NDSPI-based spartina alterniflora monitoring method

Abstract

The invention discloses a spartina alterniflora monitoring method based on NDSPI, and the method comprises the steps: firstly calculating a normalized phenological index of a time sequence through a multi-temporal Sentinel-2L2A image, and reconstructing a smooth phenological curve through time sequence harmonic analysis; the NDSPI is innovatively constructed through the combined calculation of the key phenological period, so that the difference between the spartina alterniflora and other salt marsh vegetation is enhanced. And then median filtering noise reduction is performed on the NDSPI image, and K-means + + unsupervised classification is adopted to efficiently extract spatial and temporal distribution of spartina alterniflora. The method has the advantages of full-process automation, convenient data acquisition and strong adaptability, and provides reliable technical support for coastal wetland invasive species monitoring and ecological protection.

Description

Technical Field

[0001] The invention belongs to the intersection of remote sensing technology and ecological monitoring, and in particular relates to a Spartina alterniflora monitoring method based on NDSPI. Background Art

[0002] Spartina alterniflora, a perennial grass, is a key target for ecological engineering projects due to its rapid growth and robust reproductive capacity. However, due to insufficient assessment of its invasive risk, its widespread expansion has severely damaged coastal wetland ecosystems. Its expansion has not only exacerbated the monoculture of wetland vegetation and reduced regional biodiversity, but has also altered intertidal hydrodynamics, impacting the habitats of commercial aquatic organisms such as fish and shrimp, and negatively impacting coastal fisheries and tourism. With the advancement of remote sensing technology, satellite and drone-based remote sensing imagery has become an important tool for large-scale wetland ecological monitoring due to its wide coverage, short update cycles, and low cost. In the past, academic and engineering practice have primarily used medium- to high-resolution optical satellite imagery (e.g., Landsat and Sentinel-2) or drone imagery, combined with supervised and unsupervised classification algorithms, to identify and map wetland vegetation distribution. Other approaches have also used radar data (e.g., Sentinel-1) fused with optical data to extract vegetation features through interferometry or dual-temporal differencing techniques. However, in salt marsh wetlands, where vegetation such as Phragmites australis and Suaeda salsa, whose ecological niches are similar to Spartina alterniflora, image classification methods relying solely on a single temporal phase or data source often struggle to distinguish Spartina alterniflora from non-Spartina alterniflora vegetation, resulting in high rates of false positives and missed detections.

[0003] To overcome the limitations of single image recognition, some studies have introduced time-series remote sensing images to identify species characteristics by extracting phenological information of vegetation indices (such as the Normalized Difference Vegetation Index (NDVI), the Soil Vegetation Index (SAVI), the Normalized Difference Water Index (NDWI), etc.) that change over time. Existing literature has shown that the growth cycle of Spartina alterniflora differs significantly from that of other salt marsh vegetation in terms of temporal phase, but existing studies usually use threshold-based phenological index extraction methods, which have a low degree of automation and rely on experience in threshold setting, making it difficult to generalize to different regions and sensors. At present, a number of invention technologies have been disclosed around remote sensing monitoring of Spartina alterniflora:

