A method and device for detecting changes in wetland vegetation

Abstract

The application provides a kind of wetland vegetation change detection method and device, the method comprises: obtaining the Landsat image and high time resolution MODIS image in the set time period of the area to be monitored, and pretreatment is carried out;Reconstruct long time series, high density Landsat surface reflectivity data set using space-time fusion algorithm;By analyzing the measured spectral characteristics of wetland vegetation, an optimized wetland vegetation change detection index based on spectral angle mapping principle is constructed;Through the LandTrendr time series segmentation algorithm of fusing spatial context features, the accurate and automatic detection of large-scale, long time series wetland vegetation change is realized.The method uses cloud computing capability to process massive remote sensing data, effectively solves the problem of data missing and insufficient algorithm stability in large-scale, long time series wetland vegetation monitoring, and provides efficient technical support for wetland resource management, carbon sink function evaluation and "carbon neutralization" strategy.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing monitoring and ecological environment assessment technology, specifically relating to a method and device for detecting changes in wetland vegetation. Background Technology

[0002] Wetland ecosystems possess enormous carbon sequestration potential and are one of the Earth's important "carbon reservoirs," playing an irreplaceable role in regulating global carbon balance, mitigating greenhouse gas concentrations, and maintaining global climate stability. As an important component of wetland ecosystems, wetland vegetation plays a crucial role in flood control and drought prevention, siltation and land reclamation, reducing soil erosion, degrading environmental pollution, regulating ecosystem carbon cycles, and maintaining ecosystem carbon balance.

[0003] Continuous human activities and climate change have led to the large-scale disappearance and degradation of wetlands in China. Despite national policies to protect wetland resources, wetlands still face serious threats. In the face of ongoing global change trends, developing an accurate and efficient method for detecting wetland vegetation changes, and systematically and quantitatively evaluating the spatiotemporal dynamics and patterns of wetland vegetation changes, is crucial for deepening our understanding of wetland ecosystem functions. This will play a vital role in wetland resource management, stabilizing and enhancing the carbon sequestration function of wetlands, and even in achieving China's "carbon neutrality" strategy.

[0004] Remote sensing technology, with its advantages of wide observation range, short update cycle, and large information volume, has become the main method for detecting regional wetland vegetation changes. With the development of Earth observation technology, remote sensing platforms and sensors are constantly increasing and improving, resulting in an explosive growth trend of massive, multi-source remote sensing data. Traditional desktop remote sensing processing platforms can no longer meet the current application needs of big data remote sensing. Furthermore, the "dual-carbon" strategy has elevated the importance of wetland ecosystem research to a new level. Therefore, conducting long-term wetland vegetation change detection based on remote sensing cloud computing platforms and exploring its change patterns has significant theoretical and practical value.

[0005] However, existing technologies still have many shortcomings: remote sensing data sources are easily affected by factors such as clouds and rain, leading to data loss; the selection of remote sensing indicators and the stability of change detection algorithms are insufficient, making it difficult to achieve long-term, high-precision, and spatially continuous dynamic monitoring of wetland vegetation, which restricts the accuracy and reliability of wetland vegetation change detection. Summary of the Invention

[0006] To address the aforementioned problems, this invention proposes a method and apparatus for detecting wetland vegetation changes, thereby resolving issues such as the lack of remote sensing data sources and the instability of remote sensing indicators and detection algorithms in existing technologies. The method includes: Landsat and high-temporal-resolution MODIS images of the area to be monitored within a specified time period were acquired and preprocessed. An improved spatiotemporal fusion algorithm was then used to fuse the preprocessed Landsat and MODIS images, reconstructing a long-term, high-temporal-density surface reflectance dataset. This improved spatiotemporal fusion algorithm, based on a weighted function model framework, incorporates nonlocal filtering to optimize the calculation of similar pixel weights and predicts the reflectance of missing Landsat images pixel-by-pixel by pairing high- and low-resolution images. Spectral measurement data of wetland vegetation in the area to be monitored were acquired, and the spectral reflectance characteristics of different species and growth stages of wetland vegetation were statistically analyzed. A representative standard spectral curve is extracted as a reference spectral vector. The angle between the pixel spectrum in the surface reflectance dataset and the reference spectral vector is calculated using a spectral angle mapping algorithm to obtain an optimized spectral index for enhancing wetland vegetation change signals. Based on the surface reflectance dataset and the optimized spectral index, dense time-series image data is constructed. The improved LandTrendr algorithm is used to perform temporal segmentation and trajectory fitting on the dense time-series image data to realize wetland vegetation change patches. The improved LandTrendr algorithm is a LandTrendr model that integrates spatial features, embedding spatial structure features, spatiotemporal context information, and spatiotemporal Moran index into the LandTrendr algorithm segmentation process.

