Lake phytoplankton phenological parameter extraction method and system based on multi-scale parameter migration and dynamic threshold

Abstract

The invention relates to the technical field of data extraction, and discloses a lake phytoplankton phenological parameter extraction method and system based on multi-scale parameter migration and a dynamic threshold, and the method comprises the steps: obtaining remote sensing image data of a target region, and constructing image data of a multi-scale time sequence; the method comprises the following steps: dividing a growth cycle of a plant, constructing a growth curve reference model and a parameter mapping model to obtain phenological morphological parameters of the plant in the growth cycle, and identifying and extracting key phenological parameters of the growth cycle through a preset dynamic threshold value. According to the method, the fitting stability of the high-noise sparse time sequence is improved, the fitting parameter mapping model is established, direct fitting of high-noise data is avoided, and the model robustness is improved; the dynamic self-adaptive adjustment of the phenological threshold is realized, so that the threshold is self-adaptively changed along with the algal bloom intensity, and the adaptability to lakes with different nutrition states and different years is improved; and long-time-sequence lake algal bloom phenology monitoring in different climate regions and different nutrition levels is supported.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data extraction, and in particular to a lake phytoplankton phenology parameter extraction method and system based on multi-scale parameter migration and dynamic threshold. BACKGROUND

[0002] Existing research transplants various time series smoothing methods (such as filtering method, fitting method) and phenology parameter extraction methods (such as threshold method, rate method, cumulative method) from marine phenology. Different methods and threshold selection will bring significant differences in the extraction results of phenology parameters, and these methods are originally designed for open seas with relatively uniform optical properties or terrestrial vegetation with relatively stable growth cycles. Phytoplankton, especially in eutrophic lakes, has the characteristics of short life cycle and rapid response to the environment, and its biomass (represented by remote sensing signals) changes faster and fluctuates more dramatically, resulting in noisier time series signals than marine phytoplankton or terrestrial vegetation. Therefore, directly applying existing models without adaptive optimization for the rapid change characteristics of lake phytoplankton and the high-noise data environment is the main reason for the inaccuracy and poor universality of the extraction results of phenology. SUMMARY

[0003] In view of the above existing problems, the present application is proposed. Therefore, the present application provides a lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold to solve how to effectively overcome the high-noise interference of inland water bodies and accurately capture the rapid change characteristics of lake phytoplankton, so as to make up for the problems of the prior art in the accuracy, reliability and automation of phenology parameter extraction.

[0004] To solve the above technical problems, the present application provides the following technical solutions: In a first aspect, the present application provides a lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, comprising: acquiring remote sensing image data of a target area, and preprocessing the remote sensing image data to construct multi-scale time series image data; based on the multi-scale time series image data, dividing the growth cycle of plants to identify the growth cycle, constructing a growth curve reference model and a parameter mapping model to obtain the phenology morphological parameters of plants in the growth cycle; performing pixel-level growth cycle segmentation on the original pixel sequence of the image data, and based on the trained parameter mapping model, identifying and extracting the key phenology parameters of the growth cycle through a preset dynamic threshold.

[0005] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, wherein: remote sensing image data of a target area is acquired, and the remote sensing image data is preprocessed, including: Collecting original remote sensing image data of different spatial resolutions of a target area, and extracting a floating algae coverage feature in the original remote sensing image data; Collecting global surface water data, selecting a region from the global surface water data as a reference water body mask, constructing a dynamic water body mask algorithm through a multispectral index and a reflectivity threshold, eliminating non-water body pixels and interference water body pixels in the remote sensing image data, and obtaining effective water body pixels in the remote sensing image data; Calculating a daily floating algae coverage index for the effective water body pixels, and performing multi-scale aggregation on remote sensing image data of different spatial resolutions to obtain a floating algae coverage time series.

[0006] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, wherein: the multi-scale aggregation of remote sensing image data of different spatial resolutions to obtain a floating algae coverage time series includes: The original remote sensing image data of different spatial resolutions is divided into high-resolution and low-resolution image data, and after extracting the floating algae coverage feature, a high-resolution original floating algae coverage time series and a low-resolution floating algae coverage time series are obtained; Wherein, the low-resolution image data is divided into a plurality of grid units of the same size, and the daily floating algae coverage value is determined according to the proportion of effective pixels in the grid unit, and the low-resolution floating algae coverage time series is obtained; The low-resolution floating algae coverage time series is subjected to sequence difference reconstruction to obtain a complete low-resolution floating algae coverage time series.

[0007] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, wherein: obtaining the phenology parameters of plants in the growth cycle includes: Based on the floating algae coverage index, the growth cycle of the low-resolution floating algae coverage time series is segmented to obtain a plurality of independent growth cycles; Based on the independent growth cycle, curve fitting is performed on the low-resolution floating algae coverage time series of the independent growth cycle to obtain a fitting parameter set for each growth cycle; A parameter mapping model is constructed, the model is trained through the fitting function parameter set, and a mapping relationship between low-resolution data and high-resolution data is established.

[0008] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, the parameter mapping model comprises: Based on the high-resolution original phytoplankton coverage time series of the target area in each growth cycle, a plurality of statistical features of each pixel in the sequence are calculated; A parameter mapping model is constructed by a machine model, the statistical features at high resolution are taken as model inputs, the fitted parameter set of the low-resolution sequence is taken as model outputs, the parameter mapping model is trained, a mapping relationship between the statistical features and the fitted parameter set is established, and a predicted fitted parameter of the high-resolution original phytoplankton coverage time series is obtained.

[0009] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, the parameter mapping model comprises: Through a sliding window, a phenology segmentation index of the high-resolution original phytoplankton coverage time series is calculated, and division of the growth cycle is performed; Based on each growth cycle segmented from each pixel at high resolution, the statistical features are extracted and input into the parameter mapping model, a fitted parameter set of each growth cycle of the pixel is predicted, and a fitted time series of the cycle is obtained; According to the maximum amplitude of each growth cycle, a best phenology determination threshold is dynamically calculated, and time data of plant growth is obtained from the fitted time series.

[0010] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, the parameter mapping model comprises: The high-resolution original phytoplankton coverage time series of each year is aligned in sequence according to the annual accumulated day; Through a preset sliding window, the mean, standard deviation and maximum of the phytoplankton coverage at each annual accumulated day position are calculated and fused to obtain a phenology segmentation index; The annual accumulated day corresponding to the minimum value of the phenology segmentation index is selected as an annual growth cycle segmentation point of the pixel, that is, the growth cycle of the pixel in each year starts from the annual accumulated day in the year and ends at the annual accumulated day in the next year.

[0011] As a preferred scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, the parameter mapping model comprises: The maximum amplitude determined in the fitting parameter set is used to construct a functional relationship between the maximum amplitude and an optimal activity amplitude threshold, and the optimal function form and function parameters are determined by the least square method, so as to calculate the optimal activity amplitude threshold; Based on the optimal activity amplitude threshold, from the fitting time sequence, the date corresponding to the first time when the floating algae coverage value in the growth period rises from the left baseline and exceeds the threshold is selected as the growth start day, and the date corresponding to the last time when the floating algae coverage value in the decline period drops to the threshold is selected as the growth end day.