[0004] Patent application number CN202510072404.X discloses a method for monitoring the distribution of Spartina alterniflora based on radar-optical fusion. It uses Sentinel-1 synthetic aperture radar (SAR) data and Sentinel-2 multispectral data to achieve regional classification through the time series weighted dynamic time warping (TWDTW) model, and combines ecological factors and water factors to weight the generation of graded early warning areas. Then, regular drone aerial photography and target detection models are used to achieve dynamic monitoring of diffusion. Although this method can capture diffusion trends in real time and generate early warning reports, the technical process is complicated, and the dependence on multi-source data fusion and drone aerial photography is high, and the system implementation cost and maintenance difficulty are large. Patent publication number CN119540780A discloses a method for identifying and warning Spartina alterniflora based on convolutional neural networks (CNN). It builds a ground acquisition platform to obtain multiple Spartina alterniflora images, performs image splicing, segmentation and training set construction, and then trains the CNN model for the four seasons data respectively. Finally, the hazard level of Spartina alterniflora is evaluated based on the temporal segmentation results and a visual heat map is generated. This method has high recognition accuracy on a local scale, but requires a large number of manually labeled samples and a ground collection platform, and the model generalization ability is limited to the training area and is difficult to promote quickly. Patent Publication No. CN114782842A discloses a remote sensing monitoring method based on target detection, which detects objects on high-resolution remote sensing images through a vegetation distribution recognition model, analyzes the distribution pattern of Spartina alterniflora in combination with terrain information, and generates a dynamic timeline animation to achieve online interactive monitoring. This method improves the visualization level of monitoring, but relies on a complex geographic information system (GIS) and customized software, and the automation and lightweight degree of the overall solution still need to be improved. Patent Publication No. CN118411612A and CN118506183A disclose a Spartina alterniflora monitoring and management method based on a deep semantic segmentation model (such as DeeplabV3+) and a comprehensive management and supervision system. The former performs semantic segmentation and vectorized output patches through quarterly image preprocessing and CNN model training; the latter integrates remote sensing, deep learning, and database construction to achieve the establishment of a quarterly Spartina alterniflora management process database. The above methods have made progress in segmentation accuracy and system integration, but they are highly dependent on computing resources and the update frequency is only at the quarterly level, which makes it difficult to meet real-time monitoring needs.

[0005] Although existing technologies have achieved some success in monitoring Spartina alterniflora, due to the high similarity in spectral and spatial characteristics between Spartina alterniflora and surrounding salt marsh vegetation, monitoring methods using single-phase or single-source data have inherent limitations in its identification. Phenological indicators based on thresholds require empirical thresholds to be set for specific regions, which lacks universality. Segmentation and detection methods based on deep learning require a large amount of labeling manpower, and the models are not adaptable to different sensor specifications and imaging conditions. Furthermore, most methods are updated infrequently and lack the ability to respond in real time to the rapid invasion of Spartina alterniflora, making it difficult to provide sufficient and timely monitoring information for wetland ecological protection and rapid decision-making. Summary of the Invention

[0006] In response to the problems of poor universality, low degree of automation, and strong sample dependence in the existing threshold method for phenological identification, the present invention proposes an automated monitoring method for Spartina alterniflora based on the normalized difference phenological index of Spartina alterniflora. The time series normalized phenological index is calculated by multi-phase Sentinel-2L2A image data and smoothed and reconstructed using harmonic analysis to construct the normalized difference phenological index (NDSPI) image. Combined with 3×3 median filtering noise reduction and K-means++ clustering classification, the spatiotemporal distribution map of Spartina alterniflora is automatically generated, achieving a high-precision invasion monitoring effect with low cost investment, high automation operation, low sample dependence, and cross-regional migration application. The present invention proposes a monitoring method for Spartina alterniflora based on NDSPI, which specifically includes the following steps:

[0007] S1: Obtain time series remote sensing data of coastal salt marsh wetland areas: Obtain multi-temporal Sentinel-2L2A satellite surface reflectance image data of the area to be monitored, and calculate the time series normalized phenological index image. The calculation formula is:

[0008]

[0009] Among them, NDPI represents the time series normalized phenological index, NIR, Red, and SWIR are the reflectivity values ​​of Sentinel-2L2A satellite, respectively;

[0010] S2: Harmonic reconstruction of the time series normalized phenological index image and extraction of the phenological curve: The time series normalized phenological index image calculated in step S1 is reconstructed using the time series harmonic analysis algorithm to obtain the reconstructed time series image. Specifically, the harmonic fitting model is used for calculation, and the calculation formula is:

[0011]

[0012] Among them, y(t j ) is the original NDPI sequence, is the reconstructed NDPI sequence, ε(tj ) is the residual sequence; t j is the observation time of the original sequence, where j = 1, 2, ..., N, N is the maximum number of observations of the time series data column; the value of nf represents the frequency f i The number of relevant harmonic components; a i and b j is a frequency f i The trigonometric coefficients of the components, a0 is the coefficient of the zero-frequency term;

[0013] Based on the reconstructed time series images and combined with pre-selected vegetation sample points, the phenological curves of various vegetation types are extracted;

