Method and device for monitoring change of salt marsh type based on phenology time window and storage medium

Abstract

The application discloses a kind of based on phenology time window salt marsh type change remote sensing monitoring method, equipment and storage medium, the method is aimed at the monitoring demand of salt marsh vegetation type time sequence dynamic change under the influence of high-intensity human activity, adopts kernel density estimation method to combine sample weight and filters out the index and its corresponding phenology time window with optimal ability to distinguish salt marsh vegetation type;And based on the optimal index in the optimal phenology time window, construct pixel-level time series index, carry out change point detection and time sequence segmentation;Through fourier expansion model, the trend and periodicity characteristics of the sequence after segmentation are extracted, and a comprehensive feature vector is constructed;Finally, machine learning is used to complete the time sequence classification and identification of salt marsh vegetation type;The method has the advantages of high degree of automation, strong adaptability, etc., and can be widely applied to coastal ecological restoration evaluation and other scenes, has important practical value, and is an innovative application of remote sensing information technology in the field of coastal ecology monitoring.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of remote sensing technology application and coastal wetland monitoring, and relates to a salt marsh type time series change remote sensing monitoring method based on multi-spectral remote sensing images in a phenological time window. BACKGROUND

[0002] Traditional salt marsh vegetation field investigation methods usually require a large amount of manpower and economic cost, and may cause damage to the wetland ecosystem, and it is difficult to meet the timeliness and continuity requirements of salt marsh type dynamic change monitoring. In particular, in areas that are difficult to access, such as marshes and mudflats, it is more difficult to obtain data in a timely and effective manner using traditional methods. In contrast, as a non-contact observation method, remote sensing technology has the advantages of macroscopic, rapid and continuous monitoring, and can obtain spectral information, spatial information and texture information of the ground surface, and has become an important technical means for monitoring changes in salt marsh types.

[0003] Current salt marsh vegetation remote sensing monitoring mostly uses multi-temporal remote sensing images or remote sensing time series data with sparse time phases, and the observation frequency is limited and the continuity is insufficient. In particular, in areas significantly affected by tidal flooding and salinity gradients, the monitoring accuracy is more obviously limited. Specifically, salt marsh vegetation remote sensing classification faces problems such as similar species spectra, large phenological rhythm differences, and redundant time series data, resulting in high classification difficulty and limited accuracy of traditional methods. In view of the above difficulties, the present application proposes a remote sensing index and phenological window joint screening method based on kernel density estimation, which improves the monitoring ability of salt marsh type time series change by clearly defining the remote sensing index selection and key phenological window identification mechanism, and has the advantages of high automation, high accuracy and easy promotion. SUMMARY

[0004] The present application aims to overcome the deficiencies of the prior art and provide a salt marsh type change remote sensing monitoring method based on a phenological time window. This method automatically identifies and distinguishes the most effective remote sensing index for vegetation type and its corresponding phenological window, combines variable point detection and Fourier feature extraction, and uses a random forest classifier to monitor the time series change of salt marsh types.

[0005] The present application is implemented by the following technical solutions:

[0006] A salt marsh type change remote sensing monitoring method based on a phenological time window, comprising the following steps:

[0007] (1) According to the research time and area, the ground reflectance data of multi-spectral remote sensing images are selected, and the remote sensing index is calculated;

[0008] (2) Calculate the overlap rate of different vegetation types in different time phases of each remote sensing index using kernel density estimation, and calculate the weighted overlap rate combined with the sample size to screen out the remote sensing index with the best ability to distinguish salt marsh types and the corresponding best phenology time window;

[0009] (3) Based on the screened best index and the best phenology time window, construct a per-pixel time series remote sensing index, and perform change point detection and time series segmentation;

[0010] (4) For the segmented unequal-length time series data, based on the time series model of Fourier expansion, extract periodic and trend feature parameters, and calculate statistical features to form a comprehensive feature vector;

[0011] (5) Based on the comprehensive feature vector obtained in step (4) and the known type of salt marsh plant sample, use a machine learning classifier to realize the time series classification of salt marsh vegetation types.

[0012] In the above technical solution, further, the remote sensing index is: normalized difference vegetation index NDVI (Normalized Difference Vegetation Index), enhanced vegetation index (EVI), MSAVI (modified soil-adjusted vegetation index).

