Vegetation coverage time sequence processing method and system

Abstract

The invention relates to a vegetation coverage time sequence processing method, which comprises the following steps of: inputting a time sequence remote sensing image and suburb forest and economic forest classification data of the same region; calculating a normalized differential vegetation index of the research area and carrying out time sequence sorting; calculating vegetation coverage data of the research area and carrying out time sequence sorting; synthesizing to obtain monthly FVC time sequence data of the suburb forest region and the economic forest region; adaptive filtering is carried out on the suburb forest time sequence data; carrying out seasonal decomposition based on a fixed period on the filtered suburb forest time sequence data; filtering the economic forest time series data; carrying out adaptive seasonal decomposition on the filtered economic forest time series data; combining the suburb forest time series data and the economic forest time series data obtained through decomposition; and outputting the merged time series data to obtain final optimized data. The invention also relates to a vegetation coverage time sequence processing system. According to the method, a reliable data basis can be provided for high-precision change detection and attribution analysis.

Description

Technical Field

[0001] This invention relates to a method and system for time-series processing of vegetation cover. Background Technology

[0002] In ecological environment monitoring and land use research, foliar vegetation cover (FVC) derived from remote sensing time-series data is a key indicator for characterizing vegetation dynamics and assessing ecosystem functions and services. Shenzhen, as a highly urbanized and ecologically sensitive area, has a complex green space system, typically represented by evergreen broad-leaved suburban forests dominated by ecological functions and intensive orchards (such as litchi forests) dominated by economic functions. Although these two types of vegetation share some similarities in spectral characteristics, their temporal changes are driven by both natural phenological rhythms and human activities, exhibiting distinctly different dynamic mechanisms. Therefore, accurately analyzing the inherent patterns of their FVC time-series curves is of great significance for understanding the evolution of urban ecosystems, assessing the impact of human activities, and guiding sustainable land use.

[0003] Current research on time-series analysis of remote sensing vegetation largely focuses on employing standardized preprocessing procedures, such as Savitzky-Golay filtering, Whittaker smoothing, or fixed-period-based seasonal and trend decomposition methods (STL, BFAST, etc.), to suppress noise and extract trend and periodic components. These methods typically assume that the vegetation throughout the study area follows similar change patterns or are optimized only for a single vegetation type.

[0004] Current technologies for vegetation cover (FVC) analysis based on time-series remote sensing data generally employ uniform preprocessing methods, failing to fully consider the fundamental differences in phenological rhythms and anthropogenic disturbance patterns among different vegetation function types (such as natural forests primarily for ecological functions and economic forests primarily for management functions). Especially in rapidly urbanizing subtropical areas like Shenzhen, the spectral temporal characteristics of evergreen broad-leaved suburban forests and litchi orchards are influenced by a combination of intrinsic phenological rhythms and exogenous disturbances from human management activities, resulting in significantly different FVC time-series curves exhibiting distinct change mechanisms. Directly using general filtering and decomposition methods obscures their unique trends and abrupt change signals, leading to decreased accuracy in subsequent time-series breakpoint detection (such as the DBEST algorithm) and difficulty in accurately identifying the driving factors of vegetation dynamics.

[0005] In summary, the core flaw of existing technical solutions lies in their "one-size-fits-all" preprocessing paradigm, which ignores the fundamental differences in the temporal change driving mechanisms of different vegetation functional types. Specifically: 1) Fixed parameters of general filtering methods: The widely used SG filter or Whittaker smoother usually uses globally uniform parameters (such as window size and smoothing intensity), which makes it difficult to simultaneously adapt to the shape preservation requirements of smoothing time series of evergreen forests and the strong denoising requirements of high-frequency disturbance sequences of economic forests, which can easily lead to the loss of details in the former or the noise residue in the latter.

[0006] 2) The seasonal decomposition model assumes a single assumption: mainstream decomposition methods (such as STL) assume a fixed annual cycle, which cannot effectively characterize the non-standard (such as semi-annual, bi-peak) phenological cycles that may be caused by human management in litchi orchards (such as promoting flowering and harvesting), resulting in the estimation bias of seasonal components and the misincorporation of some management signals into the residuals.