[0012] In a second aspect, the present application provides a lake phytoplankton phenology parameter extraction system based on multi-scale parameter migration and dynamic threshold, comprising: A collection processing module is configured to acquire remote sensing image data of a target area, and pre-process the remote sensing image data to construct multi-scale time sequence image data. A mapping module is configured to divide the growth cycle of plants based on the multi-scale time sequence image data, to identify the growth cycle, construct a growth curve benchmark model and a parameter mapping model, and to obtain the phenology parameters of plants in the growth cycle. An output module is configured to perform pixel-level growth cycle segmentation on the original pixel sequence of the image data, and to identify and extract the key phenology parameters of the growth cycle based on the trained parameter mapping model and a preset dynamic threshold.

[0013] In a third aspect, the present application provides a computer device, comprising: A memory and a processor; The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions, which realize the steps of the method for extracting lake phytoplankton phenology parameters based on multi-scale parameter migration and dynamic threshold.

[0014] In a fourth aspect, the present application provides a computer readable storage medium, which stores computer executable instructions, and the computer executable instructions realize the steps of the method for extracting lake phytoplankton phenology parameters based on multi-scale parameter migration and dynamic threshold when executed by a processor.

[0015] Compared with the prior art, the present application has the beneficial effects: the present application improves the fitting stability of high-noise sparse time series, establishes a fitting parameter mapping model by fusing 5km scale relatively complete time series and 250m pixel level time sequence statistical characteristics, avoids direct fitting of high-noise data, and improves the robustness of the model; realizes dynamic self-adaptive adjustment of the phenology threshold, establishes a functional relationship between the maximum amplitude and the optimal threshold, makes the threshold adaptively change with the intensity of algal blooms, and improves the adaptability to different nutrient state lakes and different years; unifies the uncertainty evaluation standard of the phenology parameter, introduces the spatial variation coefficient and the spatial autocorrelation index, quantifies the spatial consistency of the model results, and enhances the comparability of the results of different regions and different methods; supports cross-regional and long-time sequence lake algal bloom phenology monitoring, is suitable for typical lakes in different climate zones and different nutrient levels, and provides reliable technical support for early warning and ecological management of algal blooms. BRIEF DESCRIPTION OF DRAWINGS

[0016] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0017] Figure 1 The whole flowchart of a kind of lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to an embodiment of the present application.

[0018] Figure 2 The specific flowchart of a kind of lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to an embodiment of the present application.

[0019] Figure 3 The experimental data comparison chart of a kind of lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to an embodiment of the present application. DETAILED DESCRIPTION

[0020] In order to make the above-mentioned purposes, features and advantages of the present application more obvious and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0021] Example 1, refer to Figures 1-2and Table 1, an embodiment of the present application provides a lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold, comprising: S1, acquiring remote sensing image data of a target area, and pre-processing the remote sensing image data to construct multi-scale time series image data; S2, based on the multi-scale time series image data, dividing the growth cycle of plants to identify the growth cycle, constructing a growth curve reference model and a parameter mapping model to obtain the phenological form parameters of plants in the growth cycle; S3, performing pixel-level growth cycle segmentation on the original pixel sequence of the image data, and based on the trained parameter mapping model, identifying and extracting key phenological parameters of the growth cycle through a preset dynamic threshold.

[0022] It should be noted that the present application improves the fitting stability of high-noise sparse time series, establishes a fitting parameter mapping model by fusing 5km scale relatively complete time series and 250m pixel level time series statistical characteristics, avoids directly fitting high-noise data, improves the robustness of the model; Realize the dynamic adaptive adjustment of the phenological threshold, establish the functional relationship between the maximum amplitude and the optimal threshold, make the threshold adaptively change with the intensity of algal bloom, improve the adaptability of different nutrient state lakes and different years; Unified uncertainty evaluation standard of phenological parameters, introduce spatial variation coefficient and spatial autocorrelation index, quantify the spatial consistency of model results, enhance the comparability of results of different regions and different methods; Support cross-regional, long-time series lake algal bloom phenology monitoring, suitable for typical lakes in different climate zones and different nutrient levels, provide reliable technical support for early warning and ecological management of algal blooms.

[0023] Preferably, in step S1, the remote sensing image data of the target area is acquired, and the remote sensing image data is pre-processed, comprising: Collecting original remote sensing image data of different spatial resolutions of the target area, and extracting floating algae coverage features in the original remote sensing image data; Collecting global surface water data, selecting a region from the global surface water data as a reference water body mask, constructing a dynamic water body mask algorithm through a multi-spectral index and a reflectivity threshold, eliminating non-water body pixels and interference water body pixels in the remote sensing image data, and obtaining effective water body pixels in the remote sensing image data; Calculating the daily floating algae coverage index of the effective water body pixels, and performing multi-scale aggregation on remote sensing image data of different spatial resolutions to obtain the floating algae coverage time series.

[0024] This embodiment takes the floating algae coverage (FAC) index calculated by MODIS Aqua satellite data as an example, and collects high and low resolution image data, i.e. high resolution 250m original pixel sequence and low resolution 5km but relatively complete and smooth grid cell sequence. Specifically, the MODIS Aqua MYD09GQ of 250m resolution, including red and near-infrared bands, and the MYD09GA of 500m resolution data, including blue, green and short-wave infrared bands, of the research area (such as Taihu Lake, Chaohu Lake, Dianchi Lake and Hulun Lake) from 2003 to 2024 are obtained from Google Earth Engine platform, the 500m band is resampled to 250m to make the 500m data and the 250m data spatially aligned, and the JRC global land surface water data is combined to construct a dynamic water body mask through spectral index and reflectivity threshold to eliminate non-water body pixels and outliers.

[0025] Among them, the MODIS MYD09GQ dataset provides 250m resolution red and near-infrared band reflectivity for core calculation, and the MODIS MYD09GA dataset provides 500m resolution blue, green and short-wave infrared 1 and short-wave infrared 2 band reflectivity, which is bilinearly resampled to 250m resolution to unify the spatial scale.

[0026] Further, the floating algae coverage features in the original remote sensing image data are extracted, and the daily FAC value is calculated, which is represented as: ; Among them, represents the reflectivity value of the near-infrared band, with a center wavelength of 859nm; represents the reflectivity value of the red light band, with a center wavelength of 645nm; represents the reflectivity value of the short-wave infrared 1 band, with a center wavelength of 1240nm.

[0027] The occurrence layer in the JRC global land surface water dataset (GSW1_0) is used to select the area with a surface water annual occurrence frequency of ≥90% as the reference water body mask, which can ensure that the analysis area is limited within the stable water body range. The multi-spectral index includes turbid water index, cloud and flare mask, shadow and aquatic vegetation suppression index.

[0028] Among them, the turbid water index is used to eliminate turbid water pixels caused by high suspended sediment concentration, which is represented as: ; Among them, represents the reflectivity value of the red light band, with a center wavelength of 645nm, The reflectance value of the short-wave infrared 1 band, the center wavelength is 1240nm. The reflectance of turbid water in the red light band will be significantly increased due to the backscattering of suspended particles, even exceeding the reflectance in the short-wave infrared band, and the clean water body should satisfy Therefore, the turbid water pixel with a turbid water index less than 0.09 is removed.

[0029] Clouds and flares have high reflectivity in all visible light bands. The pixels affected by thick clouds and solar flares are removed by cloud and flare masks. The pixels satisfying the following conditions are determined as clouds or flares and removed, that is, , , wherein, represents the reflectance in the green light band, the center wavelength is 555nm, represents the reflectance in the blue light band, the center wavelength is 469nm.

[0030] The dark pixels such as mountain shadows, cloud shadows and the no-signal area with too low reflectance are removed by shadow. The pixels satisfying the following conditions are determined as shadows and removed, that is, , , .