[0013] Further, in step (2), for each remote sensing index, the overlap rate in different time phases is calculated based on the kernel density estimation function of two vegetation types; first, in the studied time phase range, the minimum and maximum values of the remote sensing index data corresponding to the two vegetation types are determined to generate a uniformly distributed remote sensing index value range; then, for each value, the minimum value of the kernel density estimation function of the two vegetation types is calculated, that is, the height of the overlap area; by numerically integrating the height of the overlap area, the overlap area is obtained, and the total area of the two kernel density estimation functions is also calculated. The ratio of the overlap area to the total area is the overlap rate, which quantifies the distribution overlap degree of the two vegetation types in the time phase range.

[0014] Further, in step (2), the sample size is introduced as a weighting coefficient to weight the overlap rate of each time phase, and the weighted overlap rate of each remote sensing index is calculated. Specifically: divide the studied time phase range into time segments with N days as a time segment, and in each time segment, calculate the overlap rate of the two vegetation types, and take the total number of data points of the two vegetation types in the time segment as the weight. By multiplying the overlap rate of each time segment by its weight and accumulating, the total weighted overlap rate is obtained. Finally, the total weighted overlap rate is divided by the total weight to obtain the normalized weighted overlap rate.

[0015] Further, in step (2), the weighted overlap rates of each remote sensing index in each study phase range are compared, and the remote sensing index with the lowest overlap rate and the corresponding observation phase are determined as the best index and the best phenology time window that can best distinguish vegetation types.

[0016] Further, in step (3), the BinSeg algorithm is used for change point detection and time series segmentation, wherein the minimum effective observation value is set to 12, and the minimum time span is set to 192 days; time periods that do not meet the condition are merged into adjacent segments with similar spectral characteristics.

[0017] Further, in step (4), for the segmented unequal-length remote sensing time series, a time series model in the form of Fourier expansion is used for each sequence segment to extract periodic and trend feature parameters, including baseline value, linear trend item, harmonic component, and combined with the statistical characteristics of the maximum value, minimum value, mean value and variance of each sequence segment, a comprehensive feature vector containing dynamic change information is constructed for subsequent classification model input.

[0018] The application further provides an electronic device, comprising:

[0019] one or more processors;

[0020] a memory for storing one or more programs;

[0021] When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of the above.

[0022] The application further provides a computer-readable storage medium storing computer-executable instructions, which when executed, are used to implement the method of any one of the above.

[0023] The application has the following beneficial effects:

[0024] The present application adopts the kernel density estimation method to combine the sample weight to screen out the remote sensing index with the best ability to distinguish the salt marsh vegetation type and the corresponding phenological time window; and based on the optimal remote sensing index in the optimal phenological time window, a pixel-level time series index is constructed, the BinSeg algorithm is used for change point detection and time series segmentation; further, the trend and periodicity characteristics of the segmented sequence are extracted through the Fourier expansion model, and a comprehensive feature vector is constructed combined with the statistical parameters; the machine learning completes the time series classification and identification of the salt marsh vegetation type. Through the above workflow, the remote sensing monitoring of the time series change of the salt marsh vegetation type can be completed, and the distribution of the salt marsh type year by year is obtained. The present application aims at the monitoring demand of the time series dynamic change of the salt marsh vegetation type under the influence of high-intensity human activities, and constructs a salt marsh type time series change remote sensing monitoring method based on the phenological time window vegetation index. The method has the advantages of high automation degree and strong adaptability, and can be widely applied to the scene of coastal ecological restoration evaluation, and has important practical value. It is an innovative application of remote sensing information technology in the field of coastal ecological monitoring. BRIEF DESCRIPTION OF DRAWINGS

[0025] Figure 1 The method flowchart of one specific embodiment of the present application;

[0026] Figure 2 The weighted overlay rate map calculated based on NDVI in one specific embodiment of the present application (the red box represents the position of the minimum weighted overlay rate);

[0027] Figure 3 The weighted overlay rate map calculated based on EVI in one specific embodiment of the present application (the red box represents the position of the minimum weighted overlay rate);

[0028] Figure 4 The weighted overlay rate map calculated based on MSAVI in one specific embodiment of the present application (the red box represents the position of the minimum weighted overlay rate);

[0029] Figure 5 The figure of the area change of the salt marsh vegetation type in the research area of one specific embodiment of the present application year by year. DETAILED DESCRIPTION

[0030] The technical solutions of the present application will be further described in detail below in combination with the drawings and specific examples.