[0007] 3) Lack of knowledge guidance in outlier handling: Conventional statistical outlier detection (such as the 3σ rule) cannot distinguish between "true anomalies" (signals that need to be retained) caused by agricultural activities and "false anomalies" (noise that needs to be removed) caused by noise in economic forest sequences, which may lead to the incorrect removal of important management event signals. These shortcomings together result in distortion of the preprocessed time series signal, limited improvement in signal-to-noise ratio, and consequently reduce the accuracy and reliability of subsequent advanced analyses (such as breakpoint detection and driving force analysis). Summary of the Invention

[0008] In view of this, it is necessary to provide a vegetation cover time series processing method and system that can solve the problem that the unified preprocessing model cannot adapt to the essential differences in the time series characteristics of evergreen broad-leaved suburban forests and economic forests, and can generate purified FVC sequences that clearly reflect their respective "time series skeletons", providing reliable input for subsequent high-precision breakpoint detection algorithms such as DBEST.

[0009] This invention provides a method for time-series processing of vegetation cover, comprising the following steps: S1, inputting time-series remote sensing images of the same region, and inputting classification data of suburban forests and economic forests; S2, calculating the Normalized Difference Vegetation Index (NDVI) of the study area based on the input time-series remote sensing images and sorting it temporally; S3, calculating the vegetation cover data (FVC) of the study area using the time-series NDVI data and sorting it temporally; S4, synthesizing monthly FVC time-series data of suburban forest areas and economic forest areas based on the input classification data of suburban forests and economic forests; S5, performing adaptive filtering on the suburban forest time-series data; S6, performing seasonal decomposition based on a fixed period on the filtered suburban forest time-series data; S7, filtering on the economic forest time-series data; S8, performing phenological-guided adaptive seasonal decomposition on the filtered economic forest time-series data; S9, merging the suburban forest time-series data obtained in step S6 and the economic forest time-series data obtained in step S8; S10, outputting the merged time-series data to obtain the final optimized data.

[0010] Preferably, step S2 includes: The Normalized Differential Vegetation Index (NDVI) of the study area was calculated using the input time-series Sentinel-2 data and then sorted by time series. The calculation formula is as follows:

[0011] in, Represents the normalized difference vegetation index; This represents the brightness value of the red band in Sentinel-2 data; This represents the brightness value of the near-infrared spectral band in the Sentinel-2 data.

[0012] Preferably, step S3 includes: The vegetation cover (FVC) data of the study area was calculated using time-series NDVI data and then sorted temporally. The calculation formula is as follows:

[0013] in, Represents a pixel with absolutely no vegetation cover. value; Represents pixels completely covered by vegetation value.

[0014] Preferably, step S5 includes: For suburban forest time series data Perform adaptive Savitzky-Golay filtering:

[0015] in, The normalization coefficient is the sum of all convolution coefficients within the window. , For the first The convolution coefficients of a point are determined by the order of the polynomial fitting; these coefficients can be generated by specifying the window length and the order. It is half the width of a window.

[0016] Preferably, step S6 includes: For suburban forest time series data A seasonal decomposition based on a fixed period is performed, and the decomposition algorithm is as follows:

[0017] in, It is a trend item; It is a seasonal item; It is the residual term.

[0018] Preferably, step S7 includes: Regarding time series data of economic forests Perform Whittaker filtering; the filtering algorithm is as follows:

[0019] in, To find the fitted value, minimize the objective function. get; For smoothing parameters; It is a d-order difference operator.

[0020] Preferably, step S8 includes: Step S81: Detect the dominant cycle using continuous small transformations. If there is a significant non-12-month cycle, adjust the decomposition basis according to the detection cycle. Step S82: Use BFAST-like methods for iterative decomposition. The decomposition algorithm is as follows:

[0021] in, It is a trend item; It is a seasonal item; It is the residual term; the BFAST iterative decomposition initializes the seasonal components with a period T. Iterate until convergence; Step S83, combine the agricultural calendar to analyze the residual items. The significant drop in the value was identified.

[0022] Preferably, step S9 includes: Merge the two preprocessed groups Timing: .