[0014] S3: Construction of normalized difference Spartina alterniflora phenological index image: Based on the various vegetation phenological curves extracted in step S2, the normalized difference Spartina alterniflora phenological index image is calculated and obtained. The calculation formula is:

[0015]

[0016] Among them, NDPI Sep.+Oct. NDPI is the average value of the normalized difference phenology index from September to October each year. Jul.+Aug. is the average value from July to August each year; then, the normalized difference Spartina alterniflora phenological index image was processed with a 3-pixel window-based median filter to obtain the denoised normalized difference Spartina alterniflora phenological index image;

[0017] S4: Automatic vegetation classification: The denoised normalized difference Spartina alterniflora phenological index image obtained in step S3 is automatically classified using the K-means++ algorithm to obtain a spatiotemporal distribution map of Spartina alterniflora.

[0018] As a technical preferred solution of the present invention, the satellite reflectivity in step S1 specifically includes: reflectivity values ​​of bands B8, B4 and B11.

[0019] As a technical preferred solution of the present invention, the phenological curves of various types of vegetation described in step S2 specifically include 4 different categories: phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis, and mixed vegetation.

[0020] As a preferred technical solution of the present invention, before calculating and obtaining the normalized difference Spartina alterniflora phenological index image in step S3, a phenological curve separability discrimination method is included. This method uses the J-M distance as a separability discrimination index for each type of vegetation phenological curve extracted in step S2, calculates the J-M distance between any two types of vegetation curves, and judges the separability of each type of vegetation curve based on the J-M distance value to ensure that the Spartina alterniflora curve and other salt marsh vegetation curves are significantly distinguishable in phenological characteristics. The separability JM distance is calculated as follows:

[0021] JM=2(1-e -B )

[0022]

[0023] Among them, B is the Bhattacharyya distance between the two categories, e ik represents the mean of different types of vegetation samples under the vegetation phenology curve; δ ik is the variance of different types of vegetation, where i = 1, 2, ..., 6 and k = 1, 2, ..., 4.

[0024] As a preferred technical solution of the present invention, the median filtering process based on a 3-pixel window in step S3 is performed as follows:

[0025]

[0026] Among them, I(x,y) represents the input image, represents the output image after filtering, W is a 3-pixel filtering window, and (i, j) represents the offset coordinates within the window relative to the center pixel (x, y).

[0027] As a technical preferred solution of the present invention, in step S4, the de-noised normalized difference Spartina alterniflora phenological index image obtained in step S3 is subjected to the following steps in sequence to generate a spatiotemporal distribution map of Spartina alterniflora:

[0028] S4-1: Take all pixel values ​​in the denoised normalized difference Spartina alterniflora phenological index image generated in step S3 as input, and use the K-means++ algorithm to randomly select four initial cluster centers according to the square ratio of the distance between the pixel and the selected cluster center. The specific calculation formula is:

[0029]

[0030] Among them, D(x i ) represents the data point x i The distance to the nearest selected cluster center;

[0031] S4-2: Starting from the four initial cluster centers obtained in step S4-1, a standard K-means iteration is performed on the pixel value set: each pixel is assigned to the nearest cluster center in each iteration, and the position of each cluster center is then recalculated. This process is repeated until the moving distance of all cluster centers is less than a preset threshold or the maximum number of iterations is reached;

[0032] S4-3: After completing one or more initialization and iteration processes consisting of steps S4-1 and S4-2, select a clustering result that minimizes the intra-class sum of squares, and stitch all pixels identified as Spartina alterniflora in the result in sequence according to their image acquisition time to output a spatiotemporal distribution map of Spartina alterniflora.

[0033] Compared with the prior art, the present invention has the following beneficial effects:

[0034] It has a high degree of automation and uses time series harmonic reconstruction and K-means++ clustering to replace empirical thresholds and a large amount of manual annotation, realizing the automation of the entire process from image acquisition to spatiotemporal distribution map generation, reducing human intervention and operation and maintenance costs.

[0035] The data cost is low, relying only on publicly available Sentinel-2L2A multi-temporal optical images. There is no need to purchase high-resolution or professional radar / UAV data, which greatly reduces data acquisition and processing costs.

