A multi-mode rubber plantation remote sensing identification and forest age estimation method

Abstract

This invention discloses a remote sensing identification and stand age estimation method for multi-mode rubber plantations. Utilizing multi-source remote sensing data and advanced image processing and analysis methods, it accurately monitors and estimates the distribution and stand age of rubber plantations with different modes, such as large-scale and small-scale farming. First, through data preparation and integration, combined with unsupervised image segmentation and classification optimization using Gaussian mixture models, a high-precision rubber plantation distribution map is generated. Second, based on time-series vegetation-moisture index (e.g., NDWI) data, phenological features are extracted to identify the planting year of rubber trees and estimate stand age, ultimately generating a rubber stand age distribution map, providing scientific support for plantation management and ecological assessment.

Description

Technical Field

[0001] This invention belongs to the field of remote sensing monitoring and image processing, specifically relating to a method for remote sensing identification and forest age estimation of multi-mode rubber plantations. Background Technology

[0002] The rubber tree (Hevea brasiliensis) originated in the Amazon region, but in recent years, its plantations have expanded primarily in historically unsuitable areas such as Southeast Asia, South Asia, and China. The rapid growth of global rubber cultivation is closely linked to increasing industrial demand for rubber products, and rubber plantations now extend to latitudes as far north as 25 degrees. Southeast Asia is the heart of the global rubber industry, with Thailand, Indonesia, and Malaysia collectively producing over 90% of the world's natural rubber. The sustainable development of the rubber industry has a significant impact on the global economy and farmers' livelihoods; therefore, systematically identifying rubber plantations and their expansion patterns is crucial.

[0003] Rubber plantations can be categorized into smallholder plantations and industrialized plantations. Industrialized plantations are typically larger in scale and have standardized planting structures, which facilitates identification and analysis using remote sensing technology. Smallholder plantations, on the other hand, present challenges for remote sensing monitoring due to their small size and dispersed locations. Remote sensing technology has become an important tool for assessing the distribution of rubber plantations and their ecological impacts. By integrating multi-source data (such as optical remote sensing, synthetic aperture radar, and lidar), the distribution, spatiotemporal variations, and age of rubber plantations can be efficiently monitored.

[0004] Remote sensing technology is widely used in rubber plantation monitoring, enabling the extraction of key ecological indicators and analysis of growth status, particularly leaf growth and shedding cycles. With the development of cloud computing technology, the accuracy of remote sensing data processing and phenological analysis has gradually improved. However, due to the heterogeneity of data sources and differences in processing methods, cross-regional comparisons and long-term dynamic monitoring still face challenges, especially in smallholder plantations. Therefore, deepening the application of remote sensing technology in multi-source data integration and dynamic monitoring is of great significance.

[0005] Therefore, this invention addresses the limitations of existing technologies in large-scale rubber plantation monitoring and forest age estimation, particularly the limited identification capabilities of existing methods under the influence of regional phenological differences and complex planting terrain. For example, phenological changes are significant in rubber plantations in northern Laos, while the climate in the south is stable with less pronounced phenological variations. Existing technologies struggle to eliminate phenological interference, leading to poor classification consistency and insufficient accuracy. Furthermore, in complex terrain conditions (such as mountains and hills), optical images are significantly affected by topographic shadows and vegetation canopy obstruction, and existing methods fail to fully integrate the advantages of SAR and optical data, limiting monitoring capabilities. This invention improves classification accuracy and adaptability by integrating the advantages of SAR and optical imagery, eliminating phenological interference, and combining Gaussian mixture models with time series analysis techniques. Summary of the Invention

[0006] To address the technical problems existing in the background art, this invention aims to provide a method for remote sensing identification and forest age estimation of multi-mode rubber plantations. Utilizing multi-source remote sensing data and combining advanced image processing and analysis methods, it accurately monitors and estimates the distribution and forest age of rubber plantations with different modes, such as large-scale and small-scale farming. First, through data preparation and integration, combined with unsupervised image segmentation and classification optimization using Gaussian mixture models, a high-precision rubber plantation distribution map is generated. Second, based on time-series vegetation-moisture index (such as LSWI) data, phenological features are extracted to identify the planting year of rubber trees and estimate forest age, ultimately generating a rubber forest age distribution map, providing scientific support for plantation management and ecological assessment.