[0023] This invention provides a time-series vegetation cover processing system. The system includes an input module, a sorting module, a synthesis module, a filtering module, a decomposition module, a merging module, and an output module. The input module is used to input time-series remote sensing images of the same region and classification data of suburban forests and economic forests. The sorting module is used to calculate the Normalized Difference Vegetation Index (NDVI) of the study area based on the input time-series remote sensing images and perform time-series sorting. The sorting module is also used to calculate the vegetation cover data (FVC) of the study area using the time-series NDVI data and perform time-series sorting. The synthesis module is used to synthesize data based on the input suburban forest and economic forest classification data. Monthly FVC time-series data for suburban forest areas and economic forest areas are obtained. The filtering module is used to perform adaptive filtering on the suburban forest time-series data. The decomposition module is used to perform seasonal decomposition based on a fixed period on the filtered suburban forest time-series data. The filtering module is also used to filter the economic forest time-series data. The decomposition module is also used to perform phenological-guided adaptive seasonal decomposition on the filtered economic forest time-series data. The merging module is used to merge the suburban forest time-series data and economic forest time-series data obtained by the decomposition module. The output module is used to output the merged time-series data to obtain the final optimized data.

[0024] This application ultimately outputs purified FVC sequences that clearly characterize their long-term trends, abrupt changes, and gradual transitions. The preprocessed sequences not only suppress noise to the maximum extent but also preserve and enhance the unique temporal patterns derived from natural phenology and human activities, greatly improving the sensitivity and accuracy of subsequent breakpoint detection algorithms such as DBEST in identifying the time, type, and driving force of changes. This represents a paradigm shift from "single model fitting" to "precise classification processing," providing a reliable and innovative technical approach for the refined dynamic monitoring of heterogeneous vegetation systems in urban areas. This invention has the following beneficial effects: (1) This invention achieves a paradigm shift from "universal processing" to "precise adaptation," greatly improving the fidelity of time-series signals. Existing solutions ignore the fundamental differences in the time-series generation mechanisms between natural forests and economic forests, using the same set of parameters and processing logic, inevitably leading to signal distortion or information loss. This invention proposes for the first time a differentiated preprocessing paradigm based on vegetation functional types, constructing independent and highly optimized processing chains for natural forests and economic forests respectively. This allows the high-entropy natural phenological details of suburban forests to be fully preserved, while clearly separating the strong anthropogenic disturbance signals of economic forests, thus ensuring the quality and reliability of the data required for subsequent analysis from the source, and solving the fundamental defect of the universal model being "cut to fit the shoe."

[0025] (2) A multi-source information fusion and adaptive mechanism is introduced, solving the core problem of identifying non-standard phenology and useful signals. Existing schemes mostly rely on fixed periods for seasonal decomposition, and outlier handling relies solely on statistical methods, which cannot cope with complex real-world scenarios. This invention introduces two core technologies for economic forests: First, it uses continuous wavelet transform (CWT) for adaptive detection of the dominant period, accurately capturing non-standard (e.g., half-yearly) phenological periods caused by human management (e.g., double-cropping), making the extraction of seasonal components more consistent with physical reality; Second, it innovatively integrates external prior knowledge (agricultural calendar) into the residual cleaning process, achieving "discriminative" processing of outlier signals (i.e., retaining real management events and eliminating noise), rather than simply "one-size-fits-all" elimination. This effectively avoids the risk of misjudging important abrupt change signals as noise, which is something existing purely data-driven schemes cannot achieve.

[0026] (3) It significantly improves the accuracy and interpretability of advanced time series analysis products (such as breakpoint detection), and has important application value. This invention is not an isolated data processing step; its ultimate goal is to provide higher-quality input for advanced breakpoint detection algorithms such as DBEST. After differential preprocessing by this invention, the long-term trend, seasonal cycle, and residual components of the FVC sequence are more clearly separated, enabling the DBEST algorithm to more sensitively identify true gradual or abrupt changes and significantly reduce false positives and false negatives caused by noise and mixing effects in previous data. Ultimately, this makes the attribution analysis of the driving factors of vegetation dynamics (e.g., distinguishing between natural disasters and human logging) more accurate and reliable, providing unprecedented decision support capabilities for the refined management of urban ecosystems.

[0027] In summary, this invention is not a simple improvement on existing technologies, but rather a novel, object-oriented preprocessing framework for remote sensing time series analysis established through conceptual innovation and methodological integration. Its technological advancement and application effectiveness are significantly superior to traditional solutions. Attached Figure Description

[0028] Figure 1 This is a flowchart of the vegetation coverage time-series processing method of the present invention; Figure 2 This is a hardware architecture diagram of the vegetation coverage time-series processing system of the present invention. Detailed Implementation

[0029] The present invention will now be described in further detail with reference to the accompanying drawings and specific embodiments.

