Aquatic vegetation fine classification and dynamic monitoring method, medium and equipment

Abstract

The invention discloses an aquatic vegetation fine classification and dynamic monitoring method, a medium and equipment, and belongs to the technical field of vegetation monitoring. According to the method, multi-source remote sensing data are fused, spatio-temporal information including vegetation indexes and phenological characteristics is extracted, high-precision classification of emergent aquatic-hygrophilous vegetation, submerged vegetation and floating-leaf vegetation is realized by utilizing a machine learning or deep learning model, and dynamic change monitoring is performed by means of change detection and trend analysis; comprising the following steps: S1, acquiring and preprocessing multi-source time sequence remote sensing data; s2, multi-source time sequence feature engineering construction; s3, selecting and labeling a sample; s4, fine classification based on an intelligent decision model; s5, performing classification decision and post-processing optimization; s6, dynamically monitoring and analyzing aquatic vegetation; and S7, outputting and verifying a result. According to the method, high-precision and fine classification of the aquatic vegetation can be realized, and dynamic changes of the distribution range, density and type of the aquatic vegetation can be effectively monitored.

Description

Technical Field

[0001] The present invention belongs to the technical field of vegetation monitoring, and particularly relates to a fine classification and dynamic monitoring method, medium and device for aquatic vegetation based on multi-source time-series remote sensing information and intelligent decision-making. Background Art

[0002] Aquatic vegetation in water bodies such as lakes and wetlands (including emergent-wetland, submerged, and floating-leaved vegetation, etc.) is an important part of the aquatic ecosystem, and plays a key role in maintaining biodiversity, purifying water quality, and regulating regional climate. Quickly, accurately, and dynamically grasping the types, distributions, densities, and their spatio-temporal changes of aquatic vegetation is crucial for water resource management, ecological environment protection, eutrophication prevention, and ecological restoration effect evaluation. However, traditional classification and monitoring methods have many drawbacks, including: 1) Field investigation method: It is time-consuming, laborious, and costly, difficult to cover large areas, and unable to meet the requirements of high-frequency dynamic monitoring.

[0003] 2) Remote sensing monitoring based on single-phase and single-sensor, including multispectral remote sensing, high-resolution remote sensing, and radar remote sensing; among them: Multispectral remote sensing: It is a commonly used method at present, but it is difficult to distinguish different vegetation types with similar spectral characteristics or the phenomenon of same object with different spectra only relying on single-phase multispectral images; it is easily affected by weather such as clouds and rains, and the data acquisition is unstable; the detection ability for submerged vegetation is greatly affected by factors such as water transparency and depth, and the accuracy is limited; it is difficult to achieve fine classification at the species level or community level; common methods such as threshold methods based on vegetation indices (NDVI, EVI, FAI, SVSI, etc.) or traditional supervised classification (such as maximum likelihood method) have limited model generalization ability and automation degree, and it is difficult to capture the dynamic change process of vegetation; High-resolution remote sensing: It can provide richer spatial details, but the coverage range is relatively small, the cost is high, and it also faces the problems of spectral confusion and dynamic monitoring; Radar remote sensing: It has all-weather observation ability and is sensitive to the structural information of emergent vegetation, but the recognition ability for submerged and floating-leaved vegetation is relatively weak, and the data processing is complex.

[0004] 3) Insufficient existing fusion and time-series analysis: Existing studies have tried to fuse multi-source data or conduct time-series analysis, but there are often the following problems: a. The data fusion level is relatively shallow, and the complementary information of multi-source data cannot be fully exploited; b. The time-series analysis method is simple, and the phenological characteristics of vegetation cannot be effectively utilized for differentiation; c. The intelligent degree of the classification decision model is not high, relying on artificial setting of rules or parameters, and the adaptability is not strong; d. There is a lack of a comprehensive solution that effectively combines fine classification and dynamic monitoring.