[0007] To solve the technical problem, the technical solution of the present invention is as follows:

[0008] A method for remote sensing identification and forest age estimation of multi-mode rubber plantations, the method comprising:

[0009] S1: Using multi-source remote sensing data and ground sample point data, a remote sensing feature dataset of rubber plantation is constructed. Radiometric correction, atmospheric correction and geometric correction are performed on the remote sensing feature dataset. Using the normalized vegetation index and surface moisture index, the spectral features and phenological changes of rubber plantation are extracted to obtain preprocessed remote sensing image data.

[0010] S2: Based on the preprocessed remote sensing image data, a multi-dimensional remote sensing feature space is constructed. By synthesizing images throughout the year, the adaptability of the data to regional characteristics is enhanced, and multi-source remote sensing feature data adapted to different regional characteristics is obtained.

[0011] S3: Using Gaussian mixture model to perform unsupervised image segmentation on multi-source remote sensing feature data, the spatial distribution of rubber plantations is initially identified. The expectation-maximization algorithm is introduced to optimize the classification results iteratively, and a preliminary distribution map of rubber plantations is obtained.

[0012] S4: Based on the obtained preliminary distribution map of rubber plantations, ground sample data is used to verify the accuracy, calculate the overall accuracy and Kappa coefficient, correct classification errors according to the verification results, and re-verify to obtain a high-precision distribution map of rubber plantations;

[0013] S5: Based on the time series LSWI data of Landsat imagery, analyze the changes in phenological characteristics of rubber tree planting, identify the key inflection points of the planting year, and combine the growth pattern of rubber trees to estimate the planting year of rubber trees through remote sensing vegetation feature model, thus obtaining the forest age estimation results of rubber plantations.

[0014] S6: Combine the forest age estimation results with the spatial distribution data of rubber plantations to generate a forest age distribution map of rubber plantations.

[0015] Furthermore, in step S1, the multi-source remote sensing data includes: Sentinel-1 SAR imagery, Sentinel-2 and Landsat optical imagery.

[0016] Furthermore, step S2 includes:

[0017] S201: Sentinel-1 and Sentinel-2 data fusion combines the VV and VH polarimetric SAR data from Sentinel-1 with the optical imagery data from Sentinel-2, using radar data fusion from ascending orbit ASC and descending orbit DSC; the fusion formula is as follows:

[0018]

[0019] Where: P ASC P represents the pixel value of the ascent orbit image; DSC P represents the pixel value of the descent orbit image; fused These are the merged pixel values;

[0020] S202: Extraction of the maximum Normalized Difference Vegetation Index (NDVI). The interference of phenological changes is eliminated by using the maximum annual NDVI value. Shortwave infrared (B11) band extraction is performed using Sentinel-2 B11 band data to capture moisture variation information in rubber plantations. The NDVI calculation formula is as follows:

[0021]

[0022] To eliminate the influence of seasonal variations, the method of synthesizing the annual maximum NDVI value is adopted:

[0023] NDVI max =max(NDVI) t )

[0024] Among them: NDVI t NDVI is the NDVI value at each time point in the time series; max The maximum value of NDVI throughout the year; in order to capture the sensitivity of rubber plantations to moisture changes, shortwave infrared band B11 data was extracted;

[0025] S203: Feature combination to construct a multidimensional feature space. The VV and VH polarization features of Sentinel-1 are spliced ​​with the B11 band and NDVI maximum value features of Sentinel-2 to construct a multidimensional feature space.

[0026] Furthermore, step S3 includes: using a Gaussian Mixture Model (GMM) to assume that the data consists of different Gaussian distributions, and using the EM algorithm to estimate the parameters;

[0027] The mathematical expression for the Gaussian Mixture Model (GMM) is:

[0028]

[0029] Where p(x) is the probability density of sample point x, π k It is the weight of the k-th Gaussian distribution, μ k and ∑ k Let $\mathbf{k}$ be the mean and covariance of the $k$-th Gaussian distribution, where $K$ is the number of Gaussian distributions. Select an appropriate model to distinguish rubber plantations from other land cover types and calculate the estimated values ​​of the initial mean, covariance, and weights.