[0030] See Figure 1 The diagram shown is a flowchart of a preferred embodiment of the vegetation coverage time-series processing method of the present invention.

[0031] Step S1: Input time-series remote sensing images of the same area, and input classification data for suburban forests and economic forests. Specifically: In this embodiment, time-series Sentinel-2 remote sensing images of the same region are input, covering at least 5 complete years in time and the spatial range covers the entire study area; classification data of suburban forests and economic forests are input, with the spatial range covering the entire study area.

[0032] Step S2: Based on the input temporal remote sensing images, calculate the Normalized Differential Vegetation Index (NDVI) for the study area and perform temporal sorting. Specifically: In this embodiment, the Normalized Differential Vegetation Index (NDVI) of the study area is calculated and sorted temporally using the input time-series Sentinel-2 data. The calculation formula is as follows:

[0033] in, Represents the normalized difference vegetation index; This represents the brightness value of the red band in Sentinel-2 data; This represents the brightness value of the near-infrared spectral band in the Sentinel-2 data.

[0034] Step S3: Calculate the vegetation cover (FVC) data for the study area using time-series NDVI data and perform time-series sorting. Specifically: In this embodiment, vegetation cover data (FVC) of the study area is calculated using time-series NDVI data and then sorted temporally. The calculation formula is as follows:

[0035] in, Represents a pixel with absolutely no vegetation cover. value; Represents pixels completely covered by vegetation value.

[0036] Step S4: Based on the input classification data of suburban forests and economic forests, synthesize the monthly FVC time-series data for the suburban forest area and the economic forest area. Specifically: In this embodiment, monthly FVC time-series data for the suburban forest and economic forest regions are synthesized monthly using the input classification data of suburban forest and economic forest regions. .

[0037] Step S5 involves adaptive filtering of the suburban forest time-series data. Specifically: In this embodiment, the time series data of suburban forests are used. Adaptive Savitzky-Golay filtering (hereinafter referred to as SG filtering) is performed, and the filtering algorithm is as follows:

[0038] in, The normalization coefficient is the sum of all convolution coefficients within the window (i.e., ...). In actual calculations, no explicit settings are required because the convolution kernel is automatically normalized. For the first The convolution coefficients of the points are determined by the order of the polynomial fitting (usually 2 to 3). The coefficients can be generated by specifying the window length and the order. With a window width of half, this method allows for dynamic adjustment of the window size based on the sequence noise level. The window length can be adjusted from 5 to 9. Users can evaluate the results through trial and error. Users can calculate the ratio of the root mean square error (RMSE) of the filtered sequence to the standard deviation of the original sequence. If the ratio is greater than 0.8, the window can be appropriately increased to 7 or even 9.

[0039] Step S6 involves performing a seasonal decomposition based on a fixed period on the filtered suburban forest time-series data. Specifically: In this embodiment, the time series data of suburban forests are used. A seasonal decomposition based on a fixed period is performed, and the decomposition algorithm is as follows:

[0040] in, It is a trend item; It is a seasonal item; These are residual terms, and the calculation method for each term includes: Step S61, use LOESS to fit a trend line to determine the trend term. To calculate LOESS, a smoothing parameter `span` needs to be set (the local window span, typically 0.2-0.3 for medium- to long-term trends). For a sequence length N, the trend value... The weighted polynomial fit value (weights are a three-kernel function) is the spanN data points near point t. The fitting algorithm is as follows:

[0041]

[0042] Where h is the width of the local window (h = spanN). This is the kernel function.

[0043] Step S62, Seasonal Items The calculation method is as follows: (1) Detrending:

[0044] (2) Calculate the seasonal component for each month. (k=1,...,12), calculate the average of the detrended values ​​for this month across all years:

[0045] in, It is to satisfy data points The number of; It means that for all satisfying index Summation; These are the detrended data values; Data points The corresponding months; based on the relatively stable annual cycle of 12 months in subtropical evergreen broad-leaved forests, k=1,...,12 is calculated, and the typical seasonal components of each month are extracted through a moving average window; (3) Adjust the seasonal components so that their annual sum is 0:

[0046] (4) Expand to the full sequence (That is, the component value of t corresponding to the month).