[0005] Therefore, there is a need for an automated and highly accurate technical method that can integrate multi-source and multi-temporal remote sensing information, utilize an intelligent decision-making model, and achieve refined classification and high-frequency dynamic monitoring of aquatic vegetation to overcome the limitations of existing technologies. Summary of the Invention

[0006] The present invention aims to solve at least one of the technical problems in the above-related technologies to some extent.

[0007] To this end, the object of the present invention is to provide a method, medium, and device for refined classification and dynamic monitoring of aquatic vegetation based on multi-source temporal remote sensing information and intelligent decision-making, which can solve the problems of low accuracy, limited category discrimination ability, difficulty in dynamic monitoring, and insufficient automation and intelligence in existing aquatic vegetation monitoring.

[0008] To solve the above technical problems, the present invention is implemented as follows: An embodiment of the present invention provides a method for refined classification and dynamic monitoring of aquatic vegetation. The method extracts spatio-temporal information including vegetation indices and phenological characteristics by fusing multi-source remote sensing data, and uses machine learning or deep learning models to achieve high-precision classification of emergent-hydrophytic, submerged, and floating-leaved vegetation, and monitors dynamic changes through means of change detection and trend analysis.

[0009] In addition, according to the method for refined classification and dynamic monitoring of aquatic vegetation of the present invention, the following additional technical features may also be included: In some embodiments, the steps of the method include: S1. Acquisition and preprocessing of multi-source temporal remote sensing data; S2. Construction of multi-source temporal feature engineering; S3. Sample selection and annotation; S4. Refined classification based on an intelligent decision-making model; S5. Classification decision-making and post-processing optimization; S6. Dynamic monitoring and analysis of aquatic vegetation; S7. Result output and verification.

[0010] In some embodiments, the content of step S1 includes: S11. Data source selection: According to the monitoring target and regional characteristics, select and acquire various remote sensing data that cover the study area and have a complete time series; S12. Preprocessing: Perform standardized preprocessing on each source of remote sensing data to ensure data quality and spatial consistency.

[0011] In some embodiments, the content of step S2 includes: S21. Optical index calculation: Calculate vegetation indices, water body indices, and texture features; S22. Radar feature extraction: Extract SAR backscattering coefficients, polarization decomposition parameters, interferometric coherence, and texture features; S23. Filtering, smoothing, and data extraction: Perform filtering and smoothing on the preprocessed long-term time series data, and extract key phenological parameters, time series statistics, and time series change trends; S24. Feature fusion and extraction: Fusion features from different data sources and different times, and perform feature extraction at the pixel level or feature level to obtain spatial context features.

[0012] In some of the embodiments, the content of step S3 includes: S31. Sample data source: Combine multiple channels to obtain ground truth information; S32. Determine the sampling strategy and perform annotation; S33. Dataset division: Randomly divide the labeled total sample set into three independent subsets: training set, validation set, and test set according to a certain ratio.

[0013] In some of the embodiments, the content of step S4 includes: S41. Feature selection and dimensionality reduction: Use feature selection methods to screen out the feature subset that contributes the most to classification, and use dimensionality reduction techniques to reduce data redundancy to improve the model training efficiency and generalization ability; S42. Model training: According to the classification objective and feature characteristics, select a suitable machine learning model or deep learning model, and use the ground truth samples and the optimized feature set to train the selected model; S43. Parameter tuning: During the model training process, use the cross-validation method, combined with grid search and random search, to adaptively adjust the key hyperparameters of the model to obtain the optimal model performance; S44. Model validation: Use an independent validation dataset to strictly evaluate the performance of the trained model and calculate the accuracy metrics.