[0030] For each pixel in the remote sensing image, the probability of it belonging to each category is calculated according to the Gaussian Mixture Model (GMM). For the case simplified to two Gaussian distributions, the GMM formula simplifies to:

[0031] p(x)=π1N(x|μ1,∑1)+π2N(x|μ2,∑2)

[0032] Where μ1, ∑1, and π1 represent the mean, covariance, and weight of the rubber plantation category, respectively, while μ2, ∑2, and π2 represent the corresponding parameters of farmland or oil palm plantation.

[0033] The Expectation-Maximization (EM) algorithm is used to calculate the probability that each sample point belongs to a different Gaussian distribution. That is, the posterior probability is calculated based on the current model parameters. Based on the calculated probability, the mean, covariance, and weights of the model are updated. During this process, the model parameters are gradually adjusted to maximize the likelihood of the data. The steps are repeated until the parameters converge, that is, the change is very small or the preset number of iterations is reached. Based on the calculated probability that each pixel belongs to a different category, it is classified into the category with the highest probability.

[0034] Compared with the prior art, the advantages of the present invention are as follows:

[0035] High-precision identification: By integrating multi-source remote sensing data and advanced image processing technologies (such as Gaussian mixture models), high-precision identification of rubber plantations can be achieved, especially in complex terrain and diverse planting patterns.

[0036] Automation and Efficiency: By employing unsupervised learning methods (GMM) and the expectation-maximization algorithm, manual intervention is reduced, the automation level of classification and forest age estimation is improved, and the analysis efficiency is significantly enhanced.

[0037] Highly adaptable: The method can adapt to the characteristics of different regions and effectively handle phenological changes between regions, making the monitoring results equally reliable under different geographical and climatic conditions.

[0038] Multidimensional feature space construction: By synthesizing images throughout the year and fusing features, a multidimensional remote sensing feature space is constructed to provide rich information for subsequent analysis, thereby accurately capturing the growth dynamics of rubber plantations.

[0039] Accurate forest age estimation: LSWI analysis of rubber tree growth process can accurately identify planting year and growth stage, providing a scientific basis for forest age estimation.

[0040] Sustainable agricultural management support: This approach provides important support for the sustainable management of rubber plantations, including farmland planning, management decisions, and ecological environment assessment, promoting the coordinated development of agriculture and ecology.

[0041] Data supports decision-making: The generated high-precision rubber plantation distribution maps and forest age distribution maps provide reliable information for decision-makers and farmers, supporting scientific agricultural management and policy-making.

[0042] In summary, by integrating modern remote sensing technology with machine learning methods, this invention not only improves the accuracy and efficiency of rubber plantation monitoring, but also provides effective data support and scientific basis for promoting sustainable agriculture and ecological management. Attached Figure Description

[0043] Figure 1 The main flowchart of the multi-mode rubber plantation remote sensing identification and forest age estimation method of the present invention;

[0044] Figure 2 The image shows the identification results of a multi-mode rubber plantation remote sensing identification and forest age estimation method according to the present invention. Detailed Implementation

[0045] The specific implementation of the present invention is described below with reference to embodiments:

[0046] It should be noted that the structures, proportions, sizes, etc. shown in this specification are only used to complement the content disclosed in the specification for those skilled in the art to understand and read, and are not intended to limit the conditions under which the present invention can be implemented. Any modifications to the structure, changes in the proportions, or adjustments to the size, without affecting the effects and objectives that the present invention can produce, should still fall within the scope of the technical content disclosed in the present invention.

[0047] Furthermore, the terms such as "upper," "lower," "left," "right," "middle," and "one" used in this specification are merely for clarity of description and are not intended to limit the scope of the invention. Any changes or adjustments to their relative relationships, without substantially altering the technical content, should also be considered within the scope of the invention.