[0047] Step S63, Residual Term The calculation method is as follows:

[0048] Among them, residual term Including noise and outliers, the residual terms need to be analyzed. application The criteria exclude significant outliers. The criterion assumes that the residuals follow a normal distribution, calculates the mean μ and standard deviation σ of the residual sequence, and eliminates those that satisfy the following criteria. Points (considered outliers) will be used to identify outliers. Replace with The corrected seasonal and trend components are then reconstructed into a purified sequence. .

[0049] Step S7 involves filtering the time-series data of economic forests. Specifically: In this embodiment, time series data of economic forests are used. Perform Whittaker filtering; the filtering algorithm is as follows:

[0050] in, To find the fitted value, minimize the objective function. get; For smoothing parameters, it is recommended to set it to [value]. To enhance smoothness; For a d-order difference operator, d is usually set to 2; by adjusting... It effectively suppresses high-frequency man-made disturbance noise.

[0051] Step S8 involves performing phenological-guided adaptive seasonal decomposition on the filtered economic forest time-series data. Specifically: In this embodiment, time series data of economic forests are used. Perform phenology-guided adaptive seasonal decomposition, including: Step S81: The dominant cycle is detected using the Continuous Wavelet Transform (CWT). CWT refers to performing a continuous wavelet transform on the detrended sequence using Morlet wavelets as the basis function. The wavelet power spectrum of the cycle corresponding to scale s (cycle = scale × sampling interval) is calculated. By detecting the dominant cycle, if a significant non-12-month cycle exists (such as double-cropping due to management), the decomposition basis is adjusted according to the detected cycle. The criteria for determining a significant cycle are: using the red noise background spectrum test, if the power of a certain cycle exceeds the 95% confidence level, it is considered a significant cycle; if the largest significant cycle is not 12 months (such as 6 months), then this cycle T is recorded as the seasonal decomposition basis.

[0052] Step S82: Use a BFAST (Breaks For Additive Season and Trend) class method for iterative decomposition. The decomposition algorithm is as follows:

[0053] in, It is a trend item; It is a seasonal item; It is the residual term; the BFAST iterative decomposition initializes the seasonal components with a period T (detection value or default 12). The iteration continues until convergence. The iterative decomposition steps are as follows: (1) Fitting trend: for Using LOESS (span=0.3) yields

[0054] (2) Detrending:

[0055] (3) Seasonal updates: for Aggregate according to period T (same as the suburban forest method) to obtain

[0056] (4) Calculate the changes: Stop when (ε=0.001) (5) Calculation results:

[0057]

[0058]

[0059] Step S83: Combine the agricultural calendar (such as known pruning and harvest times) to analyze the residual terms. Identify significant drops in volume: if If so, it indicates synchronization with agricultural events and is considered a "true signal" and retained; if This is considered noise removal. The final output is a purified sequence. .

[0060] Step S9: Merge the time-series data of suburban forests obtained in step S6 and the time-series data of economic forests obtained in step S8. Specifically: In this embodiment, the two preprocessed groups are merged. Timing:

[0061] Step S10: Output the merged time-series data to obtain the final optimized data. Specifically: In this embodiment, the final optimized data product is output. This data product can be directly used in subsequent research such as breakpoint detection and driving factor analysis of the DBEST algorithm.

[0062] See Figure 2 The diagram shown is a hardware architecture diagram of the vegetation coverage time-series processing system 10 of the present invention. The system includes: an input module 101, a sorting module 102, a synthesis module 103, a filtering module 104, a decomposition module 105, a merging module 106, and an output module 107.

[0063] The input module 101 is used to input time-series remote sensing images of the same area, and to input classification data of suburban forests and economic forests. Specifically: In this embodiment, the input module 101 inputs time-series Sentinel-2 remote sensing images of the same region, covering at least 5 complete years in time and the entire research area in spatial scope; it also inputs classification data of suburban forests and economic forests, with the spatial scope covering the entire research area.