[0014] In some of the embodiments, the content of step S5 includes: S51. Model integration: Integrate multiple independently trained models, and combine their prediction results through voting and weighted average strategies to form the final classification decision; S52. Region-wide classification prediction: Apply the finally selected and trained single model or integrated model to the feature data of the entire study area to generate a preliminary fine classification map of aquatic vegetation at the pixel level or object level. S53. Uncertainty assessment: While generating the classification results, calculate the classification uncertainty or confidence of each taxonomic unit by using the probability output of the model itself or the prediction differences among integrated models, and generate an uncertainty thematic map to provide a basis for the reliability analysis and application of the results. S54. Post-classification processing: Post-process and optimize the preliminary classification results as needed, remove small patches with too small areas, or correct or mark low-confidence regions in combination with the uncertainty assessment results.

[0015] In some of these implementation manners, the preprocessing in step S12 includes radiometric calibration, atmospheric correction, geometric precise correction and registration, noise suppression, cloud / shadow detection and masking, and water body boundary extraction.

[0016] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it realizes the content of the aquatic vegetation fine classification and dynamic monitoring method described in any one of the above.

[0017] The embodiment of the present invention also provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor. The processor executes the computer program to realize the content of the aquatic vegetation fine classification and dynamic monitoring method described in any one of the above.

[0018] Compared with the prior art, the present invention has at least the following beneficial effects: In the embodiment of the present invention, the provided aquatic vegetation fine classification and dynamic monitoring method, through deeply integrating multi-source remote sensing data such as optical and radar, combining temporal features such as vegetation phenology, and using advanced machine learning or deep learning models for intelligent decision-making, realizes high-precision and refined classification of aquatic vegetation (emergent-hydrophytic, submerged, floating-leaved), and can effectively monitor the dynamic changes in its distribution range, density, and type, providing strong technical support for the precise management and scientific research of the water ecological environment. In the embodiment of the present invention, the provided aquatic vegetation fine classification and dynamic monitoring method, through the fusion of multi-source temporal information and intelligent decision-making, can greatly improve the classification accuracy and reliability of aquatic vegetation compared with traditional methods. In the embodiment of the present invention, the provided aquatic vegetation fine classification and dynamic monitoring method can distinguish more aquatic vegetation subclasses or community types, provide more detailed vegetation structure information, and meet the needs of ecological research and refined management. In the embodiment of the present invention, the provided aquatic vegetation fine classification and dynamic monitoring method can capture the intra-annual and inter-annual dynamic changes in the distribution range, density, and type of aquatic vegetation, providing a basis for timely warning, effect evaluation, and adaptive management. In the embodiments of the present invention, the provided method for fine classification and dynamic monitoring of aquatic vegetation improves the processing efficiency with an automated and intelligent process. Combining the wide-area coverage ability of satellite remote sensing, high-frequency monitoring of large-scale water areas can be achieved. In the embodiments of the present invention, the provided method for fine classification and dynamic monitoring of aquatic vegetation has a stronger adaptability of the intelligent model to different water environments and different data qualities, reducing the dependence on specific regions or empirical thresholds. In the embodiments of the present invention, the provided method for fine classification and dynamic monitoring of aquatic vegetation can directly serve application fields such as water environment management, ecological restoration planning, and biodiversity protection with accurate and dynamic aquatic vegetation distribution information.

[0019] Additional aspects and advantages of the present invention will be given in part in the following description, become apparent in part from the following description, or be learned through the practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a flowchart of the method for fine classification and dynamic monitoring of aquatic vegetation disclosed in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

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

[0022] The embodiments of the present invention will be described in detail below with reference to the drawings through specific embodiments and their application scenarios.

[0023] Please refer to Figure 1 As shown, in some embodiments of the present invention, a method for fine classification and dynamic monitoring of aquatic vegetation is provided, mainly including the following steps: Step 1. Acquisition and preprocessing of multi-source time-series remote sensing data 1.1 Data source selection: According to the monitoring objectives and regional characteristics, select and acquire various remote sensing data that cover the study area and have a complete time series.