[0036] The monitoring accuracy has been significantly improved. The normalized difference Spartina alterniflora phenological index (NDSPI) has magnified the numerical differences between Spartina alterniflora and other salt marsh vegetation during the phenological period, and the noise has been removed through median filtering to improve the accuracy and stability of classification.

[0037] It has strong anti-noise ability and introduces a 3×3 pixel median filter link on the NDSPI image to effectively suppress high-frequency noise such as cloud shadows and sensor artifacts, reducing the risk of misclassification, especially in the vegetation transition zone.

[0038] It has good generalization and migration capabilities, the harmonic reconstruction model adapts to different time series characteristics, and K-means++ clustering does not require prior sample labels. It can be quickly extended to different coastal salt marsh types and geographical regions, realizing Spartina alterniflora monitoring in multiple scenarios.

[0039] The method has strong timeliness. The 5-day temporal resolution of Sentinel-2 and the annual monitoring frequency of 73 scenes enable this method to capture the rapid expansion of Spartina alterniflora, supporting early invasion warning and dynamic change analysis. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flow chart of a method for monitoring Spartina alterniflora based on NDSPI provided by the present invention;

[0041] Figure 2 An operational flow chart of a method for monitoring Spartina alterniflora based on NDSPI is provided as an embodiment of the present invention;

[0042] Figure 3 The present invention provides a schematic diagram of the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation in 2022, which are extracted based on the reconstructed time series normalized phenological index image in one year;

[0043] Figure 4 A schematic diagram of the separability discrimination results of the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation provided in an embodiment of the present invention;

[0044] Figure 5 A schematic diagram of the spatiotemporal distribution of salt marsh vegetation classified using the K-means++ algorithm based on the normalized difference Spartina alterniflora phenological index image with high-frequency noise removed is provided in an embodiment of the present invention;

[0045] Figure 6 A schematic diagram of the accuracy evaluation of the spatiotemporal distribution results of salt marsh vegetation provided by an embodiment of the present invention;

[0046] Figure 7 The present invention provides the comparative experimental results of the 2022 Linhong River Estuary Wetland Vegetation Classification in an embodiment; wherein, (a) is the K-means++ classification result of calculating NDSPI based on NDVI, and (b) is the RF classification result with the NDSPI structural parameters unchanged. DETAILED DESCRIPTION

[0047] The present invention is further described below with reference to the accompanying drawings. However, the present invention can be implemented in many different ways and should not be construed as limited to the embodiments shown; rather, these embodiments provide those skilled in the art with implementation methods that meet applicable legal requirements.

[0048] Example 1: Specifically, Figure 1 This is a flow chart of a method for monitoring Spartina alterniflora based on NDSPI provided in an embodiment of the present application.

[0049] like Figure 2 As shown, the Spartina alterniflora monitoring method based on the normalized difference phenological index includes the following steps:

[0050] In step S101, the multi-phase Sentinel-2L2A satellite surface reflectance image data of the coastal salt marsh wetland area to be monitored is obtained, and the time series normalized phenological index image is calculated. The multi-phase Sentinel-2L2A satellite surface reflectance image data of the coastal salt marsh wetland area to be monitored is downloaded through the relevant platform. For example, the coastal salt marsh wetland can be set as the Haizhou Bay Linhong River Estuary Wetland, and the time series range is from 2019 to 2023. In the actual implementation process, the time series Sentinel-2L2A satellite surface reflectance image data of the Haizhou Bay Linhong River Estuary Wetland with cloud cover less than 10% from 2019 to 2023 can be obtained, and then the normalized phenological index (NDPI) of the obtained image is calculated. The NDPI calculation formula is as follows:

[0051]

[0052] NIR, Red, and SWIR are the reflectance values ​​of the B8, B4, and B11 bands of the Sentinel-2 satellite, respectively.

[0053] In step S102, the time series normalized phenological index image is reconstructed using a time series harmonic analysis algorithm, the vegetation phenological characteristics are analyzed, and a normalized difference Spartina alterniflora phenological index is proposed.

[0054] In an embodiment of the present application, a time series normalized phenological index image is reconstructed using a time series harmonic analysis algorithm, vegetation phenological characteristics are analyzed, and a normalized difference Spartina alterniflora phenological index is proposed, including: extracting phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis, and mixed vegetation within a year based on the reconstructed time series normalized phenological index image. The normalized difference Spartina alterniflora phenological index is proposed based on the phenological curve, and the normalized difference Spartina alterniflora phenological index image is generated.