[0007] Furthermore, the process of fusing the preprocessed Landsat and MODIS images using an improved spatiotemporal fusion algorithm specifically includes: First, the preprocessed Landsat and MODIS images are paired, with Landsat images of similar dates being paired with MODIS images to obtain unpaired MODIS image sets, paired MODIS image sets, and paired Landsat image sets. Then, the three image sets are unified to the same spatial resolution, and similar pixel search and weight calculation are performed. The idea of ​​nonlocal filtering is introduced to optimize the weight allocation of similar pixels, and the reflectance of Landsat images in the missing phases is predicted pixel by pixel. Finally, the prediction results are integrated to generate a high-precision long-term sequence image.

[0008] Furthermore, the step of using the improved LandTrendr algorithm to perform temporal segmentation and trajectory fitting on the dense time-series image data to realize wetland vegetation change patches specifically includes: First, an improved LandTrendr algorithm is used to segment the temporal imagery. Spatial structural features are introduced during the segmentation process. Through a multi-scale spatial sliding window, the spatial contrast and landscape heterogeneity index of the central pixel's neighborhood are calculated, integrating local spatial pattern information into the change detection process. Second, potential abrupt change points in the sequence are identified. Then, the authenticity of pixel changes is determined by calculating the spatiotemporal Moran index of the temporal residuals of pixels within the sliding window, and pixel-level spectral change trajectories are fitted. Finally, by combining the temporal segmentation results, spectral change trajectories, and spatial context information, change patches of wetland vegetation are extracted and identified.

[0009] Furthermore, the improved spatiotemporal fusion algorithm is based on a weighted function model framework and introduces the idea of ​​nonlocal filtering to optimize the calculation of similar pixel weights. The weight calculation formula is as follows: , In the formula, G is the Gaussian kernel, and R represents the kernel size in pixels ( ). x i ，y i The radius of the spatial neighborhood window centered on ) x i ，y i ) is the first i The positions of similar pixels, P(x) i y i R, t0) is in the pixel ( x i ，y i The region of the location is C, where C is the reflectance of the MODIS image at times t0 and t1, h is the filtering parameter, t0 represents the reference time phase, and t1 is the predicted date.

[0010] Furthermore, the specific formula for calculating the optimized spectral index is as follows: , in, R 0 As a reference vector, it consists of two parts: the first part is the spectral value of the measured spectral curve in a certain band. r 1 The second part is a high-dimensional zero vector. R i The target spectrum is the value of the measured spectral curve within a certain wavelength range. R 0 In r 1 and R i In r 1 equal, ri This represents the spectral value of the measured pixel in the i-th spectral band.

[0011] Furthermore, the formula for calculating the spatiotemporal Moran index is as follows: , In the formula, n is the number of neighborhood spatiotemporal elements involved in the calculation. For spatiotemporal elements, it represents the observation value of the central pixel i at time t. A weighted index for the interconnections between different spatiotemporal elements. Let t' represent the mean value of all spatial location pixels at time t', and j represent the number of the neighboring pixel. This represents the average value of all pixels within the sliding window at time t; This represents the observed value of the central pixel i at time t'.

[0012] Furthermore, the preprocessing includes radiometric calibration, atmospheric correction, reprojection, resampling, and cloud and cloud shadow masking.

[0013] Furthermore, accuracy verification was conducted using manual sampling points, GPS field surveys, and historical high-resolution remote sensing images, and was evaluated using overall accuracy (OA), Kappa, user accuracy (UA), and producer accuracy (PA).

[0014] The present invention also provides a wetland vegetation change detection device, comprising at least a microprocessor and a memory, characterized in that the microprocessor is programmed or configured to execute the steps of the above-described wetland vegetation change detection method, or the memory stores a computer program programmed or configured to execute the above-described wetland vegetation change detection method.

[0015] The present invention also provides a computer-readable storage medium storing a computer program programmed or configured to perform the above-described wetland vegetation change detection method.