[0048] Example 1:

[0049] This invention aims to accurately monitor and estimate the distribution and age of rubber plantations under different models, such as large-scale and small-scale farms, using multi-source remote sensing data combined with advanced image processing and analysis methods. First, through data preparation and integration, and by combining unsupervised image segmentation and classification optimization using a Gaussian mixture model, a high-precision rubber plantation distribution map is generated. Second, based on time-series Normalized Difference Vegetation Index (NDVI) and Surface Water Index (LSWI) data, phenological features are extracted to identify the planting year of rubber trees and estimate their age, ultimately generating a rubber plantation age distribution map, providing scientific support for plantation management and ecological assessment. Figure 1 As shown, the specific steps are as follows:

[0050] 1. Theoretical Basis and Data Preparation

[0051] Using multi-source remote sensing data such as Sentinel and Landsat, combined with ground sample point data, a remote sensing feature dataset for rubber plantations was constructed. Preprocessing of the remote sensing data, including radiometric correction, atmospheric correction, and geometric correction, ensured data quality and consistency. Based on time-series imagery, the Normalized Difference Vegetation Index (NDVI) and Surface Moisture Index (LSWI) were extracted to dynamically analyze the spectral characteristics and phenological changes of rubber plantations, providing solid data support for subsequent classification and stand age estimation.

[0052] 2. Integration of multi-source remote sensing data

[0053] To address the significant differences in phenological variations between the north and south (with pronounced phenological characteristics in the north and a more gradual change in the south), phenological interference is effectively eliminated through the synthesis of year-round imagery, enhancing the technology's adaptability to different regions. A multi-dimensional remote sensing feature space is constructed by integrating Sentinel-1 SAR imagery (VV and VH polarization features) with Sentinel-2 and Landsat optical imagery (red band B4 and shortwave infrared band B11). SAR imagery, due to its excellent penetration capabilities, is suitable for complex terrain, while optical imagery provides high spectral resolution vegetation features, enabling precise monitoring of the spatial distribution of plantations in both plains and mountainous areas.

[0054] 3. Gaussian Mixture Model for Unsupervised Image Segmentation

[0055] Unsupervised image segmentation was performed using a Gaussian Mixture Model (GMM) to construct a feature space from multi-source remote sensing data. Statistical modeling methods were used to initially classify rubber plantations for a single year, identifying their potential spatial distribution patterns. To further improve classification accuracy, the Expectation-Maximization (EM) algorithm was introduced to iteratively optimize the initial classification results, thereby obtaining more accurate classification probability results. Finally, a distribution map of rubber plantations for a single year was output, providing a reference for subsequent forest age identification and accuracy verification.

[0056] 4. Precision verification of rubber plantations

[0057] To ensure the reliability of the rubber plantation classification results, ground sample data were used to assess the accuracy of the classification results, calculating the overall accuracy (OA) and Kappa coefficient. The classification results were then corrected and re-validated. The resulting high-precision rubber plantation distribution map provides accurate basic data support for subsequent research.

[0058] 5. Forest Age Identification and Estimation

[0059] Based on time-series LSWI data from Landsat imagery, this method identifies key inflection points in rubber tree planting by analyzing monthly interannual phenological characteristics. By combining planting patterns with remote sensing vegetation characteristic models, the planting year of the rubber trees is estimated, and the stand age is determined accordingly. This method fully utilizes abrupt changes in rubber tree planting, enabling high-precision identification of stand age and providing crucial reference data for production management and ecological assessment of rubber plantations.

[0060] 6. Distribution map of rubber plantation age

[0061] By integrating the results of forest age identification with the spatial distribution information of rubber plantations, a distribution map of rubber forest age is generated.

[0062] Example 2:

[0063] This embodiment applies to Embodiment 1. Specifically, the first step is to perform multi-source data fusion to prepare the dataset; then, the Gaussian mixture model is called to achieve unsupervised automatic classification of rubber plantations; based on the plantation map of a single year (e.g., 2024) that has already been generated, Landsat data is used to identify LSWI inflection points, thereby determining the planting year.

[0064] 1. Data fusion and feature extraction (based on Sentinel-1 and Sentinel-2 feature combination)

[0065] Implementation steps:

[0066] ①Sentinel-1 and Sentinel-2 Data Fusion: This method combines VV and VH polarimetric SAR data from Sentinel-1 with optical imagery data from Sentinel-2, employing ascent orbit (ASC) and descent orbit (DSC) radar data fusion to effectively mitigate the impact of mountain shadows and terrain complexity on the imagery.