[0064] The sorting module 102 is used to calculate the Normalized Differential Vegetation Index (NDVI) of the study area and perform temporal sorting based on the input temporal remote sensing images. Specifically: In this embodiment, the sorting module 102 calculates the Normalized Difference Vegetation Index (NDVI) of the study area using the input time-series Sentinel-2 data and performs time-series sorting. The calculation formula is as follows:

[0065] in, Represents the normalized difference vegetation index; This represents the brightness value of the red band in Sentinel-2 data; This represents the brightness value of the near-infrared spectral band in the Sentinel-2 data.

[0066] The sorting module 102 is also used to calculate the vegetation cover data (FVC) of the study area using time-series NDVI data and perform time-series sorting. Specifically: In this embodiment, the sorting module 102 calculates the vegetation cover data (FVC) of the study area using time-series NDVI data and performs time-series sorting. The calculation formula is as follows:

[0067] in, Represents a pixel with absolutely no vegetation cover. value; Represents pixels completely covered by vegetation value.

[0068] The synthesis module 103 is used to synthesize monthly FVC time-series data for suburban forest areas and economic forest areas based on the input classification data of suburban forests and economic forests. Specifically: In this embodiment, the synthesis module 103 synthesizes monthly FVC time-series data for suburban forest and economic forest regions on a monthly basis using the input classification data of suburban forest and economic forest. .

[0069] The filtering module 104 is used for adaptive filtering of time-series data from suburban forests. Specifically: In this embodiment, the filtering module 104 is used for time-series data of suburban forests. Adaptive Savitzky-Golay filtering (hereinafter referred to as SG filtering) is performed, and the filtering algorithm is as follows:

[0070] in, The normalization coefficient is the sum of all convolution coefficients within the window (i.e., ...). In actual calculations, no explicit settings are required because the convolution kernel is automatically normalized. For the first The convolution coefficients of the points are determined by the order of the polynomial fitting (usually 2 to 3). The coefficients can be generated by specifying the window length and the order. With a window width of half, this method allows for dynamic adjustment of the window size based on the sequence noise level. The window length can be adjusted from 5 to 9. Users can evaluate the results through trial and error. Users can calculate the ratio of the root mean square error (RMSE) of the filtered sequence to the standard deviation of the original sequence. If the ratio is greater than 0.8, the window can be appropriately increased to 7 or even 9.

[0071] The decomposition module 105 is used to perform seasonal decomposition based on a fixed period on the filtered suburban forest time-series data. Specifically: In this embodiment, the decomposition module 105 targets the time-series data of suburban forests. A seasonal decomposition based on a fixed period is performed, and the decomposition algorithm is as follows:

[0072] in, It is a trend item; It is a seasonal item; These are residual terms, and the calculation of each term specifically includes: The decomposition module 105 uses the LOESS fitting trend line to determine the trend item. To calculate LOESS, a smoothing parameter `span` needs to be set (the local window span, typically 0.2-0.3 for medium- to long-term trends). For a sequence length N, the trend value... The weighted polynomial fit value (weights are a three-kernel function) is the spanN data points near point t. The fitting algorithm is as follows:

[0073]

[0074] Where h is the width of the local window (h = spanN). This is the kernel function.

[0075] The decomposition module 105 calculates seasonal items. ,include: (1) Detrending:

[0076] (2) Calculate the seasonal component for each month. (k=1,...,12), calculate the average of the detrended values ​​for this month across all years:

[0077] in, It is to satisfy data points The number of; It means that for all satisfying index Summation; These are the detrended data values; Data points The corresponding months; based on the relatively stable annual cycle of 12 months in subtropical evergreen broad-leaved forests, k=1,...,12 is calculated, and the typical seasonal components of each month are extracted through a moving average window; (3) Adjust the seasonal components so that their annual sum is 0:

[0078] (4) Expand to the full sequence (That is, the component value of t corresponding to the month).

[0079] The decomposition module 105 calculates the residual term. ,include:

[0080] Among them, residual term Including noise and outliers, the residual terms need to be analyzed. application The criteria exclude significant outliers. The criterion assumes that the residuals follow a normal distribution, calculates the mean μ and standard deviation σ of the residual sequence, and eliminates those that satisfy the following criteria. Points (considered outliers) will be used to identify outliers. Replace with The corrected seasonal and trend components are then reconstructed into a purified sequence. .