[0055] Specifically, in the embodiments of this application, a time series harmonic analysis algorithm is used to decompose the time series NDPI image into the sum of a finite number of harmonics and a constant term. The harmonic coefficients are obtained through repeated least squares fitting, and then the reconstructed normalized phenological index image is obtained. The following formula is mainly used:

[0056]

[0057] Where y is the original NDPI sequence, is the reconstructed NDPI sequence, and ε is the residual sequence. j is the observation time of the original sequence, where j = 1, 2, ..., N, and N is the maximum number of observations (number of samples) of the time series data column.

[0058] It should be noted that, in the embodiment of the present application, the time resolution of acquiring the surface reflectance image data of Sentinel-2A is set to 5 days, and the time series length is calculated as one year, so the maximum number of observations N is 73. The value of nf represents the frequency f i The number of relevant harmonic components. a i and b j is a frequency f i The trigonometric coefficients of the components, a0 is the coefficient of the zero-frequency term. Usually, a harmonic sequence consists of a fundamental frequency signal and several signals that are integer multiples of the fundamental frequency.

[0059] In an embodiment of the present application, based on the reconstructed time series normalized phenological index image, the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation within one year are extracted.

[0060] Specifically, combining prior knowledge with visual interpretation, pixel points of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation were selected as samples within the monitoring area, and the mean curve of the normalized phenological index of the selected pixels was calculated for each year from 2019 to 2023, which is the phenological curve of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation. Figure 3 Shown is a schematic diagram of the phenological curve results of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation in 2022 according to the embodiment of the present application.

[0061] Optionally, in an embodiment of the present application, before constructing the normalized difference Spartina alterniflora phenological index, the method further includes: performing separability discrimination on the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation within the year.

[0062] Specifically, based on the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation in 2022, the JM distance was used as the discrimination index to calculate the separability of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation. The calculation formula of the JM distance is as follows:

[0063] JM=2(1-e -B )

[0064]

[0065] Where B is the Bhattacharyya distance between the two categories. ik Represents the mean of different types of vegetation samples under the vegetation phenology curve. Similarly, δ ik is the variance of different types of vegetation (i=1,2,...,6,k=1,2,...,4).

[0066] JM distance considers both the first-order statistical variables (mean) and the second-order statistical variables (covariance) at the same time. In the evaluation of multi-category separability, it has a good effect on quantitatively evaluating the separability of two types of statistical variables. Figure 4 Shown is a schematic diagram of the separability discrimination results of the phenological curves of Spartina alterniflora, Suaeda salsa, Phragmites australis and mixed vegetation in the embodiment of the present application.

[0067] In an embodiment of the present application, the normalized difference Spartina alterniflora phenological index is constructed according to the phenological curve, and the normalized difference Spartina alterniflora phenological index image is generated.

[0068] Specifically, the normalized difference Spartina alterniflora phenological index (NDSPI) is proposed based on the phenological curve, and the NDSPI formula is as follows:

[0069]

[0070] Among them, NDPI Sep.+Oct. NDPI is the average value of the normalized difference phenology index from September to October each year. Jul.+Aug. The index is the average value from July to August each year. This index further amplifies the differences between Spartina alterniflora and other salt marsh vegetation. The calculation formula of the normalized difference enhances the stability and robustness of the index, which is crucial for the identification of Spartina alterniflora.

[0071] The normalized difference Spartina alterniflora phenological index images of the monitoring area from 2019 to 2023 were generated according to the NDSPI calculation formula.

[0072] It should be noted that the number of normalized difference Spartina alterniflora phenological index images generated is one image per year, and a total of five images were generated from 2019 to 2023.

[0073] In step S103, the normalized difference Spartina alterniflora phenological index image with high-frequency noise removed is used to automatically classify coastal salt marsh vegetation using the K-means++ algorithm to obtain a high-precision spatiotemporal distribution of Spartina alterniflora.