[0067] ② Extraction of the maximum NDVI value: By using the maximum NDVI value throughout the year to eliminate the interference of phenological changes, the spatial distribution of rubber plantations becomes more stable, which can effectively reduce phenological interference and enhance the representativeness and stability of vegetation identification.

[0068] ③ Extraction of shortwave infrared band (B11): Sentinel-2 B11 band data was used to capture information on moisture changes in rubber plantations.

[0069] This invention improves the spatial distribution identification accuracy of rubber plantations by combining Sentinel-1 synthetic aperture radar (SAR) data and Sentinel-2 optical imagery data using a feature fusion method, demonstrating greater stability and applicability, particularly in complex terrain environments. Sentinel-1 SAR data includes VV and VH polarization features, providing information on surface structure and vegetation scattering. This invention employs the fusion of ascending orbit (ASC) and descending orbit (DSC) radar data, effectively mitigating the impact of mountain shadows and terrain complexity on the imagery. The fusion formula is as follows:

[0070]

[0071] Where: P ASC P represents the pixel value of the ascent orbit image; DSC P represents the pixel value of the descent orbit image; fused These are the pixel values ​​after fusion. Track fusion processing eliminates noise caused by terrain occlusion in single-track images, while improving the stability of rubber plantation identification in complex terrain environments. In Sentinel-2 optical image processing, this invention removes interference from phenological changes by extracting the maximum NDVI value throughout the year. The NDVI calculation formula is as follows:

[0072]

[0073] To eliminate the influence of seasonal changes (such as greening and withering), the method of synthesizing the annual NDVI maximum value is adopted:

[0074] NDVI max =max(NDVI) t )

[0075] Among them: NDBI tNDVI is the NDVI value at each time point in the time series; max The maximum NDVI value is the annual NDVI value. Extracting the maximum NDVI value throughout the year effectively reduces phenological interference and enhances the representativeness and stability of vegetation identification. Simultaneously, to capture the sensitivity of rubber plantations to moisture changes, this invention extracts shortwave infrared (B11) data. The B11 band reflects the canopy moisture content and vegetation growth status of rubber trees, providing crucial support for accurate identification of rubber plantations. Regarding feature combination, this invention splices the VV and VH polarization features of Sentinel-1 with the B11 band and the maximum NDVI value features of Sentinel-2 to construct a multi-dimensional feature space.

[0076] 2. Automatic Classification of Rubber Plantations Based on Gaussian Mixture Model (GMM)

[0077] Implementation steps:

[0078] ①GMM modeling: GMM is used to assume that the data consists of different Gaussian distributions (e.g., Gaussian distributions of rubber plantations and tropical rainforests), and the EM algorithm is used to estimate the parameters.

[0079] ② Classification process: This section explains how to calculate the probability of each pixel belonging to a certain category based on the GMM, thereby achieving automatic classification of rubber plantations in remote sensing images.

[0080] A Gaussian Mixture Model (GMM) is a probability-based statistical model that assumes a dataset consists of multiple Gaussian distributions (normal distributions) with different parameters. Each Gaussian distribution represents a latent class or cluster in the data. GMM describes the distribution of the entire dataset by weighting and combining the data. In remote sensing image analysis, GMM is often used for automatic land cover classification, such as distinguishing rubber plantations from other vegetation or land cover types. The mathematical expression for GMM is:

[0081]

[0082] Where p(x) is the probability density of sample point x, π k It is the weight of the k-th Gaussian distribution, μ k and ∑ k Here, K and K represent the mean and covariance of the k-th Gaussian distribution, respectively, where K is the number of Gaussian distributions. For complex land cover types, Gaussian Mixture Models (GMMs) can automatically separate different types of land cover based on the spectral characteristics or other attributes of pixels. In practical applications, when distinguishing between rubber plantations and farmland or oil palm plantations, the GMM can be simplified to two Gaussian distributions (i.e., K = 2). In this case, the formula for the GMM simplifies to:

[0083] p(x)=π1N(x|μ1,∑1)+π2N(x|μ2,∑2)