[0031] According to the salt marsh type change remote sensing monitoring method based on the phenological time window of the present application, the technical roadmap of the method is as shown in Figure 1 According to one specific embodiment of the present application, the method comprises the following steps:

[0032] (1) Obtain the multispectral image surface reflectance data in the study area and the study period. In the example study area, Landsat multispectral data are used as the data source. Landsat data are ideal for monitoring long-term changes in salt marsh wetlands due to their long time series. In this example, all Landsat Collection2 surface reflectance Tier 1 data covering the study area from 1986 to 2023 are used, a total of 2031 images, with a spatial resolution of 30 meters. Time series remote sensing vegetation index calculation is carried out based on the obtained remote sensing data. The method for calculating the index is:

[0033]

[0034]

[0035]

[0036] where NIR is the near-infrared band reflectance, and Red is the red band reflectance.

[0037] (2) Use kernel density estimation to calculate the overlap rate of different vegetation types at different time phases of each remote sensing index, and introduce the number of samples as a weighting coefficient to calculate the total weighted overlap rate. According to the principle of minimum total weighted overlap rate, the best remote sensing index and its corresponding best phenology time window are selected. The specific method is as follows:

[0038] Kernel density estimation is a non-parametric probability density function estimation method that can estimate the probability density distribution of continuous variables from limited data samples. Given a set of data points , the formula for kernel density estimation is:

[0039]

[0040] where is the kernel function (usually a Gaussian kernel function), is the bandwidth parameter, which controls the smoothing degree of the kernel function, is the number of data points.

[0041] Then, by calculating the minimum value integral of the two density curves in the same index value range, the overlap rate at this time phase is obtained, reflecting the spectral similarity of the two types of vegetation. The calculation of the overlap rate is based on the comparison of the kernel density estimation functions of the two types of vegetation. First, for each remote sensing index, determine the minimum and maximum values of the two types of vegetation data at all time phases to generate a uniformly distributed value range. Then, for each The minimum value of the two kernel density estimation functions is calculated, which represents the height of the overlap area. By numerically integrating the height of the overlap area, the overlap area is obtained. Meanwhile, the total area of the two kernel density estimation functions is calculated. Finally, the overlap rate is defined as the ratio of the overlap area to the total area, which is represented by the formula:

[0042]

[0043] wherein, represents the overlap rate, and are the kernel density estimation functions of the two vegetation types, respectively. Through this method, the degree of distribution overlap between the two vegetation types in the time series can be quantified.

[0044] To improve the objectivity of the evaluation, the number of samples is further introduced as a weighting coefficient to weight the overlap rates of each phase, and the weighted overlap rate (WOR) of each index is calculated. Specifically, the time series is segmented into 10-day time periods, and within each time period, the overlap rate of the two vegetation types is calculated, and the total number of data points of the two vegetation types in the time period is used as the weight. By multiplying the overlap rate of each time period by its weight and accumulating, the total weighted overlap rate is obtained. Finally, the total weighted overlap rate is divided by the total weight to obtain the normalized weighted overlap rate. The formula is represented as:

[0045]

[0046] wherein, is the weighted overlap rate, is the overlap rate of the th time period, is the weight of the th time period. This weighting method can effectively reflect the influence of the number of data points in different time periods on the overlap rate, thereby improving the accuracy and reliability of the analysis.

[0047] Then, by comprehensively comparing the weighted overlap rates of each remote sensing index at each phase, the remote sensing index with the lowest overlap degree and its corresponding observation phase are determined according to the principle of minimum weighted overlap rate, i.e., the best index and the best phenology time window that best distinguish between vegetation types.

[0048] The salt marsh vegetation types generally include the following: reed, sea rush, salt bush, and mutual flower rice. According to the actual situation, two vegetation types that need to be distinguished in the study area are selected. If the vegetation types in the study area exceed two, the above method is iterated to extract each type.