[0031] The water vegetation suppression index is used to remove the interference of water vegetation in wetlands and shallow water areas, and is represented as: ; The index enhances the green peak signal of green vegetation by introducing the short-wave infrared band to construct a baseline. The determination condition of the vegetation coverage area is CMI>0.02, and the interference pixels are removed.

[0032] It should be noted that the present application can effectively remove the non-water body pixels and the interference water body pixels with complex optical characteristics in the remote sensing image by constructing a dynamic and high-precision water body mask, so as to provide a pure and reliable water body pixel dataset for subsequent phytoplankton phenology parameter extraction.

[0033] Preferably, in the step S1, the multi-scale aggregation is performed on the remote sensing image data with different spatial resolutions to obtain the time sequence of the floating algae coverage, including: The original remote sensing image data with different spatial resolutions are divided into high-resolution and low-resolution image data. After extracting the floating algae coverage feature, the high-resolution original floating algae coverage time sequence and the low-resolution floating algae coverage time sequence are obtained. Wherein, the low-resolution image data is divided into a plurality of grid units with the same size. According to the proportion of the effective pixels in the grid unit, the daily floating algae coverage value is determined, and the low-resolution floating algae coverage time sequence is obtained. ​The low-resolution floating algae coverage time series is reconstructed by sequence difference to obtain a complete low-resolution floating algae coverage time series.

[0034] In the embodiment, the high-resolution original floating algae coverage time sequence is the original FAC time sequence of 250-meter pixels; the low resolution refers to the 5-km image data collected, and the lake is divided into multiple units using a 5-km grid; if the number of valid pixels of each unit on a certain day accounts for ≥20%, the mean value of the FAC of these pixels is taken as the FAC value of the unit on that day; otherwise, the FAC value is marked as missing.

[0035] Specifically, the boundaries of the four lakes are divided into a plurality of 5*5-km grid square units using a 5-km fishing net; due to the irregular shape of the lakes, some areas are less than 5 km but are still regarded as independent units; it is assumed that a certain unit contains a total number of pixels n on a certain day, and the number of non-mask pixels m; when m≥n / 5, the mean value of the FAC of the non-mask pixels is calculated as the FAC value of the unit on that day; if m

[0036] Further, 5-km sequence interpolation reconstruction is performed, and features are selected, including meteorological factors, historical FAC dynamic characteristics, and day of year (DOY), etc.; the meteorological factors can be daily mean temperature, daily maximum temperature, daily / monthly maximum precipitation, u / v direction wind speed, daily mean atmospheric pressure, solar radiation, etc.; the historical FAC dynamic characteristics can be the FAC mean value and maximum value in a 51-day sliding window centered on the current day; the XGBoost algorithm is used to divide the effective FAC observation data and corresponding features of each 5-km unit into a training set and a test set in a ratio of 4:1, and the hyperparameters are optimized through GridSearchCV. For units with sparse data, the data of the adjacent four units are combined for joint training, and then the trained model is used to process the FAC time sequence interpolation of the target unit to finally obtain a complete daily FAC sequence.

[0037] Specifically, first, the original daily FAC observation data of the 5-km image data and the corresponding multi-dimensional feature data, such as water body optical parameters and weather conditions, are read, and are aligned and combined by date to generate a complete daily time axis within the collection period, ensuring that the subsequent interpolation results are continuous and uninterrupted in time.

[0038] Further, key timing features are created, including sliding window mean features and sliding window peak features. The sliding window mean feature calculates the average of all valid high FAC observations within a 25-day window for each date, capturing the short-term background state of the algal bloom. The sliding window peak feature calculates the average of the top three FAC values within a 25-day window for each date, representing the potential algal bloom outbreak intensity in the local area. Linear interpolation is used to fill in missing values in the calculated timing features, ensuring the integrity of the feature matrix.

[0039] For specific 5km grid cells with insufficient FAC observation data, a spatial proximity strategy is used to merge the data of the target cell with its four directly adjacent cells (east, south, west, and north), generating an expanded and more data-rich training dataset.

[0040] An XGBoost regression model is trained using the merged dataset. The trained model is used to interpolate the feature data of the target cell, including the interpolated timing features, to obtain the daily FAC interpolation results for the target cell. RandomizedSearchCV is used to perform random search within a predefined parameter space, and K-fold cross-validation (K=4) is used to evaluate parameter performance. The optimized key hyperparameters include the number of trees, the maximum depth of the tree, the learning rate, the subsampling ratio, the feature sampling ratio, and the regularization parameter. Multiple rounds of training and validation are performed, and the optimal combination of hyperparameters is selected based on the performance on the cross-validation set to build the final model.

[0041] It should be noted that this step builds a FAC (algal coverage) replacement model for inland water bodies, which is not suitable for the inland water bodies. The model provides more reliable time series data for subsequent phenology analysis. The time series reconstruction and noise filtering process is optimized for the characteristics of phytoplankton and MODIS data, effectively improving the quality and usability of time series data.

[0042] Preferably, in step S2, the phenological morphological parameters of the plant in the growth cycle are obtained by: Based on the floating algae coverage index, the growth cycle of the low-resolution floating algae coverage time series is segmented to obtain multiple independent growth cycles. Based on the independent growth cycle, the low-resolution floating algae coverage time series of the independent growth cycle is curve-fitted to obtain multiple optimal fitting parameters. A parameter mapping model is constructed, and the model is trained by the optimal fitting parameters to establish the mapping relationship between the low-resolution data and the high-resolution data.

[0043] In this embodiment, for each 5km grid cell, the sliding window detection method is used to identify the FAC minimum value trough points between adjacent years, and the long time series is divided into multiple independent growth cycles. For example, 22 years of data are divided into 22 cycles, each of which contains a complete ecological process from the beginning, development to decline. For each growth cycle, the FAC dynamic change in each growth cycle is fitted using an asymmetric Gaussian function, which is expressed as: ; where t is the tth day in the growth cycle; is the daily FAC active index after reconstruction of the time series; is the maximum value of FAC in the growth cycle, i.e. the peak amplitude; represents the peak position; is the baseline value before the growth season; is the baseline after the growth season, and control the steepness on both sides of the peak.

[0044] To ensure the convergence of the fitting process and the robustness of the results, the model parameters are initialized based on the phenology strategy. The peak amplitude is initialized as the mean value of the high percentile interval (such as the top 10%) of all valid FAC observations in the growth cycle, which can effectively approximate the true biomass peak and avoid initialization bias caused by individual extreme outliers. The peak position is initialized as the date sequence DOS corresponding to the maximum FAC observation in the growth cycle, which is used as a benchmark to capture the core time node of algal bloom outbreak. The left and right baselines , are initialized as the statistical quantities (such as the median or mean) of all valid FAC observations in a complete phenological window (such as 1 to 2 months) before and after the start of the growth cycle, which can robustly estimate the background activity level of algae in the non-active growth season. The initial values of the width parameters , are set as an empirical proportion value (such as 1 / 10 to 1 / 5 of the total duration) associated with the total duration of the growth cycle, which ensures that the initial curve shape can adapt to the actual time scale of the phenological process, providing a reasonable starting point for subsequent optimization.

[0045] Further, parameter optimization is performed, and the Levenberg-Marquardt algorithm is used for nonlinear least squares fitting to minimize the residual between the fitting curve and the observation value, to obtain the optimal fitting parameter set of each growth cycle at a resolution of 5 km .