[0084] Here, μ1, ∑1, and π1 represent the mean, covariance, and weight of the rubber plantation category, respectively, while μ2, ∑2, and π2 represent the corresponding parameters for farmland or oil palm plantations. By leveraging prior knowledge (such as the fact that rubber plantations typically have different spectral reflectance characteristics), rubber plantations can be assigned to clusters with different means, thus achieving efficient classification. To estimate the parameters of these Gaussian distributions, this study uses the Expectation-Maximization (EM) algorithm. The EM algorithm employs two iterative steps: the first step calculates the probability of each sample belonging to each Gaussian distribution (expectation step); the second step updates the model parameters based on these probabilities (maximization step). Through multiple iterations, the EM algorithm can estimate the optimal model parameters, thereby achieving automatic classification of remote sensing images. After parameter estimation, the Gaussian Model (GMM) can calculate the probability of each pixel belonging to a specific category, thus achieving efficient land cover classification.

[0085] 3. Use Landsat data to determine the planting year.

[0086] Implementation steps:

[0087] ①Landsat data selection: Select Landsat-5, Landsat-7 and Landsat-8 Level-2 Collection-2 Tier-1 products to ensure data quality and consistency.

[0088] ②LSWI Change Analysis: Using LSWI (Surface Moisture Index) to analyze the inflection point of land use change, and then to infer the growth process and planting year of rubber plantations.

[0089] This study used Landsat time-series data (Landsat-5, Landsat-7, and Landsat-8) to estimate the planting year of rubber plantations. Since 1984, the Landsat satellite series has provided optical imagery with a spatial resolution of 30 meters, with a revisit time of 16 days for each satellite. To ensure data quality, we used the Level-2 Collection-2 Tier-1 products of Landsat-5, Landsat-7, and Landsat-8, which are atmospherically corrected and provide surface reflectance data suitable for long-term remote sensing analysis. Landsat data accurately captures different growth stages of rubber plantations, from initial land preparation to the seedling stage after planting, and then to mature rubber trees. Therefore, this study used the Surface Moisture Index (LSWI) to monitor moisture changes and growth processes in rubber plantations. The formula for calculating LSWI is:

[0090]

[0091] NIR represents the near-infrared band (B5 in Landsat-8), and SWIR represents the short-wave infrared band (B7 in Landsat-8). LSWI is sensitive to moisture changes and can reflect the growth process of rubber plantations. Compared to vegetation indices that rely on the visible spectrum, LSWI has less noise and is less sensitive to atmospheric conditions (such as clouds and water vapor), making it particularly suitable for tropical regions. Taking rubber plantations as an example, during the land preparation stage, LSWI values ​​are low, indicating soil exposure. As the rubber seedlings grow after planting, LSWI values ​​steadily increase, reflecting increased water absorption by the vegetation. As the rubber trees reach maturity and form a complete canopy, LSWI values ​​plateau. By analyzing the changes in LSWI over Landsat time series, the planting year of the rubber plantation can be clearly inferred, such as... Figure 2 As shown.

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the above embodiments. Within the scope of knowledge possessed by those skilled in the art, various changes can be made without departing from the spirit of the present invention.

[0097] Many other changes and modifications can be made without departing from the concept and scope of this invention. It should be understood that this invention is not limited to the specific embodiments, and the scope of this invention is defined by the appended claims.