[0081] The filtering module 104 is also used to filter the time-series data of economic forests. Specifically: In this embodiment, the filtering module 104 targets the time-series data of economic forests. Perform Whittaker filtering; the filtering algorithm is as follows:

[0082] in, To find the fitted value, minimize the objective function. get; For smoothing parameters, it is recommended to set it to [value]. To enhance smoothness; For a d-order difference operator, d is usually set to 2; by adjusting... It effectively suppresses high-frequency man-made disturbance noise.

[0083] The decomposition module 105 is also used to perform phenological-guided adaptive seasonal decomposition on the filtered economic forest time-series data. Specifically: In this embodiment, the decomposition module 105 targets the time-series data of economic forests. Perform phenology-guided adaptive seasonal decomposition, including: The decomposition module 105 uses the Continuous Wavelet Transform (CWT) to detect the dominant cycle. CWT refers to selecting Morlet wavelets as the basis function and performing a continuous wavelet transform on the detrended sequence. The wavelet power spectrum of the cycle corresponding to scale s (cycle = scale × sampling interval) is calculated. By detecting the dominant cycle, if a significant non-12-month cycle exists (such as double-cropping due to management), the decomposition basis is adjusted according to the detected cycle. The criteria for determining a significant cycle are: using a red noise background spectrum test, if the power of a certain cycle exceeds the 95% confidence level, it is considered a significant cycle; if the largest significant cycle is not 12 months (such as 6 months), then this cycle T is recorded as the seasonal decomposition basis.

[0084] The decomposition module 105 uses a BFAST (Breaks For Additive Season and Trend) class method for iterative decomposition. The decomposition algorithm is as follows:

[0085] in, It is a trend item; It is a seasonal item; It is the residual term; the BFAST iterative decomposition initializes the seasonal components with a period T (detection value or default 12). Iterate until convergence. The iterative decomposition includes: (1) Fitting trend: for Using LOESS (span=0.3) yields

[0086] (2) Detrending:

[0087] (3) Seasonal updates: for Aggregate according to period T (same as the suburban forest method) to obtain

[0088] (4) Calculate the changes: Stop when (ε=0.001) (5) Calculation results:

[0089]

[0090]

[0091] The decomposition module 105 combines the agricultural calendar (such as known pruning and harvest times) to analyze the residual items. Identify significant drops in volume: if If so, it indicates synchronization with agricultural events and is considered a "true signal" and retained; if This is considered noise removal. The final output is a purified sequence. .

[0092] The merging module 106 is used to merge the time-series data of suburban forests and economic forests obtained by the decomposition module 105. Specifically: In this embodiment, the merging module 106 merges the two preprocessed sets. Timing:

[0093] The output module 107 is used to output the merged time-series data to obtain the final optimized data. Specifically: In this embodiment, the output module 107 outputs the final optimized data product. This data product can be directly used in subsequent research such as breakpoint detection and driving factor analysis of the DBEST algorithm.

[0094] This invention extracts the temporal "skeleton" information of suburban forests and economic forests through adaptive filtering and phenological-guided seasonal decomposition, providing a reliable data foundation for high-precision change detection and attribution analysis. This invention constructs two independent and highly optimized preprocessing pipelines, targeting the temporal characteristics of evergreen broad-leaved suburban forests and economic forests respectively, performing adaptive filtering, phenologically guided seasonal decomposition, and knowledge-guided signal purification. Specifically, for suburban forests, highly conformal adaptive SG filtering and fixed-period LOESS decomposition are employed to meticulously preserve their natural phenological details and long-term trends; for economic forests, more noise-resistant Whittaker smoothing and CWT-based adaptive periodic decomposition are used, along with the integration of agronomical knowledge for discriminative cleaning of residuals to accurately separate management signals from noise. Ultimately, purified FVC sequences that clearly reflect their respective "temporal skeletons" are generated, providing reliable input for subsequent high-precision breakpoint detection algorithms such as DBEST, thereby achieving accurate identification and comparative analysis of the driving factors of vegetation change in both types. This invention not only enhances the analytical value of FVC time-series products but also provides a new technological paradigm for refined remote sensing monitoring of heterogeneous vegetation areas.

[0095] Although the present invention has been described with reference to the present preferred embodiments, those skilled in the art should understand that the above preferred embodiments are only used to illustrate the present invention and are not intended to limit the scope of protection of the present invention. 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.