[0046] Preferably, in step S2, the parameter mapping model is constructed by: Based on the high-resolution original phytoplankton coverage time series of the target area in each growth cycle, a plurality of statistical features of each pixel in the sequence are calculated; The parameter mapping model is trained by constructing the parameter mapping model through the machine model, taking the statistical features as the model input, and taking the fitting parameter set of the low-resolution sequence as the model output, to establish the mapping relationship between the plurality of statistical features and the fitting parameter set, so as to obtain the predicted fitting parameter.

[0047] It should be noted that the phytoplankton biomass dynamics has significant transient characteristics, and its aggregation-dispersion process in a small scale range is often completed within a few hours. However, due to the limitation of the observation ability of the MODIS satellite, that is, the daily revisit period and the limited water penetration depth, the observation data at a resolution of 250 m is difficult to accurately capture such rapid change process. Specifically, most pixels can only record the low biomass background value after the algal bloom disappears, and a few pixels accidentally capture the algal aggregation peak value, resulting in significant high-value outliers in the data. This "all or nothing" observation mode produces obvious signal discontinuity problems at the pixel scale, so that the FAC time series constructed based on the original observation data has significant noise.

[0048] In this embodiment, in step S2, the parameter mapping model is constructed as follows: For the original FAC observation data of each growth cycle , the weighted center point is calculated first as the initial value of the peak value parameter, and the data is divided into growth period and decline period , that is, before time is the growth period, and after time is the decline period. Divide into four subsets in ascending order and four subsets in descending order, to obtain eight subsets to . For each subset, the FAC mean value and the weighted mean value of are calculated, to obtain a total of 16 statistical features.

[0049] Firstly, the validated fitting parameter set in 5 km grid cell is extracted as a reliable phenology benchmark, and the statistical characteristics of the original FAC time series of all 250 m pixels in the corresponding grid cell are calculated, including mean, variance, skewness, kurtosis, and each quantile, etc. Through the establishment of the machine learning mapping relationship between the above statistical characteristics and the six fitting parameters, the effective downscaling of the phenology parameters from 5 km to 250 m scale is realized, which enables the estimation of pixel-level phenology parameters based on 250 m original observation data directly.

[0050] Through experiments, 16 statistical characteristic values in the growth cycle are finally determined, which respectively form 6 groups of characteristics, and 6 parameter mapping models are trained respectively. Assuming that there are original data in a certain growth cycle in a pixel , wherein, is the original FAC observation value, is the corresponding observation time, i.e. the day in the growth cycle.

[0051] Specifically, the initial value of the parameter is calculated first , which is the weighted center point of the growth cycle, and the growth cycle is divided into left and right parts. The initial value is expressed as: ; wherein, represents the i-th observation value in the original sequence, and the original data is divided into two parts, The original data set of the former is recorded as the growth period , The original data set of the latter is recorded as , the former is sorted in ascending order according to , and it is divided into four equal parts, recorded as - , and the latter is sorted in descending order according to , and it is also divided into four equal parts, recorded as - , the mean value of in and the weighted mean value of are calculated respectively, which correspond to the characteristics , , and the calculation formula is as follows: ; ; wherein, represents the k-th set divided according to the four quantiles of FAC size in the growth and decline periods, represents the number of original observation values in the k-th set, represents the FAC mean value in the k-th set, The average value of the FAC observation in the kth set as the weight corresponding to the time (DOY).

[0052] In addition, the parameters and in the asymmetric Gaussian function have no specific phenological significance, and it is difficult to establish a mapping relationship. Two dynamic threshold parameters and are used as an alternative, where , satisfy the following formula, which is expressed as: ; ; where , indicates that the two parameters are substituted into the asymmetric Gaussian function value of the interannual curve fitting FAC; and and and are expressed as: ; ; These two parameters represent the t values on the left and right sides when the function F(t) reaches 20% of the maximum amplitude, respectively. 20% is a compromise between sensitivity and stability, and replaces the parameters and to control the growth / decay rate.

[0053] Finally, the model is trained, and the above 16 statistical features are combined into 6 groups, which are used as independent variables to train machine learning models to predict 6 fitting parameters. The machine model can be XGBoost, random forest, support vector regression, etc. To improve the interpretability and stability of the model, dynamic threshold parameters and are used, i.e., the dates when the function value reaches 20% of the maximum amplitude, to build six independent parameter mapping models.

[0054] Among them, machine learning algorithms such as XGBoost, random forest and support vector regression are selected to build parameter mapping models, different feature combinations are used as input, and the prediction model structure of the corresponding parameters is established by setting parameters such as the number of decision trees, depth and kernel function. Calculate R 2 , mean absolute error, root mean square error and relative error to evaluate the performance of the model. Select the optimal model for comprehensive evaluation as the mapping model, and build the optimal prediction model for the six parameters, as shown in Table 1.

[0055] Table 1: Parameter table Parameter Phenological significance Characteristic A FAC peak value FAC mean value in the mth set, m = 3, 4, 5, 6 p Peak time Time mean value corresponding to the FAC observation value in the mth set as the weight, m = 3, 4, 5, 6 b L ]]> Left baseline value FAC mean value in the mth set, m = 1, 2, 3, 4 b R ]]> Right baseline value FAC mean value in the mth set, m = 5, 6, 7, 8 Q L ]] Growth average rate Time mean value corresponding to the FAC observation value in the mth set as the weight, m = 1, 2, 3, 4 Q R ]] Decline average rate Time mean value corresponding to the FAC observation value in the mth set as the weight, m = 5, 6, 7, 8

[0056] Preferably, in step S3, identifying the key phenological parameters of the growth cycle includes: By a sliding window, the phenological segmentation index of the high-resolution original floating algae coverage time series is calculated to divide the growth cycle; Based on each growth cycle segmented by each pixel of high resolution, statistical features are extracted and input into a parameter mapping model to predict the fitting parameter set of each growth cycle of the pixel, and a fitting time series of the cycle is obtained; According to the maximum amplitude of each growth cycle, the best phenological determination threshold is dynamically calculated, and the time data of plant growth is obtained from the fitting time series.

[0057] Preferably, the phenological segmentation index of the high-resolution original floating algae coverage time series is calculated to divide the growth cycle, including: The high-resolution original floating algae coverage time series of each year is aligned in sequence according to the annual accumulated day; By a preset sliding window, the mean, standard deviation and maximum of the floating algae coverage at each annual accumulated day position are calculated, and weighted fusion is performed to obtain the phenological segmentation index; The annual accumulated day corresponding to the minimum value of the phenological segmentation index is selected as the annual growth cycle segmentation point of the pixel, that is, the growth cycle of the pixel each year starts from the annual accumulated day of the year and ends at the annual accumulated day of the next year.

[0058] Specifically, for the original FAC sequence of each 250-meter pixel in 22 years, a segmentation algorithm suitable for sparse high-noise sequences is proposed. The data of each year is aligned according to the annual accumulated day DOY, DOY=1, 2,..., 366, a 21-day sliding window is used to calculate the FAC mean, standard deviation and maximum at each DOY position, and after normalization, the three indicators are weighted and fused to construct the phenological segmentation index , denoted as: ; Among them, represents the normalized FAC mean, represents the normalized FAC standard deviation, represents the normalized FAC maximum.

