division threshold to classify the time-series images into normal time-series images that conform to the actual change patterns and abnormal time-series images that do not conform to the actual change patterns. Specifically, it includes: The absolute residuals are arranged in order of magnitude. Based on the residual thresholds determined by the median, interquartile range, skewness, and kurtosis of the absolute residuals, the time series images are divided into time series images that conform to the actual changes and time series images that do not conform to the actual changes. The expression for the residual threshold determined by the absolute residual median, interquartile range, skewness, and kurtosis is as follows: 、 、 、 、 In the formula, T is the residual threshold, M is the residual median, IQR is the interquartile range, Skewness is the skewness, Kurtosis is the kurtosis, α is the adjustment coefficient, β, γ, and δ are empirical coefficients, and k1 and k2 are slope parameters. Membership function representing skewness. Membership function representing kurtosis; Time series images that conform to actual changes are considered normal time series images, while time series images that do not conform to actual changes are considered abnormal time series images. An adaptive filtering module is used to construct a filtering window based on spatial and temporal neighborhoods for abnormal time-series image parts that do not conform to the actual change pattern, and to determine the weight distribution according to Euclidean distance for spatiotemporal joint filtering based on the time-series image segmentation results. For normal time-series image areas that conform to the actual change pattern, a time-domain filtering algorithm is used for lightweight reconstruction. Finally, the two types of filtering results are fused to obtain the reconstructed time-series remote sensing image.

7. A spatiotemporal filtering device with dynamic residual threshold adaptive selection, characterized in that, The spatiotemporal filtering device for adaptive selection of dynamic residual thresholds includes: a memory, a processor, and a spatiotemporal filtering program for adaptive selection of dynamic residual thresholds stored in the memory and executable on the processor. When the spatiotemporal filtering program for adaptive selection of dynamic residual thresholds is executed by the processor, it implements the steps of the spatiotemporal filtering method for adaptive selection of dynamic residual thresholds as described in any one of claims 1 to 5.

8. A storage medium, characterized in that, The storage medium stores a spatiotemporal filtering program for adaptive selection of dynamic residual thresholds. When the spatiotemporal filtering program for adaptive selection of dynamic residual thresholds is executed by the processor, it implements the steps of the spatiotemporal filtering method for adaptive selection of dynamic residual thresholds as described in any one of claims 1 to 5.

Images