[0016] This invention provides a wetland vegetation change detection method and apparatus that acquires a long-term, dense Landsat dataset through a spatiotemporal fusion algorithm; further, it constructs new remote sensing indicators for wetland vegetation change detection through measured spectral analysis of wetland vegetation; simultaneously, by improving the LandTrendr segmentation algorithm, it achieves rapid and accurate detection of large-scale, long-term wetland vegetation changes. By constructing a change detection method based on a remote sensing cloud platform, large-scale, long-term wetland vegetation change detection can be implemented, providing technical support for wetland resource surveys and management, and reducing the workload and cost of field surveys while ensuring accuracy requirements. Attached Figure Description

[0017] Figure 1This is a flowchart of wetland vegetation change detection based on a novel remote sensing index.

[0018] Figure 2 This describes the spatiotemporal fusion algorithm flow based on the GEE remote sensing cloud platform.

[0019] Figure 3 This refers to the hypothetical spectral curves of different ground features and the spectral ranges of their variations.

[0020] Figure 4 The flowchart of the FSF-LandTrendr change detection algorithm that integrates spatial features. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0022] like Figure 1 As shown, this embodiment relates to a wetland vegetation change detection method, using the Dongting Lake wetland as the monitoring area, specifically applied to the Google Earth Engine (GEE) cloud computing platform, and includes the following steps: The first step is to acquire Landsat images and high temporal resolution MODIS images of the area to be monitored within a set time period, and then perform preprocessing.

[0023] Based on the GEE platform, Landsat series images of the monitored area from 2000 to 2022 were acquired, specifically Landsat 5 TM, Landsat 7 ETM+, Landsat 8 / 9 OLI (30 m), and high temporal resolution MODISMOD09A1 (500 m) images.

[0024] The preprocessing includes radiometric calibration, atmospheric correction, cloud masking, image registration, and spatial resolution resampling to form a standardized data layer, ensuring consistency and comparability among multi-source data.

[0025] The second step involves using an improved spatiotemporal fusion algorithm to fuse the preprocessed Landsat and MODIS images, thereby reconstructing a long-time, high-temporal-density surface reflectance dataset.

[0026] The improved spatiotemporal fusion algorithm is based on a weighted function model framework. It introduces the idea of ​​nonlocal filtering to optimize the weight calculation of similar pixels. By pairing high- and low-resolution images and using the improved weight formula, it predicts the reflectance of missing Landsat images pixel by pixel, reconstructing a long-term, high-density surface reflectance dataset with high temporal frequency (specifically, monthly) and a spatial resolution of 30 meters. This effectively fills the data gaps caused by clouds and rain. The weight calculation formula is as follows: , In the formula, G is the Gaussian kernel, and R represents the kernel size in pixels ( x i ，y i The radius of the spatial neighborhood window centered on is used to construct the neighborhood block P(x). i y i (R, t0), x i ，y i ) is the first i The positions of similar pixels, P(x) i y i R, t0) is in the pixel ( x i ，y i The neighborhood block at position t0, C is the reflectance of the MODIS image at times t0 and t1, and h is the filtering parameter.

[0027] In this step, the spatiotemporal fusion algorithm proposed to be built based on the GEE remote sensing cloud platform improves the heterogeneity problem while addressing the issues of high computational load and time costs. It can effectively improve fusion accuracy when land cover type changes are more complex. This algorithm is implemented through the Python interface of the GEE platform, such as... Figure 2 As shown, the system is mainly divided into three modules: image set processing, group prediction of high-precision images, and result output. The image set processing module pairs the preprocessed Landsat images and MODIS images, matching Landsat images with similar dates to MODIS images to form three types of image sets: unpaired MODIS image set, paired MODIS image set, and paired Landsat image set. The group prediction of high-precision images module first unifies the three types of image sets to a spatial resolution of 30 meters, then performs similar pixel search and weight calculation, introduces nonlocal filtering to optimize the weight allocation of similar pixels, suppresses noise, and predicts the reflectance of Landsat images for missing phases pixel by pixel. The result output module integrates the above prediction results to generate long-term high-precision images.

[0028] The third step is to acquire spectral measurement data of wetland vegetation in the area to be monitored, analyze its spectral reflectance characteristics, and calculate the angle between the pixel spectrum and the reference spectral vector in the spectral measurement data based on the spectral angle mapping algorithm to obtain the optimized spectral index for enhancing the wetland vegetation change signal.