[0049] (3) Based on the selected optimal index and optimal phenological time window, construct the per-pixel remote sensing index time series data, and use the BinSeg algorithm for change point detection and time series segmentation. The specific method is as follows: First, based on the selected optimal remote sensing index of the optimal phenological time window, for each pixel, construct the time series remote sensing index. Then, based on the BinarySegmentation (BinSeg) change point detection algorithm (reference Truong C., Oudre L., Vayatis N. Selective review of offline change point detection methods [J]. Signal Processing, 2020, 167, i.e. Selective review of offline change point detection methods [J]. Signal Processing, 2020, 167.), segment the remote sensing index time series. This method divides the time series into multiple sub-sequences with internal consistency by recursively finding the optimal change point, thereby identifying the vegetation change process. The purpose of change point detection is to divide the time series into multiple sub-sequences with similar spectral characteristics, thereby more clearly revealing the dynamic change pattern of vegetation. In order to ensure that each sub-sequence obtained after change point detection has sufficient observations, this study sets two key parameters to control the rationality of the time series segmentation results: First, set the minimum effective observation value of each sub-sequence to 12 to ensure that each segment has sufficient sample size; second, set the minimum time span to 192 days. For time series fragments with less than 12 observations or less than 192 days of time interval, this study will merge them into adjacent and spectrally similar segments. After merging, the time span of the shortest time series segment is 192 days, and the time span of the longest time series segment can reach 13306 days. By introducing these constraints, the instability of the segmentation results caused by insufficient observations is effectively avoided, while ensuring that each sub-sequence can fully reflect the change characteristics of the time series.

[0050] (4) For the segmented unequal-length time series data, this study constructs a time series model based on Fourier expansion to extract periodic and trend feature parameters, and calculates statistical features, thereby constructing the comprehensive feature vector required for classification. The specific method is as follows: First, based on the sub-sequences obtained in step (3), use the Fourier expansion form of the time series model to extract features, and represent each sub-sequence as a combination of multiple harmonic components and linear trends. The time series model formula is as follows:

[0051]

[0052] Where:

[0053] is the number of days from the start date of the time series (01 / 01 / 1986) to the current date;

[0054] is the time frequency of the harmonic component ( );

[0055] is the period length, which is set to 365 days in this study to reflect the annual periodicity;

[0056] is the constant term, representing the baseline value of the time series;

[0057] and are the intra-annual variation coefficients, representing the cosine and sine coefficients of the th harmonic component, respectively, which are used to capture the periodicity of the time series;

[0058] is the inter-annual variation coefficient, representing the slope of the linear trend term, which is used to describe the long-term trend of the time series;

[0059] is the remote sensing index predicted by the model on the th day since the start date of the time series (01 / 01 / 1986).

[0060] Based on the fitting results of the above time series model, the periodic model feature parameters are extracted, including the constant term (intercept), linear trend coefficient (slope), cosine coefficients of the three harmonic components (Cos1, Cos2, Cos3), and sine coefficients (Sin1, Sin2, Sin3). These parameters can accurately describe the baseline level, long-term trend, and intra-annual periodicity of the time series. Meanwhile, based on the harmonic coefficients, derived parameters such as amplitude and phase are further derived, and the root mean square error of the model is calculated as the goodness-of-fit indicator. These model features together constitute the key indicators that describe the periodicity, trend, and volatility of the time series.

[0061] In terms of statistical features, descriptive statistics such as the maximum, minimum, mean, and variance of the remote sensing index are calculated for each sub-time series. These features can reflect the numerical distribution range, central tendency, and dispersion degree of the time series.

[0062] Combining the feature parameters extracted from the above time series model segmentation (such as baseline value, trend term, harmonic coefficient, etc.), and the statistical features of each sub-time series (maximum, minimum, mean, variance, etc.), together as a comprehensive feature vector for the input of the next classification model.

[0063] (5) Using a machine learning classifier to carry out supervised classification of salt marsh vegetation types, visually interpreting 200 sample points for each of the different salt marsh vegetation types, using the comprehensive feature vector synthesized in step (4) as the classification feature, and carrying out supervised classification based on machine learning for each sub-time sequence obtained in step (3) to obtain annual salt marsh type distribution time sequence data. In this example, a random forest model is used, with the following parameters: maximum depth 10, minimum number of leaf nodes 1, minimum number of node splits 2, and number of decision trees 100.

[0064] Based on the above technical process, the monitoring of the temporal dynamic changes of salt marsh vegetation types is finally completed, and the method has strong stability and generalizability, and is suitable for large-scale salt marsh type remote sensing classification tasks.

[0065] The present application aims to meet the demand for monitoring the temporal dynamic changes of salt marsh vegetation types under the influence of high-intensity human activities, and constructs a salt marsh type temporal change remote sensing monitoring method based on the phenological time window vegetation index. This method does not require manual image screening and is suitable for large-scale, long-time sequence monitoring tasks, has high automation and strong adaptability, and can be widely used in coastal ecological restoration evaluation and other scenarios, has important practical value, and is an innovative application of remote sensing information technology in the field of coastal ecology monitoring.