Claims

1. A method for time-series processing of vegetation cover, characterized in that, The method includes the following steps: S1, Input time-series remote sensing images of the same area, and input classification data of suburban forests and economic forests; S2, Based on the input time-series remote sensing images, calculate the Normalized Differential Vegetation Index (NDVI) of the study area and perform time-series sorting. S3, using time-series NDVI data, calculates the vegetation cover data (FVC) of the study area and sorts it by time series; S4. Based on the input classification data of suburban forests and economic forests, synthesize the monthly FVC time series data of suburban forest areas and economic forest areas. S5 performs adaptive filtering on time-series data of suburban forests; S6, for the filtered suburban forest time series data, perform seasonal decomposition based on a fixed period; S7, filtering for time-series data of economic forests; S8 performs phenology-guided adaptive seasonal decomposition on the filtered time-series data of economic forests; S9. Merge the suburban forest time series data obtained from step S6 and the economic forest time series data obtained from step S8. S10 outputs the merged time series data to obtain the final optimized data.

2. The method as described in claim 1, characterized in that, Step S2 includes: The Normalized Differential Vegetation Index (NDVI) of the study area was calculated using the input time-series Sentinel-2 data and then sorted by time series. The calculation formula is as follows: in, Represents the normalized difference vegetation index; This represents the brightness value of the red band in Sentinel-2 data; This represents the brightness value of the near-infrared spectral band in the Sentinel-2 data.

3. The method as described in claim 2, characterized in that, Step S3 includes: The vegetation cover (FVC) data of the study area was calculated using time-series NDVI data and then sorted temporally. The calculation formula is as follows: in, Represents a pixel with absolutely no vegetation cover. value; Represents pixels completely covered by vegetation value.

4. The method as described in claim 3, characterized in that, Step S5 includes: For suburban forest time series data Perform adaptive Savitzky-Golay filtering: in, The normalization coefficient is the sum of all convolution coefficients within the window. , For the first The convolution coefficients of a point are determined by the order of the polynomial fitting; these coefficients can be generated by specifying the window length and the order. It is half the width of a window.

5. The method as described in claim 4, characterized in that, Step S6 includes: For suburban forest time series data A seasonal decomposition based on a fixed period is performed, and the decomposition algorithm is as follows: in, It is a trend item; It is a seasonal item; It is the residual term.

6. The method as described in claim 5, characterized in that, Step S7 includes: Regarding time series data of economic forests Perform Whittaker filtering; the filtering algorithm is as follows: in, To find the fitted value, minimize the objective function. get; For smoothing parameters; It is a d-order difference operator.

7. The method as described in claim 6, characterized in that, Step S8 includes: Step S81: Detect the dominant cycle using continuous small transformations. If there is a significant non-12-month cycle, adjust the decomposition basis according to the detection cycle. Step S82: Use BFAST-like methods for iterative decomposition. The decomposition algorithm is as follows: in, It is a trend item; It is a seasonal item; It is the residual term; the BFAST iterative decomposition initializes the seasonal components with a period T. Iterate until convergence; Step S83, combine the agricultural calendar to analyze the residual items. The significant drop in the value was identified.

8. The method as described in claim 7, characterized in that, Step S9 includes: Merge the two preprocessed groups Timing: 。 9. A vegetation cover time-series processing system, characterized in that, The system includes an input module, a sorting module, a synthesis module, a filtering module, a decomposition module, a merging module, and an output module, wherein: The input module is used to input time-series remote sensing images of the same area and input classification data of suburban forests and economic forests. The sorting module is used to calculate the Normalized Differential Vegetation Index (NDVI) of the study area and sort it in time series based on the input time-series remote sensing images. The sorting module is also used to calculate the vegetation cover data (FVC) of the study area using time-series NDVI data and to sort it in time series. The synthesis module is used to synthesize monthly FVC time-series data for suburban forest areas and economic forest areas based on the input classification data of suburban forests and economic forests. The filtering module is used for adaptive filtering of time-series data of suburban forests; The decomposition module is used to perform seasonal decomposition based on a fixed period on the filtered suburban forest time series data. The filtering module is also used to filter time-series data of economic forests; The decomposition module is also used to perform phenological-guided adaptive seasonal decomposition on the filtered economic forest time series data; The merging module is used to merge the time-series data of suburban forests and economic forests obtained by the decomposition module. The output module is used to output the merged time series data to obtain the final optimized data.

Images