[0074] Optionally, in an embodiment of the present application, the normalized difference Spartina alterniflora phenological index image is subjected to noise reduction processing to obtain a normalized difference Spartina alterniflora phenological index image from which high-frequency noise is removed;

[0075] Specifically, to improve clustering robustness, the normalized difference Spartina alterniflora phenological index image was subjected to a 3-pixel window median filter using the following formula:

[0076]

[0077] Among them, I(x,y) represents the input image, The output image after filtering is represented by W, where W is the 3-pixel filter window and (i, j) represents the offset coordinates within the window relative to the center pixel (x, y). Median filtering effectively reduces high-frequency noise caused by transient cloud shadows and sensor artifacts. This processing step minimizes the risk of misclassification in the ecological transition zone where Spartina alterniflora coexists with Suaeda salsa and Phragmites australis.

[0078] The normalized difference Spartina alterniflora phenological index image after high-frequency noise removal is input into the K-means++ algorithm for classification, and Spartina alterniflora pixels are identified from the normalized difference Spartina alterniflora phenological index image after high-frequency noise removal to obtain the spatiotemporal distribution map of Spartina alterniflora.

[0079] Specifically, in the embodiment of the present application, the unsupervised classification method, K-means++ algorithm, is used to classify vegetation on the normalized difference Spartina alterniflora phenological index images from 2019 to 2023 after removing high-frequency noise. The core expression of K-means++ initialization involves calculating the probability of each data point being selected as a new cluster center. Set the cluster center, the data point is x i , then the probability of selecting the next cluster center is as follows:

[0080]

[0081] Among them, D(x i ) represents the data point x i The distance to the nearest selected cluster center. K-means++'s intelligent initialization method makes it easier for the algorithm to find appropriate cluster centers in vegetation classification tasks. Due to the calculation of NDSPI, the numerical characteristics of Spartina alterniflora differ significantly from those of other salt marsh vegetation, improving the accuracy and stability of classification in our study.

[0082] It should be noted that in the implementation, we set the number of clusters to 4, representing the four categories of Spartina alterniflora, Suaeda salsa, Phragmites australis, and mixed vegetation. After 10 initialization iterations and 300 maximum iterations, the final coastal salt marsh vegetation classification results were obtained. Figure 5 The figure shows a schematic diagram of the spatiotemporal distribution of salt marsh vegetation classified using the K-means++ algorithm based on the normalized difference Spartina alterniflora phenological index image with high-frequency noise removed in an embodiment of the present application.

[0083] Optionally, in an embodiment of the present application, after realizing the automatic classification of coastal salt marsh vegetation using the K-means++ algorithm, the method includes: performing verification based on the classification result of the coastal salt marsh to determine the accuracy of the classification algorithm.

[0084] Specifically, in the embodiment of the present application, the classification results are verified using sample points obtained through field sampling. The classification accuracy is evaluated by calculating the accuracy, recall, precision, and F1 Score, as shown in the following formula:

[0085]

[0086] TP stands for true positives, which refers to the number of samples that are actually positive and correctly predicted as positive by the model; TN stands for true negatives, which refers to the number of samples that are actually negative and correctly predicted as negative by the model; FP stands for false positives, which refers to the number of samples that are actually negative but incorrectly predicted as positive by the model; and FN stands for false negatives, which refers to the number of samples that are actually positive but incorrectly predicted as negative by the model. Accuracy represents the ratio of correctly classified samples to the total number of samples and represents the overall accuracy of the classifier. Recall represents the ratio of true positives to all actual positive samples and focuses on the classifier's coverage of positive examples. Precision represents the ratio of true positives to samples classified as positive and focuses on the accuracy of the classifier. The F1 score comprehensively considers both precision and recall, and its value ranges from 0 to 1, with values ​​closer to 1 indicating better model performance.

[0087] It should be noted that in the embodiment of this application, only the classification results of 2022 are evaluated for accuracy. Figure 6 This figure illustrates the accuracy evaluation of the spatiotemporal distribution of salt marsh vegetation in this example. The accuracy evaluation results show that S. alterniflora performed exceptionally well in identification, achieving a precision of 0.82, a recall of 0.88, and an F1 score of 0.85—the highest among all categories. This excellent balance of high precision and recall demonstrates the robustness of NPSPI-based S. alterniflora extraction.