[0066] Embodiment

[0067] In this embodiment, the Hangzhou Bay is taken as an example, and the salt marsh type temporal change remote sensing monitoring method based on the phenological time window is applied. The method includes the following steps:

[0068] The first step is to obtain Landsat image data of the Hangzhou Bay. In this example, all Landsat Collection 2 land surface reflectance Tier 1 data covering the study area from 1986 to 2023 are obtained based on Google Earth Engine, a total of 2031 images, with a spatial resolution of 30 meters. Based on the obtained remote sensing image data, the remote sensing indices NDVI, EVI, and MSAVI are calculated.

[0069] The second step is to calculate the overlap rate of different vegetation types in different time phases of each remote sensing index using kernel density estimation, and introduce the sample size as a weighting coefficient to calculate the total weighted overlap rate. According to the principle of minimum total weighted overlap rate, the best remote sensing index and its corresponding best phenology time window are selected. 200 sample points of Scirpus mariqueter and 200 sample points of Spartina alterniflora (including part of Phragmites australis) are selected on the south coast of Hangzhou Bay, a total of 400 sample points, to calculate the weighted overlap rate. The calculation results show that when the number of effective pixels is greater than or equal to 12, the weighted overlap rate of Scirpus mariqueter and Spartina alterniflora in the 160-360 time phase range is the smallest. Specifically, the weighted overlap rate of EVI is 1.50, the weighted overlap rate of MSAVI is 1.51, and the weighted overlap rate of NDVI is the smallest, only 0.50 (as shown in Figures 2-4 ). This shows that NDVI has high resolution in distinguishing between the two types of vegetation, and the best phenology window is 160-360 days.

[0070] The third step is to construct long time series NDVI data based on the NDVI index of the phenology time window (160-360 days) of each year, and use the BinSeg algorithm to detect change points and time series segmentation of the time series NDVI curve.

[0071] The fourth step is to construct a time series model based on Fourier expansion to extract periodic and trend feature parameters for the segmented unequal length time series data, and calculate statistical features to construct the comprehensive feature vector required for classification.

[0072] The fifth step is to construct a training dataset containing 200 Scirpus mariqueter, 200 Spartina alterniflora and 200 bare land sample points based on the visual interpretation results of high-resolution remote sensing images in 2021, and select time series segmentation containing 2021 as training samples. The parameters of the random forest model are set as maximum depth 10, minimum leaf sample size 1, minimum sample size for node splitting 2, and number of decision trees 100.

[0073] Based on the above technical process, the annual salt marsh type distribution results of Hangzhou Bay from 1986 to 2023 are obtained. As Figure 5As shown, the monitoring results show that the vegetation area of Spartina alterniflora and Spartina anglica in the study area presents a significant fluctuation trend from 1986 to 2022. Specifically, the area of Spartina alterniflora reached a peak in 1992, about 140 square kilometers, and dropped to a minimum in 2006, about 40 square kilometers. In contrast, the area of Spartina anglica increased significantly after 2018. Driven by reclamation engineering, the land area increased steadily from 1986 to 2022, while the light beach area decreased significantly. In particular, during 2002-2006, the areas of Spartina anglica and Spartina alterniflora decreased synchronously. In addition, due to the strong niche competition advantage of Spartina anglica, since 2008, it has shown a significant expansion trend on newly accreted light beaches. Based on the remote sensing monitoring results, precision verification based on high-resolution UAV images shows that the overall accuracy (OA) of the cover types in the Hangzhou Bay salt marsh wetland is 90%, and the kappa is 0.85, which reflects good monitoring accuracy.

[0074] The present application aims at the monitoring demand of temporal dynamic change of salt marsh vegetation types under the influence of high-intensity human activities, and constructs a salt marsh type temporal change remote sensing monitoring method based on the vegetation index of the phenological time window. The method has the advantages of high automation and strong adaptability, and can be widely applied to scenes such as coastal ecological restoration evaluation, and has important practical value. It is an innovative application of remote sensing information technology in the field of coastal ecology monitoring.

[0075] Those skilled in the art will appreciate that embodiments of the application can be provided as methods, systems, or computer program products. Accordingly, the application can take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the application can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROMs, optical storage media, etc.) having computer-usable program code embodied in the medium.