[0029] By analyzing the spectral measurement data of wetland vegetation obtained from field measurements, the spectral reflectance characteristics of different types and growth stages of wetland vegetation were statistically analyzed. Representative standard spectral curves were extracted as reference spectral vectors. The specific calculation formula for the optimized spectral index is as follows: , in, R 0 As a reference vector, it consists of two parts: the first part is the spectral value of the measured spectral curve in a certain band. r 1 The second part is a high-dimensional zero vector. R i The target spectrum is the value of the measured spectral curve within a certain wavelength range. R 0 In r 1 and R i In r 1 equal, r i This represents the spectral value of the measured pixel in the i-th spectral band.

[0030] Optimized spectral indices quantify spectral similarity by calculating the angle between the pixel spectrum and a specifically constructed reference spectral vector. This approach can keenly capture subtle spectral changes in wetland vegetation and enhance its resistance to background disturbances such as water bodies and moist soil. Figure 3 This refers to the spectral curves of different ground features and the spectral ranges of their variations.

[0031] The fourth step involves using the improved LandTrendr algorithm, which integrates spatial structure features, to detect changing patches in the dense time series data based on dense time series data and optimized spectral indices, thereby extracting wetland vegetation changes.

[0032] The improved LandTrendr algorithm incorporates spatial contrast, landscape heterogeneity, and the spatiotemporal Moran index into the LandTrendr segmentation process. By setting a moving window, it calculates the spatiotemporal Moran index of the pixel time series within the window. The formula for calculating the spatiotemporal Moran index is as follows: , In the formula, n is the number of neighborhood spatiotemporal elements involved in the calculation. For spatiotemporal elements, it represents the observation value of the central pixel i at time t. A weighted index for the interconnections between different spatiotemporal elements. Let t' represent the mean value of all spatial location pixels at time t', and j represent the number of the neighboring pixel. This represents the average value of all pixels within the sliding window at time t; This represents the observed value of the central pixel i at time t'.

[0033] like Figure 4 As shown, the improved LandTrendr algorithm incorporates spatial context information into the standard LandTrendr time series segmentation process. Specifically, it sets a moving window and calculates the spatiotemporal Moran index of the pixel time series within the window to measure the spatial autocorrelation between the central pixel and its neighboring pixels in terms of change. The STI is used as a constraint condition in the identification of change vertices and the fitting of segmentation lines, thereby suppressing isolated noise points and making the detected change patches more continuous in space, and better reflecting the spatial processes such as wetland vegetation expansion, contraction, and fragmentation.

[0034] Pixels in the FSF-LandTrendr output whose changes exceed a set threshold are extracted as changed pixels and categorized into types such as "vegetation degradation" and "vegetation restoration" based on the direction of change. Using 139 wetland vegetation change verification points collected in the Dongting Lake area from 2021 to 2022, combined with additional verification points generated from visual interpretation of historical high-resolution Google Earth imagery, a confusion matrix is ​​constructed. The overall accuracy (OA), Kappa coefficient, producer accuracy (PA), and user accuracy (UA) for each category are calculated to evaluate the change detection accuracy of this method. The overall accuracy reaches over 85%, and the Kappa coefficient is greater than 0.9.

Claims

1. A method for detecting changes in wetland vegetation, characterized in that, The method includes: Acquire Landsat and high temporal resolution MODIS images of the area to be monitored within a specified time period, and perform preprocessing. An improved spatiotemporal fusion algorithm is used to fuse preprocessed Landsat and MODIS images to reconstruct a long-time, high-temporal-density land surface reflectance dataset. The improved spatiotemporal fusion algorithm is based on a weighted function model framework, introduces the idea of ​​nonlocal filtering to optimize the calculation of similar pixel weights, and predicts the reflectance of Landsat images with missing phases pixel by pixel by pairing high- and low-resolution images. Spectral measurement data of wetland vegetation in the area to be monitored are obtained. By analyzing the spectral measurement data of wetland vegetation measured in the field, the spectral reflectance characteristics of different types and growth stages of wetland vegetation are statistically analyzed. Representative standard spectral curves are extracted as reference spectral vectors. The angle between the pixel spectrum in the surface reflectance dataset and the reference spectral vector is calculated using a spectral angle mapping algorithm to obtain an optimized spectral index for enhancing the signal of wetland vegetation change. Based on the surface reflectance dataset and optimized spectral indices, dense time-series image data is constructed. The improved LandTrendr algorithm is used to perform temporal segmentation and trajectory fitting on the dense time-series image data to realize wetland vegetation change patches. The improved LandTrendr algorithm is a LandTrendr model that integrates spatial features. It embeds spatial structure features, spatiotemporal context information and spatiotemporal Moran indices into the LandTrendr algorithm segmentation process.