Claims

1. A method for remote sensing identification and forest age estimation of multi-mode rubber plantations, characterized in that, The method includes: S1: Using multi-source remote sensing data and ground sample point data, a remote sensing feature dataset of rubber plantation is constructed. Radiometric correction, atmospheric correction and geometric correction are performed on the remote sensing feature dataset. The normalized vegetation index NDVI and surface water index LSWI are used to extract the spectral features and phenological changes of rubber plantation to obtain preprocessed remote sensing image data. Wherein, NIR represents the near-infrared band, and SWIR represents the short-wave infrared band; S2: Based on the preprocessed remote sensing image data, a multi-dimensional remote sensing feature space is constructed. By synthesizing images throughout the year, the adaptability of the data to regional characteristics is enhanced, and multi-source remote sensing feature data adapted to different regional characteristics is obtained. S201: Sentinel-1 and Sentinel-2 data fusion combines the VV and VH polarimetric SAR data from Sentinel-1 with the optical imagery data from Sentinel-2, using radar data fusion from ascending orbit ASC and descending orbit DSC; the fusion formula is as follows: Where: P ASC P represents the pixel value of the ascent orbit image; DSC P represents the pixel value of the descent orbit image; fused These are the merged pixel values; S202: Extraction of the maximum NDVI value. The interference from phenological changes is eliminated by using the annual maximum NDVI value. Extraction of the shortwave infrared band B11. Sentinel-2 B11 band data is used to capture moisture variation information in rubber plantations. The interference from phenological changes is eliminated by extracting the annual maximum NDVI value. The NDVI calculation formula is as follows: To eliminate the influence of seasonal variations, the method of synthesizing the annual maximum NDVI value is adopted: NDVI max =max(NDVI t ) Among them: NDVI t NDVI is the normalized difference vegetation index (NDVI) value at each time point in the time series; max The maximum value of the Normalized Difference Vegetation Index (NDVI) for the whole year was used; shortwave infrared (B11) data was extracted to capture the sensitivity of rubber plantations to water changes. S203: Feature combination to construct a multi-dimensional feature space. The VV and VH polarization features of Sentinel-1 are spliced ​​with the B11 band and NDVI maximum value features of Sentinel-2 to construct a multi-dimensional feature space. S3: Using Gaussian mixture model to perform unsupervised image segmentation on multi-source remote sensing feature data, the spatial distribution of rubber plantations is initially identified. The expectation-maximization algorithm is introduced to optimize the classification results iteratively, and a preliminary distribution map of rubber plantations is obtained. S4: Based on the obtained preliminary distribution map of rubber plantations, ground sample data is used to verify the accuracy, calculate the overall accuracy and Kappa coefficient, correct classification errors according to the verification results, and re-verify to obtain a high-precision distribution map of rubber plantations; S5: Based on the time series LSWI data of Landsat imagery, analyze the phenological characteristics during the rubber tree planting process and identify the key inflection points of the planting year; combined with the growth pattern of rubber trees, estimate the planting year of rubber trees through remote sensing vegetation feature models to obtain the forest age estimation results of rubber plantations. S6: Combine the forest age estimation results with the spatial distribution data of rubber plantations to generate a forest age distribution map of rubber plantations.

2. The method for remote sensing identification and forest age estimation of multi-mode rubber plantations according to claim 1, characterized in that, Step S3 includes: using a Gaussian Mixture Model (GMM) to assume that the data is composed of different Gaussian distributions, and using the EM algorithm to estimate the parameters; The mathematical expression for the Gaussian Mixture Model (GMM) is: Where p(x) is the probability density of sample point x, π k It is the weight of the k-th Gaussian distribution, μ k and ∑ k Let $\mathbf{k}$ be the mean and covariance of the $k$-th Gaussian distribution, where $K$ is the number of Gaussian distributions. Select an appropriate model to distinguish rubber plantations from other land cover types, and calculate the estimated values ​​of the initial mean, covariance, and weights. For each pixel in the remote sensing image, the probability of it belonging to each category is calculated according to the Gaussian Mixture Model (GMM). For the case simplified to two Gaussian distributions, the GMM formula simplifies to: p(x)=π1N(x|μ1,∑1)+π2N(x|μ2,∑2) Where μ1, ∑1, and π1 represent the mean, covariance, and weight of the rubber plantation category, respectively, while μ2, ∑2, and π2 represent the corresponding parameters of farmland or oil palm plantation. The Expectation-Maximization (EM) algorithm is used to calculate the probability that each sample point belongs to a different Gaussian distribution. That is, the posterior probability is calculated based on the current model parameters. Based on the calculated probability, the mean, covariance, and weights of the model are updated. During this process, the model parameters are gradually adjusted to maximize the likelihood of the data. The steps are repeated until the parameters converge, that is, the change is very small or the preset number of iterations is reached. Based on the calculated probability that each pixel belongs to a different category, it is classified into the category with the highest probability.

Images