[0076] The present application is described with reference to flowcharts and / or block diagrams of the method, device (system), and computer program product according to the embodiments of the present application. It should be understood that each flow and / or block in the flowcharts and / or block diagrams, and the combination of the flows and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device produce a device that implements the flowcharts and / or block diagrams. Figure 1 one flow or multiple flows and / or blocks Figure 1means for performing the function specified in the block or blocks.

[0077] These computer program instructions can also be stored in a computer readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer readable memory produce an article of manufacture including instructions which implement the flow Figure 1 one or more flowcharts and / or blocks Figure 1 means for performing the function specified in the block or blocks.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the flow Figure 1 one or more flowcharts and / or blocks Figure 1 means for performing the function specified in the block or blocks.

[0079] The above-described embodiments are merely some of the preferred schemes of the present application, and are not intended to limit the present application. Those skilled in the art, without departing from the spirit and scope of the present application, can make various changes and modifications. Therefore, all technical schemes obtained by equivalent replacement or equivalent transformation shall fall within the scope of the present application.

Claims

1. A method for monitoring salt marsh type change by remote sensing based on phenology time window, characterized in that, It comprises the following steps: (1) According to the research time and area, the surface reflectance data of multispectral remote sensing image is screened out, and the remote sensing index is calculated; the remote sensing index is: normalized difference vegetation index NDVI, enhanced vegetation index EVI, improved soil adjusted vegetation index MSAVI; (2) The kernel density estimation is used to calculate the overlap rate of different vegetation types in each remote sensing index at different time phases, and the weighted overlap rate is calculated combined with the sample number, and the remote sensing index with the best ability to distinguish salt marsh type and the corresponding best phenology time window are screened out; the specific screening method is: by comprehensively comparing the weighted overlap rates of each remote sensing index in each research phase range, the remote sensing index with the lowest overlap degree and the corresponding observation phase are determined according to the principle of minimum weighted overlap rate, that is, the best index and the best phenology time window which can best distinguish the vegetation type are determined; (3) Based on the screened best index and the best phenology time window, a pixel-by-pixel time series remote sensing index is constructed, and the BinSeg algorithm is used for change point detection and time series segmentation, and the minimum effective observation value is set to 12 and the minimum time span is set to 192 days in the segmentation process; the time periods that do not meet the condition are merged into the adjacent segments with similar spectral characteristics; (4) For the segmented unequal length time series data, based on the time series model of Fourier expansion, the periodic and trend feature parameters are extracted, and the statistical features are calculated to form a comprehensive feature vector; specifically, for each segmented unequal length remote sensing time series segment, the time series model of Fourier expansion is used to extract the periodic and trend feature parameters, including baseline value, linear trend item, harmonic component, and the statistical features of the maximum value, minimum value, mean value and variance of each sequence segment remote sensing index are combined to form a comprehensive feature vector containing dynamic change information, which is used as the input of the subsequent classification model; (5) Based on the comprehensive feature vector obtained in step (4) and the known type of salt marsh plant sample, a machine learning classifier is used to realize the time series classification of salt marsh vegetation type.

2. The method for monitoring salt marsh type change by remote sensing based on phenology time window according to claim 1, characterized in that, In step (2), the overlap rate in different time ranges is calculated for each remote sensing index based on the kernel density estimation function of the two vegetation types; first, the minimum and maximum values of the remote sensing index data corresponding to the two vegetation types are determined in the studied time range, and a uniformly distributed remote sensing index value range is generated ; Subsequently For each value, the minimum of the two kernel density estimation functions, i.e. the height of the overlap region, is calculated; by numerically integrating the height of the overlap region, the overlap area is obtained, while the total area of the two kernel density estimation functions is calculated, the ratio of the overlap area to the total area is the overlap rate, which quantifies the degree of distribution overlap of the two vegetation types in the time phase range. 3.The method according to claim 1, wherein, In step (2), the sample number is introduced as a weighting coefficient to weight the overlap rate of each phase, and the weighted overlap rate of each remote sensing index is calculated. Specifically, the research phase range is segmented by N days as a time period, the overlap rate of two vegetation types is calculated in each time period, and the total number of data points of the two vegetation types in the time period is taken as the weight. The total weighted overlap rate is obtained by multiplying the overlap rate of each time period by its weight and accumulating. Finally, the total weighted overlap rate is divided by the total weight to obtain the normalized weighted overlap rate.

4. An electronic device, comprising: It comprises: One or more processors; Memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1-3.

5. A computer readable storage medium storing computer executable instructions, the instructions when executed to implement the method of any one of claims 1-3.

Images