[0059] The DOY corresponding to the minimum value of PSI is selected as the annual growth cycle segmentation point of the pixel, that is, the growth cycle of the pixel each year starts from the DOY day of the year and ends at the DOY day of the next year. For Hulun Lake with ice period, two segmentation points DOY S(start) and DOY E (end), and are limited in the DOY ranges of [120, 150] and [300, 330] respectively.

[0060] Further, for each growth cycle of each pixel, 16 statistical features are extracted and input into the six parameter mapping models trained in step S2 to predict the fitting function parameters of the growth cycle of the pixel , so as to obtain the fitting time series of the cycle. Using these parameters and the fitting function, the smooth fitting curve of the pixel in the cycle can be obtained. The optimal threshold of the growth cycle is determined by the optimal threshold relationship determined in the fitting function parameters, and the starting and ending days in the growth cycle SOS, EOS are extracted, and the two data are converted into the day of the year SOY, EOY.

[0061] Preferably, the optimal phenology determination threshold is dynamically calculated, and the time data of plant growth is obtained from the fitting time series, including: The maximum amplitude determined in the fitting parameter set is used to construct a function relationship between the maximum amplitude and the optimal activity amplitude threshold, and the optimal function form and function parameters are determined by the least square method to calculate the optimal activity amplitude threshold; Based on the optimal activity amplitude threshold, the date when the floating algae coverage value first rises from the left baseline to exceed the threshold in the growth period is selected as the growth start day, and the date when the floating algae coverage value last falls to the threshold in the decline period is selected as the growth end day.

[0062] Specifically, a function relationship between the maximum amplitude A and the optimal activity amplitude threshold threshold is constructed to realize threshold self-adaptation. A batch of reference phenological periods of growth cycles are determined by visual interpretation, i.e. the 15-day interval in which SOS and EOS of the i-th growth cycle in a 5km unit are located, the corresponding threshold is calculated, and A is fitted. This embodiment proposes three candidate function forms, including: Exponential decay function, expressed as: ; wherein, , , are constants.

[0063] Modified power function, expressed as: ; wherein, , , are constants.

[0064] Sigmoid decay function, expressed as: ;

[0065] The maximum amplitude A after fitting each growth cycle function is the independent variable, and the threshold value is the dependent variable, which is substituted into the function, and the parameters of the three fitting functions are determined by using the least square method, and the determination coefficient R 2 The function relationship between the final A and the optimal threshold value is determined, and the optimal activity amplitude threshold value is finally calculated, and the higher threshold value corresponds to the larger amplitude to suppress noise, and the lower threshold value corresponds to the larger amplitude to capture the decay signal.

[0066] For each growth cycle of each pixel, according to the maximum amplitude A, the adaptive threshold value is calculated through the selected optimal function relationship, and then, on the fitting curve, the date corresponding to the first time that the FAC value in the growth period rises above the threshold value from the left baseline The date corresponding to the last time that the FAC value in the decay period falls below the threshold value is the growth end date EOS.

[0067] It should be noted that the present application significantly improves the accuracy and robustness of the extraction of the phenology parameter under high noise and sparse data, and effectively avoids the problem of failure or large deviation when directly fitting the 250-meter pixel-level data with high noise and high missing by using the innovative strategy of 'low-resolution fitting and high-resolution migration'. The present application makes use of the advantages of relatively continuous and high signal-to-noise ratio of 5-kilometer scale data to establish a robust phenology curve fitting model at this scale, and then learns the complex mapping relationship between the original time series statistical characteristics and the fitting function parameters through a machine learning model, and migrates this relationship to the pixel level, so that the pixel-level phenology extraction not only depends on the fragile data cleaning and direct fitting, but also inherits the reliable ecological process rule established at a large scale, so that stable and reliable results can still be obtained when facing serious data missing and noise interference.

[0068] Meanwhile, the present application also realizes the adaptive dynamic optimization of the phenology extraction threshold, improves the applicability and accuracy of the model in different lakes and different years, and overcomes the poor adaptability of the traditional fixed threshold method in the inland water body with strong spatio-temporal heterogeneity. The present application can automatically adapt to various scenes from oligotrophic to eutrophication, from normal years to abnormal years. By constructing the quantitative function relationship between the maximum amplitude and the optimal activity amplitude threshold, the threshold can be dynamically adjusted according to the intensity of algal blooms. For weak algal blooms, a higher threshold is used to effectively suppress the misjudgment of background noise; for strong algal blooms, a lower threshold is used to accurately capture the starting and dissipation signals of algal blooms, so that the identification of the phenology parameter is more in line with the actual ecological process, and the identification bias introduced by improper threshold selection is significantly reduced.

[0069] The application also has good universality and scalability, wide application range, can be applied to multiple typical lakes in different climate zones and different nutrition levels, and verifies its effectiveness on different phytoplankton characterization indexes (such as FAC and Chl-a). Not dependent on specific sensors or indexes, through the inventive concept of "sequence reconstruction-parameter mapping-threshold self-adaptation", it can be easily migrated to the phenology extraction of other medium-high resolution remote sensing data (such as OLCI, VIIRS) or other water body optical parameters (such as NDVI, PC), and provides a general technical solution for large-scale, long-time sequence lake water ecological environment monitoring.

[0070] In step S2 of embodiment 2, the growth curve benchmark model can include: Step one, obtain the meteorological data of the low-resolution image data unit grid and the historical FAC data, and extract the multi-peak data of plant growth through wavelet transform; Specifically, the environmental data and historical FAC data of a 5km unit grid are obtained, wherein the environmental data includes water table temperature, solar radiation, precipitation, wind speed, etc. The environmental data and the FAC data are aligned in time and standardized.

[0071] Further, the multi-peak data of plant growth is extracted through wavelet transform, the number and position of potential peaks of the FAC time series of a 5km unit are identified through continuous wavelet transform (CWT), the local maximum lines on the scale-time plane are analyzed, and the initial peak candidate set is extracted K is the upper limit of the number of candidate peaks, and is set to 3-5.

[0072] Step two, construct a multi-peak Gaussian mixture model, the parameters of each Gaussian component are controlled by environmental factors (environmental data), the model parameters and peak number are optimized by variational Bayes method, the significant peaks are determined by posterior probability, and the actual peak number and peak position are determined.

[0073] Specifically, the multi-peak Gaussian mixture model is represented as: ; Wherein, is the FAC time series at time t, b is the baseline value, which represents the background FAC level without algal bloom; is the amplitude of the kth Gaussian component, i.e. the peak intensity; is the peak position of the kth Gaussian component; is the standard deviation of the kth Gaussian component, which is used to control the width of the peak.

[0074] Wherein, the amplitude parameter is dynamically adjusted by environmental factors, and is represented as: ; where, is the base amplitude coefficient of the kth Gaussian component, is the weight coefficient of the kth Gaussian component for the jth environmental feature transformation function, is the jth environmental feature transformation function, which can be a quadratic polynomial, a spline basis function, etc. is the normalized value of the n environmental variables at time t, the environmental variables are water temperature, solar radiation, precipitation, wind speed, and m is the number of environmental feature transformation functions.

[0075] The variational Bayesian method is used to optimize the model parameters and the number of peaks. The joint probability of the latent variables of the observed data FAC and the existence of the peaks under given parameters is determined by the posterior probability. The threshold is set to 0.75, and the actual number of peaks is finally determined and position .