[0088] Example 2: This example selects the salt marsh wetland at the mouth of the Linhong River in Haizhou Bay, Lianyungang, Jiangsu Province as the study area. The typical salt marsh vegetation in this area is mainly Spartina alterniflora, Phragmites australis, Suaeda salsa and their mixed vegetation, which are significantly affected by tides. In order to verify the feasibility and superiority of the method of the present invention, a total of 365 scenes of Sentinel-2L2A surface reflectance images with open access from Copernicus were used from 2019 to 2023. Images with cloud cover of less than 10% were screened for analysis. The spatial resolution was 10m, the temporal resolution was about 5 days, and the corresponding annual maximum number of observations was N=73.

[0089] To further evaluate the classification results, we conducted comparative experiments in two aspects: on the one hand, by changing the calculation parameters of NDSPI, replacing NDPI with NDVI, and continuing to use the K-means++ algorithm for classification; on the other hand, only changing the classifier, and using the supervised classification random forest algorithm for classification.

[0090] Figure 7 This is the comparison result of the study area in 2022. The results show that: Figure 7 The classification capabilities of (a) and (b) are obviously weaker than Figure 5 , they were unable to extract complete S. alterniflora patches, and misclassified S. alterniflora from S. salsa and P. australis to a serious extent. The classification ability for mixed weeds was almost lost. Figure 7 (b) The patch fragmentation is relatively high, and artificial facilities and vegetation in landscape areas are missed.

[0091] Table 1 Classification results (NDSPI structural parameters remain unchanged)

[0092]

[0093] Table 1 is Figure 7 The accuracy evaluation results of the two classification methods only show the overall accuracy of the classification methods and the relevant accuracy index values ​​of S. alterniflora. Figure 6 As a result, the classification accuracy of S. alterniflora was significantly reduced, and the overall accuracy of the classifier also dropped significantly. Figure 7 (a) is slightly more accurate than (b), but the classification accuracy for S. alterniflora is only 0.55, indicating that this method is insufficient for S. alterniflora recognition. NDVI+K-means++ has an overall accuracy of only 0.76, and an accuracy of 0.55 for Spartina alterniflora. While NDSPI+Random Forest slightly improves the accuracy to 0.79, the accuracy for Spartina alterniflora is only 0.50, with an F1 score of 0.58. In comparison, the NDSPI+K-means++ method of the present invention has significant advantages in recognition ability, model stability, and reliance on small sample sizes.

[0094] In summary, this example verifies that the automated monitoring method based on the time series normalized phenological index and the normalized difference phenological index combined with harmonic reconstruction and K-means++ clustering has high precision, low cost, low sample dependence and good generalization ability, and can provide effective technical support for real-time invasion monitoring and early warning of Spartina alterniflora in coastal salt marsh wetlands.

[0095] The above embodiments merely illustrate the implementation methods of the present invention. Although the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention. It should be noted that a person skilled in the art may make various modifications and improvements without departing from the scope of the present invention, and such modifications and improvements are all within the scope of protection of the present invention.

Claims

1. A method for monitoring Spartina alterniflora based on NDSPI, characterized in that: The following steps are involved: S1: Obtain time series remote sensing data of coastal salt marsh wetland areas: Obtain multi-temporal Sentinel-2L2A satellite surface reflectance image data of the area to be monitored, and calculate the time series normalized phenological index image. The calculation formula is: Among them, NDPI represents the time series normalized phenological index, NIR, Red, and SWIR are the reflectivity values ​​of Sentinel-2L2A satellite, respectively; S2: Harmonic reconstruction of the time series normalized phenological index image and extraction of the phenological curve: The time series normalized phenological index image calculated in step S1 is reconstructed using the time series harmonic analysis algorithm to obtain the reconstructed time series image. Specifically, the harmonic fitting model is used for calculation, and the calculation formula is: Among them, y(t j ) is the original NDPI sequence, is the reconstructed NDPI sequence, ε(t j ) is the residual sequence; t j is the observation time of the original sequence, where j = 1, 2, ..., N, N is the maximum number of observations of the time series data column; the value of nf represents the frequency f i The number of relevant harmonic components; a i and b j is a frequency f i The trigonometric coefficients of the components, a0 is the coefficient of the zero-frequency term; Based on the reconstructed time series images and combined with pre-selected vegetation sample points, the phenological curves of various vegetation types are extracted; S3: Construction of the Normalized Difference Spartina alterniflora Phenological Index (NDSPI) image: Based on the various vegetation phenological curves extracted in step S2, the Normalized Difference Spartina alterniflora Phenological Index (NDSPI) image is calculated and obtained. The calculation formula is: Among them, NDPI Sep.+Oct. NDPI is the average value of the normalized difference phenology index from September to October each year. Jul.+Aug. is the average value from July to August each year; then, the normalized difference Spartina alterniflora phenological index image was processed with a 3-pixel window-based median filter to obtain the denoised normalized difference Spartina alterniflora phenological index image; S4: Automatic vegetation classification: The denoised normalized difference Spartina alterniflora phenological index image obtained in step S3 is automatically classified using the K-means++ algorithm to obtain a spatiotemporal distribution map of Spartina alterniflora.