2. The wetland vegetation change detection method according to claim 1, characterized in that, The process of fusing preprocessed Landsat and MODIS images using an improved spatiotemporal fusion algorithm specifically includes: First, the preprocessed Landsat and MODIS images are paired. Landsat images with similar dates are paired with MODIS images to obtain unpaired MODIS image sets, paired MODIS image sets, and paired Landsat image sets. Then, the three types of image sets are unified to the same spatial resolution, and then similar pixel search and weight calculation are performed. The idea of ​​nonlocal filtering is introduced to optimize the weight allocation of similar pixels, and the reflectance of Landsat images in the missing phase is predicted pixel by pixel. Finally, the prediction results are integrated to generate high-precision long-term sequence images.

3. The wetland vegetation change detection method according to claim 1, characterized in that, The step of using the improved LandTrendr algorithm to perform temporal segmentation and trajectory fitting on the dense time-series image data to realize wetland vegetation change patches specifically includes: First, the improved LandTrendr algorithm is used to segment the time-series images. Spatial structure features are introduced during the segmentation process. Through a multi-scale spatial sliding window, the spatial contrast and landscape heterogeneity index of the central pixel neighborhood are calculated, and local spatial pattern information is integrated into the change detection process. Secondly, potential mutation points in the sequence are identified, and then the spatiotemporal Moran index of the pixel temporal residual within the sliding window is used to determine the authenticity of pixel changes and fit the pixel-level spectral change trajectory. Finally, by combining the temporal segmentation results, spectral change trajectories, and spatial context information, we extracted and identified the change patches of wetland vegetation.

4. The wetland vegetation change detection method according to claim 1, characterized in that, The improved spatiotemporal fusion algorithm is based on a weighted function model framework and introduces the idea of ​​nonlocal filtering to optimize the calculation of similar pixel weights. The weight calculation formula is as follows: ， In the formula, G is the Gaussian kernel, and R represents the kernel size in pixels ( ). x i ，y i The radius of the spatial neighborhood window centered on ) x i ，y i ) is the first i The positions of similar pixels, P(x) i y i R, t0) is in the pixel ( x i ，y i The region of the location is C, where C is the reflectance of the MODIS image at times t0 and t1, h is the filtering parameter, t0 represents the reference time phase, and t1 is the predicted date.

5. The wetland vegetation change detection method according to claim 1, characterized in that, The specific formula for calculating the optimized spectral index is as follows: ， in, R 0 As a reference vector, it consists of two parts: the first part is the spectral value of the measured spectral curve in a certain band. r 1 The second part is a high-dimensional zero vector. R i The target spectrum is the value of the measured spectral curve within a certain wavelength range. R 0 In r 1 and R i In r 1 equal, r i This represents the spectral value of the measured pixel in the i-th spectral band.

6. The wetland vegetation change detection method according to claim 1, characterized in that, The formula for calculating the spatiotemporal Moran index is as follows: ， In the formula, n is the number of neighborhood spatiotemporal elements involved in the calculation. For spatiotemporal elements, it represents the observation value of the central pixel i at time t. A weighted index for the interconnections between different spatiotemporal elements. Let t' represent the mean value of all spatial location pixels at time t', and j represent the number of the neighboring pixel. This represents the average value of all pixels within the sliding window at time t; This represents the observed value of the central pixel i at time t'.

7. The wetland vegetation change detection method according to claim 1, characterized in that, The preprocessing includes radiometric calibration, atmospheric correction, reprojection, resampling, and cloud and cloud shadow masking.

8. The wetland vegetation change detection method according to claim 1, characterized in that, Accuracy verification was conducted using manual sampling points, GPS field surveys, and historical high-resolution remote sensing images, and was evaluated using overall accuracy (OA), Kappa, user accuracy (UA), and producer accuracy (PA).

9. A wetland vegetation change detection device, comprising at least a microprocessor and a memory, characterized in that, The microprocessor is programmed or configured to perform the steps of the wetland vegetation change detection method according to any one of claims 1 to 8, or the memory stores a computer program programmed or configured to perform the wetland vegetation change detection method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that is programmed or configured to perform the wetland vegetation change detection method according to any one of claims 1 to 5.

Images