[0076] The constraint conditions are set, and the parameter optimization and model verification are performed. The constraint conditions include: b is greater than or equal to 0, is greater than or equal to 0; is located within the corresponding seasonal window, for example, the spring peak should be in March-June; satisfies the ecological constraints, for example, the growth period is steep, , indicating that the decline period is gentler than the growth period. At the same time, the model is evaluated and verified by the root mean square error.

[0077] Step three, based on the above steps, input the low-resolution floating algae coverage time series, and obtain multiple growth peak parameter sets for each low-resolution unit, i.e. the phenology parameter set of each 5km unit growth peak .

[0078] Further, through the spatio-temporal perception graph attention transfer network, a parameter mapping model is constructed, including: The spatio-temporal graph structure is constructed, and the lake area is constructed as a graph G=(V,E), V represents a node set, each high-resolution pixel is a node, and the node features in the node set include the original FAC sequence and its statistical features. E is an edge set, which is constructed through hydrological connectivity and geographical distance, and is represented as: ; where, is the Euclidean distance between pixels i and j, is the distance attenuation coefficient, i.e. the influence degree of geographical distance on edge weight, which is 1km, is the hydrological connectivity coefficient, which can be determined based on historical flow field or FAC correlation.

[0079] The spatio-temporal feature encoder includes a time encoding module, a spatial attention mechanism module, and a spatial aggregation module. The time encoding module constructs a multi-scale LSTM encoder for each pixel's FAC sequence and its 16 statistical features (the extraction of statistical features is the same as in Embodiment 1) to capture short-term and long-term dependencies. The spatial attention mechanism module calculates the attention coefficients between nodes through operations such as concatenation of attention vectors to measure the strength of spatial influence. The spatial aggregation module aggregates neighbor information through attention weighting, i.e., based on the neighbor set of the node and the attention coefficients between nodes, and performs weighted aggregation through a nonlinear activation function.

[0080] The multi-task parameter prediction layer outputs the predicted set of phenological parameters for each pixel. A multi-head prediction branch is designed, with each branch corresponding to a set of phenological parameters. wherein, represents the reference parameter migrated from the 5km scale, is a parameter mapping function, i.e., a multi-layer perceptron, represents the parameter value in the predicted set of phenological parameters, which is at the 5km scale in this embodiment, represents the final representation of node i, i.e., the feature after the fusion of spatio-temporal information through the spatio-temporal feature encoder, and represents the reference parameter migrated from the low-resolution pixel. A physical consistency constraint loss is also introduced, including the peak timing increment and the steeper growth period than the decline period , and the constraint loss value is calculated by weighted summation.

[0081] In step S3, as in Embodiment 1, the high-resolution original floating algae coverage time series of each year is aligned by year cumulative day. The mean, standard deviation, and maximum of floating algae coverage at each year cumulative day position are calculated through a pre-set sliding window, and are fused to obtain the phenological segmentation index. The year cumulative day corresponding to the minimum value of the phenological segmentation index is selected as the annual growth cycle segmentation point of the pixel. Based on step S2 in this embodiment, the 16 statistical features of the high-resolution 250m FAC time series feature within the growth cycle and the environmental data feature are obtained, and the 16 FAC statistical features and 4 environmental features of each growth cycle of each pixel are output to form a 20-dimensional feature vector as the input of the parameter mapping model. The predicted number of winds and the set of phenological parameters of 250m are output based on the parameter mapping model of the 5km pixel, and then the fitting FAC curve of each growth cycle of each 250m pixel is reconstructed based on the predicted parameters.

[0082] Further, in this embodiment, based on the trained parameter mapping model, the key phenological parameters of the growth cycle are identified and extracted through a pre-set dynamic threshold, as follows: Step one, by setting the reference threshold, combined with the environmental threshold and signal quality threshold, that is, by the environmental index and signal quality index to adjust it dynamically, calculate the first threshold; Specifically, the reference threshold can be set to 0.2, and the environmental threshold and signal quality threshold are calculated respectively, and then the first threshold is obtained, wherein, first, based on water temperature, solar radiation and precipitation, the temperature and solar radiation are calculated by the Gaussian response function by combining the real-time environmental data with the suitable environmental data of phytoplankton, indicating that there is an optimal value, at the same time, the solar radiation is also calculated by the exponential function, to judge whether the actual solar radiation value is greater than the minimum solar radiation required for growth, to obtain the water temperature threshold and the solar radiation threshold, the precipitation is calculated by the linear negative effect, indicating that the precipitation inhibits the growth of phytoplankton, and the threshold values of the three are weighted and comprehensively calculated to obtain the environmental index, and the environmental index is smoothed by the Sigmoid function to obtain the environmental threshold, ensuring that the threshold value increases when the environment is extremely unsuitable, and the threshold value decreases when the environment is suitable.

[0083] The environmental index is represented as: ; Wherein, , , are the weight coefficients of temperature, radiation and precipitation respectively, is the water surface temperature of the current year and the current day, is the suitable water surface temperature of phytoplankton, which is taken as 22° in this embodiment, is the width parameter of the temperature response curve, which is taken as 8, is the solar radiation of the current year and the current day, is the minimum solar radiation required for the growth of phytoplankton, is the width parameter of the radiation response curve, which is taken as 100, is an indicator function, which is 1 when the condition is met, and 0 otherwise, is the normalized precipitation of the current year and the current day.

[0084] The environmental threshold is represented as: ; Wherein, is the steepness coefficient of the Sigmoid function, which controls the transition speed and is taken as 3, is the midpoint of the Sigmoid function, which corresponds to the moderate suitable condition and is taken as 0.5.

[0085] Further, the FAC signal quality of each pixel at each time point is evaluated, the signal-to-noise ratio is calculated by comparing the fitting value of the reconstructed fitting FAC curve with the background noise level, a lower threshold is used for a high-quality signal area, and a higher threshold is used for a low-quality signal area to avoid noise interference, wherein the background noise level can be the moving median of the past 30 days, the signal-to-noise ratio is also smoothed using a Sigmoid function, and a minimum adjustment coefficient is set to ensure that the threshold is not too low. 0.3, to obtain the signal quality threshold.

[0086] Further, the product of the reference threshold, the environmental threshold and the signal quality threshold is taken as the first threshold.

[0087] Step two, each identified peak is assigned a corresponding threshold, and based on the first threshold, a second threshold is calculated in combination with a set peak weight.

[0088] Specifically, the weight is assigned according to the amplitude ratio of each peak value, i.e. the ratio of the predicted amplitude of the current peak value to the predicted amplitude of the total peak value, and a stronger peak value obtains a higher weight, and then based on the first threshold and the peak weight, a second threshold is calculated The threshold is lowest near the peak position and gradually increases to both sides, represented as: ; Wherein, represents the threshold of the kth peak at position (x, y) on the dth day of the year, is the first threshold, is the peak weight, and d is the dth day of the year, is the predicted position of the kth peak, i.e. the number of days in the growth cycle, is the standard deviation of the Gaussian function, which is 20.

[0089] Step three, based on the predicted position of the peak , the peak time range is determined, and by taking the range of ±2 standard deviations of the Gaussian distribution characteristics, the expected start time of the kth peak and the expected end time of the kth peak are obtained.

[0090] Step four, considering that phytoplankton growth is generally faster than decline, a growth-decline asymmetric threshold adjustment strategy is developed, the growth period threshold gradually increases from the trough, the decline period threshold gradually increases from the peak position, and the decline period slope is more gentle.