2. a kind of Spartina alterniflora monitoring method based on NDSPI according to claim 1, is characterized in that: The satellite reflectivity in step S1 specifically includes reflectivity values ​​of bands B8, B4, and B11.

3. a kind of Spartina alterniflora monitoring method based on NDSPI according to claim 1, is characterized in that: The phenological curves of various types of vegetation described in step S2 specifically include phenological curves of four different categories: Spartina alterniflora, Suaeda salsa, Phragmites australis, and mixed vegetation.

4. a kind of Spartina alterniflora monitoring method based on NDSPI according to claim 1, is characterized in that: Before calculating and obtaining the normalized difference Spartina alterniflora phenological index image in step S3, a phenological curve separability discrimination method is included. This method uses the J-M distance as a separability discrimination index for each type of vegetation phenological curve extracted in step S2, calculates the J-M distance between any two types of vegetation curves, and judges the separability of each type of vegetation curve based on the J-M distance value to ensure that the Spartina alterniflora curve is significantly distinguishable from other salt marsh vegetation curves in phenological characteristics.

5. a kind of Spartina alterniflora monitoring method based on NDSPI according to claim 4, is characterized in that: The separability JM distance is calculated as follows: JM=2(1-e -B ) Where B is the Bhattacharyya distance between two categories; e ik represents the mean of different types of vegetation samples under the vegetation phenology curve; δ ik is the variance of different types of vegetation; where i = 1, 2, ..., 6; k = 1, 2, ..., 4.

6. a kind of Spartina alterniflora monitoring method based on NDSPI according to claim 1, is characterized in that: The median filtering process based on a 3-pixel window in step S3 is performed as follows: Among them, I(x,y) represents the input image, represents the output image after filtering, W is a 3-pixel filtering window, and (i, j) represents the offset coordinates within the window relative to the center pixel (x, y).

7. A method for monitoring Spartina alterniflora based on NDSPI according to claim 1, characterized in that: In step S4, the following steps are performed in sequence on the de-noised normalized difference Spartina alterniflora phenological index image obtained in step S3 to generate a spatiotemporal distribution map of Spartina alterniflora: S4-1: Take all pixel values ​​in the denoised normalized difference Spartina alterniflora phenological index image generated in step S3 as input, and use the K-means++ algorithm to randomly select four initial cluster centers according to the square ratio of the distance between the pixel and the selected cluster center. The specific calculation formula is: Among them, D(x i ) represents the data point x i The distance to the nearest selected cluster center; S4-2: Starting from the four initial cluster centers obtained in step S4-1, a standard K-means iteration is performed on the pixel value set. Each pixel is assigned to the nearest cluster center in each iteration. The position of each cluster center is then recalculated and the process is repeated until the moving distance of all cluster centers is less than a preset threshold or the maximum number of iterations is reached. S4-3: After completing one or more initialization and iteration processes consisting of steps S4-1 and S4-2, select a clustering result that minimizes the intra-class sum of squares, and stitch all pixels identified as Spartina alterniflora in the result in sequence according to their image acquisition time to output a spatiotemporal distribution map of Spartina alterniflora.

Images