[0091] Specifically, the asymmetric threshold function is set, represented as: ; ; Wherein, is the threshold of the kth peak in the growth period, is the threshold value for the kth peak in the growth phase, is the adjustment coefficient, controlling the strength of the time-varying threshold value, taking a value in the range of 0-1, and taking a value of 0.4 in the embodiment, is the expected start time of the kth peak, is the expected end time of the kth peak.

[0092] It should be noted that the present step is based on the peak threshold value as the basis, and the asymmetry is adjusted according to the growth stage. For example, at the beginning of the growth phase, the threshold value is ; at the peak position, that is, , the threshold value is reduced by 40% ( =0.4), which can ensure the accurate capture of SOS (growth start day) and EOS (growth end day).

[0093] Step five, determining the final threshold value based on the asymmetric threshold function, extracting phenological parameters from the fitted curve, including the SOS and EOS of each peak, and the overall growth period phenological parameters.

[0094] Specifically, for single-peak phenological parameter extraction, for each peak k, the date when the FAC first exceeds the growth phase threshold value in the growth phase is taken as , and the date when the FAC last exceeds the recession phase threshold value in the recession phase is taken as , if there is no d that meets the condition, then and are set to NaN.

[0095] The phenological parameters of multiple peaks are integrated into overall growth season parameters, and the main peak with the maximum biomass is identified, wherein the start day of the entire growth season is the date when the growth start day of the single peak is the smallest, the end day of the entire growth season is the date when the growth end day of the single peak is the largest, the total growth duration is the end day of the entire growth season minus the start day of the entire growth season, and the main peak is obtained by calculating the amplitude parameter corresponding to each peak multiplied by the maximum value of the growth duration. Finally, the number of days in the growth period is calculated from the split point, converted back to the annual accumulated day, and the final key phenological parameters of the 250m pixel are obtained.

[0096] It should be noted that the present application solves the problem of unstable quality of high-resolution data by establishing a multi-scale parameter migration from a 5km reference unit to a 250m pixel, considers environmental factors, accurately identifies multiple peak events through a multi-peak Gaussian process model, establishes a dynamic function relationship between environmental variables and growth parameters, ensures the asymmetry of the growth-recession process through parameter constraints, and can more truly reflect the life cycle of phytoplankton.

[0097] The above is a schematic scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to the embodiment. It should be noted that the technical scheme of the lake phytoplankton phenology parameter extraction system based on multi-scale parameter migration and dynamic threshold belongs to the same concept as the technical scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold described above. The technical details of the lake phytoplankton phenology parameter extraction system based on multi-scale parameter migration and dynamic threshold in this embodiment are not described in detail, and can be referred to the description of the technical scheme of the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold described above.

[0098] Embodiment 3 provides a lake phytoplankton phenology parameter extraction system based on multi-scale parameter migration and dynamic threshold, comprising: The acquisition processing module is configured to obtain remote sensing image data of a target region, and pre-process the remote sensing image data to construct multi-scale time series image data. The mapping module is configured to divide the growth cycle of the plant based on the multi-scale time series image data, to identify the growth cycle, construct a growth curve reference model and a parameter mapping model, and obtain the phenology morphological parameters of the plant in the growth cycle. The output module is configured to perform pixel-level growth cycle segmentation on the original pixel sequence of the image data, and identify and extract the key phenology parameters of the growth cycle based on the trained parameter mapping model and the preset dynamic threshold.

[0099] The embodiment also provides a computer device suitable for lake phytoplankton phenology parameter extraction based on multi-scale parameter migration and dynamic threshold, comprising: The memory is configured to store computer executable instructions, and the processor is configured to execute the computer executable instructions to implement the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to the above embodiment.

[0100] The embodiment also provides a storage medium having a computer program stored thereon, which is executed by a processor to implement the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold according to the above embodiment.

[0101] The storage medium proposed in the embodiment belongs to the same inventive concept as the lake phytoplankton phenology parameter extraction method based on multi-scale parameter migration and dynamic threshold proposed in the above embodiment. The technical details not described in detail in this embodiment can be referred to the above embodiment, and the embodiment has the same beneficial effects as the above embodiment.

[0102] From the above description of the embodiments, those skilled in the art can clearly understand that the present application can be realized by means of software and necessary general hardware, and of course can also be realized by hardware, but in many cases the former is a better embodiment. Based on such understanding, the technical solutions of the present application can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a floppy disk, a read-only memory (ROM), a random access memory (RAM), a FLASH memory, a hard disk, or an optical disc, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to perform the methods of various embodiments of the present application.

[0103] Embodiment 4, refer to Table 2 and Figure 3 For an embodiment of the present application, a lake phytoplankton phenology parameter extraction method based on multi-scale parameter transfer and dynamic threshold is provided, in order to verify its beneficial effects, scientific demonstration is carried out through economic benefit calculation and simulation experiment.

[0104] In order to better evaluate the performance improvement of the present application, four comparison methods C1-C4 are proposed for comparison, as shown in Table 2, the data are FAC time series extracted based on MODIS-Aqua resolution images, and three amplitude thresholds 10%, 20%, and 50% are used to determine SOS and EOS, as follows: Comparison method 1 uses the fitting method proposed in the present application to smooth each year's time, and only replaces the active amplitude threshold with the fixed three amplitude thresholds to extract the parameters; Comparison method 2 calculates the average value of FAC in the time window with a unit of 10 d, if there are missing values in a window, the average value of the window is the missing state, the time series is compressed to 1 / 10 of the original length, the window value is filled by linear interpolation, then the sequence is smoothed by harmonic filtering, and then the time series is restored to the original length. Since it is difficult to smooth the time series of each year into the form of strictly first monotonically increasing and then monotonically decreasing, the maximum value in each year after filtering is taken as the maximum amplitude A, the minimum value on the left side of A is taken as the left baseline value b L , and the minimum value on the right side of A is taken as the right baseline value b R , the time corresponding to the first time the amplitude threshold is reached on the left side is SOS, and the time corresponding to the last time the amplitude threshold is reached on the right side is EOS; Comparison method 3 is consistent with the operation process in method 2, and the harmonic filtering is replaced by Gaussian filtering to smooth the time series; Method 4, using the improved Bayesian land surface phenology model, first uses a double logistic function to fit the FAI time series based on the Bayesian hierarchical model, and then uses a fixed three amplitude threshold to extract SOS and EOS.

[0105] Table 2: Comparison method summary table Stage Comparison method 1 Comparison method 2 Comparison method 3 Comparison method 4 Data and characterization index FAC time series of 250m FAC time series of 250m FAC time series of 250m FAC time series of 250m Reconstruction time series Method of the present application HANTS Gaussian filtering BLSP fitting Extract phenological parameters Fixed amplitude threshold Fixed amplitude threshold Fixed amplitude threshold Fixed amplitude threshold By integrating the start and end times of different years in the sampling pixels as true values and comparing them with the model extraction parameters, the performance of the algorithm is evaluated by R², absolute percentage difference (APD), unbiased absolute percentage difference (UAPD), bias (Bias), root mean square error (RMSE), and unbiased relative percentage root mean square error (URMSE).

[0106] The experimental results are shown in Figure 3 The selected 50 sampling points cover different years, and a total of 330 sets of phenology true values are used to verify the extraction accuracy. The N value is the effective extraction value of the phenology parameter, indicating the total number of successfully extracted corresponding phenology parameters after reconstructing the time series. The optimal fixed threshold values for comparison methods 1-4 are 15%, 50%, 50%, and 15%, respectively, and the scatter plot shows the optimal fixed threshold results of the comparison methods. As can be seen from the scatter plot, the method of the present application performs best among all comparison methods, with the determination coefficient R² of the extracted start date (SOS), end date (EOS), and growing season duration (LOS) reaching 0.90, 0.64, and 0.78, respectively, significantly better than other comparison methods.

[0107] The extraction accuracy of comparison methods 1 and 4 is second, with R² between 0.40 and 0.60. The optimal fixed threshold values of these two methods are both 15%. Among them, the results of comparison method 1 using the reconstructed time series method proposed by the present application are slightly better than those of comparison method 4 using the LSP method, and the effective extraction value (N) of each parameter of comparison method 1 is higher than that of comparison method 4, which reflects that the reconstructed time series method proposed by the present application has higher robustness. The accuracy of comparison method 2 based on harmonic decomposition and comparison method 3 based on Gaussian filtering is relatively low, with R² of each phenology parameter being less than 0.5 and the effective extraction value (n) being lower than that of the other two comparison methods.

[0108] The main reason for the low accuracy of methods 2 and 3 is the specific climate conditions of the study area (Taihu and Chaohu). The Meiyu season lasts for about 1-2 months every year, resulting in large-scale systematic data missing in remote sensing observations. Methods such as harmonic analysis and Gaussian filtering are highly dependent on the continuity of the time series and are extremely sensitive to such long-term data gaps, making it difficult to accurately reconstruct complete phenology curves, thereby introducing large deviations in the extraction of phenology parameters.

[0109] It should be noted that the above examples are only used to illustrate the technical solutions of the present application but not to limit the present application. Although the present application is described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present application can be modified or equivalently replaced, without departing from the spirit and scope of the technical solutions of the present application, which should be covered in the scope of the claims of the present application.

Claims

1. A method for extracting phenological parameters of lake phytoplankton based on multi-scale parameter migration and dynamic thresholding, characterized in that, include: Acquire remote sensing image data of the target area, and preprocess the remote sensing image data to construct multi-scale time series image data; Based on the multi-scale time series image data, the plant growth cycle is divided to identify the growth cycle, and a growth curve benchmark model and parameter mapping model are constructed to obtain the phenological morphological parameters of the plant during the growth cycle. The original pixel sequence of the image data is segmented into pixel-level growth cycles, and based on the trained parameter mapping model, key phenological parameters of the growth cycle are identified and extracted through a preset dynamic threshold.

2. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 1, characterized in that, Acquire remote sensing image data of the target area and preprocess the remote sensing image data, including: Collect raw remote sensing image data of the target area at different spatial resolutions, and extract the algae coverage features from the raw remote sensing image data; Global surface water data is collected, and a region is selected from the global surface water data as a reference water body mask. A dynamic water body mask algorithm is constructed using multispectral index and reflectance threshold to remove non-water body pixels and interfering water body pixels from the remote sensing image data, thereby obtaining the effective water body pixels in the remote sensing image data. The daily algae coverage index is calculated for the effective water body pixels, and multi-scale aggregation is performed on remote sensing image data with different spatial resolutions to obtain the algae coverage time series.

3. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 2, characterized in that, Multi-scale aggregation of remote sensing image data at different spatial resolutions was performed to obtain the time series of algae cover, including: The original remote sensing image data with different spatial resolutions are divided into high-resolution and low-resolution image data. After extracting the algae coverage features, high-resolution original algae coverage time series and low-resolution algae coverage time series are obtained. The low-resolution image data is divided into several grid cells of the same size. The daily algae coverage value is determined based on the proportion of effective pixels in the grid cells, thus obtaining the low-resolution algae coverage time series. The low-resolution algae coverage time series was reconstructed by sequence difference to obtain the complete low-resolution algae coverage time series.

4. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 2, characterized in that, Obtaining phenological parameters of plants during their growth cycle includes: Based on the algae coverage index, the growth cycle of the low-resolution algae coverage time series is segmented to obtain multiple independent growth cycles. Based on the independent growth cycle, curve fitting is performed on the low-resolution algae coverage time series of the independent growth cycle to obtain the fitting parameter set for each growth cycle. A parameter mapping model is constructed, and the model is trained using the parameter set of the fitting function to establish a mapping relationship between low-resolution data and high-resolution data.

5. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 4, characterized in that, Constructing the parameter mapping model includes: Based on the high-resolution original algal bloom coverage time series of the target area within each growth cycle, multiple statistical features of each pixel in the series are calculated. A parameter mapping model is constructed using a machine learning model. The high-resolution statistical features are used as the model input, and the low-resolution sequence of the fitted parameter set is used as the model output. The parameter mapping model is trained to establish the mapping relationship between the multiple statistical features and the fitted parameter set, so as to obtain the prediction fitting parameters of the high-resolution original algal cover time series.

6. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 2, characterized in that, Identifying and extracting key phenological parameters for the growth cycle includes: By using a sliding window, the phenological segmentation index of the high-resolution original algal cover time series is calculated to divide the growth cycle. Based on each growth cycle segmented by each high-resolution pixel, the statistical features are extracted and input into the parameter mapping model to predict the fitting parameter set for each growth cycle of the pixel, thereby obtaining the fitting time series of the cycle. Based on the maximum amplitude of each growth cycle, the optimal phenological threshold is dynamically calculated, and the time data of plant growth are obtained from the fitted time series.

7. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 6, characterized in that, Calculate the phenological segmentation index of the high-resolution original algal cover time series to divide the growth cycle, including: The high-resolution original algal bloom coverage time series for each year are aligned by year-to-day. The mean, standard deviation, and maximum value of algae coverage at each location on each day of the year are calculated using a preset sliding window and then fused to obtain the phenological segmentation index. The day corresponding to the minimum value of the phenological segmentation index is selected as the annual growth cycle segmentation point of the pixel. That is, the annual growth cycle of the pixel starts from the day of the year and ends on the day of the year of the following year.

8. The method for extracting lake phytoplankton phenological parameters based on multi-scale parameter migration and dynamic thresholding as described in claim 7, characterized in that, Dynamically calculate the optimal phenological threshold and obtain plant growth time data from the fitted time series, including: The maximum amplitude determined by the set of fitting parameters is used to construct a functional relationship between the maximum amplitude and the optimal active amplitude threshold. The optimal function form and function parameters are determined by the least squares method, and the optimal active amplitude threshold is calculated. Based on the optimal activity amplitude threshold, from the fitted time series, the date on which the algal bloom coverage value first rises above the threshold from the left baseline during the growth period is selected as the growth start date, and the date on which the algal bloom coverage value last falls to the threshold during the decline period is selected as the growth end date.

9. A system for extracting phenological parameters of lake phytoplankton based on multi-scale parameter migration and dynamic thresholding, employing the method for extracting phenological parameters of lake phytoplankton based on multi-scale parameter migration and dynamic thresholding as described in any one of claims 1 to 8, characterized in that, include: The acquisition and processing module is used to acquire remote sensing image data of the target area and preprocess the remote sensing image data to construct multi-scale time series image data. The mapping module is used to divide the plant growth cycle based on the multi-scale time series image data, so as to identify the growth cycle, construct a growth curve benchmark model and a parameter mapping model, and obtain the phenological morphological parameters of the plant during the growth cycle. The output module is used to perform pixel-level growth cycle segmentation on the original pixel sequence of the image data, and based on the trained parameter mapping model, identify and extract key phenological parameters of the growth cycle through a preset dynamic threshold.

Images