[0024] Optical data: Medium-high resolution multi-spectral data (such as Sentinel-2, Landsat 8 / 9, GF-1 / 2 / 6, etc.), providing rich spectral and texture information; high-resolution data or hyperspectral data can be selectively added to improve the fine classification ability.

[0025] Radar data: Synthetic Aperture Radar data (such as Sentinel-1, GF-3, HJ2E / F, etc.), leveraging its all-weather observation capabilities and sensitivity to vegetation structure, especially for the identification of emergent vegetation.

[0026] Auxiliary data: DEM digital elevation model, measured water depth data, water quality parameters, meteorological data, historical vegetation distribution maps, ground survey sample point data, etc.

[0027] 1.2 Preprocessing: Perform standardized preprocessing on each source data, including radiometric calibration, atmospheric correction (optical), geometric precise correction and registration, noise suppression (SAR), cloud / shadow detection and masking, water body boundary extraction, etc., to ensure data quality and spatial consistency.

[0028] Step 2: Construction of multi-source time-series feature engineering 2.1 Calculation of optical indices: Calculate various vegetation indices (such as NDVI, EVI, NDWI, MNDWI, FAI, EVSI, SVSI, etc.), water body indices, and texture features.

[0029] 2.2 Extraction of radar features: Extract SAR backscattering coefficients, polarization decomposition parameters, interferometric coherence, texture features, etc.

[0030] 2.3 For the preprocessed long-term time-series data, perform filtering and smoothing, and extract key phenological parameters, time-series statistics, time-series change trends, etc.

[0031] 2.4 Fuse features from different sensors and different times. Perform fusion at the pixel level or at the feature level to extract spatial context features.

[0032] Step 3: Sample selection and annotation 3.1 Sources of sample data: Combine multiple channels to obtain ground truth information, including but not limited to: Field surveys: Use devices such as GPS, cameras, drones, and fishing traps to record information such as the exact location, vegetation type, community structure, and growth status of sample points / sample plots; High-resolution image interpretation: Utilize aerial images, drone images, or commercial high-resolution satellite images with similar transit times to medium-resolution satellites, and delineate representative aquatic vegetation patches or sample points through visual interpretation or semi-automatic methods; Reference to historical materials: Refer to existing aquatic vegetation investigation reports, historical classification maps, literature materials, etc. in the study area as auxiliary verification or sample sources.

[0033] 3.2 Sampling strategy and annotation Sampling strategy: Adopt scientific sampling methods to ensure the representativeness and coverage of aquatic vegetation samples; Sample annotation: According to the preset classification criteria, each collected aquatic vegetation sample is accurately labeled with its category, and a standardized sample database is established, which includes information such as sample ID, geographical coordinates, category labels, sources, collection time, etc.

[0034] 3.3 Dataset division: The total labeled sample set is randomly divided into three independent subsets, namely the training set, validation set, and test set, according to a certain ratio (such as 70% / 15% / 15% or 60% / 20% / 20%).

[0035] Step 4: Fine classification based on the intelligent decision-making model 4.1 Feature selection / dimensionality reduction: Intelligent feature selection methods (such as model-based recursive feature elimination RFE, sparsity-based LASSO, etc.) are used to screen out the feature subset that contributes the most to classification, and dimensionality reduction techniques are adopted to reduce data redundancy, so as to improve the model training efficiency and generalization ability.

[0036] 4.2 Model training: According to the classification objective and feature characteristics, a suitable machine learning model or deep learning model is selected. Using high-quality ground truth samples and combined with the optimized feature set, the selected model is trained.

[0037] 4.3 Parameter tuning: During the model training process, methods such as cross-validation are adopted, combined with techniques such as grid search and random search, to adaptively adjust the key hyperparameters of the model to obtain the optimal model performance.

[0038] 4.4 Model validation: Use an independent validation dataset to strictly evaluate the performance of the trained model and calculate the accuracy metrics.

[0039] Step 5: Classification decision-making and post-processing optimization 5.1 Model integration: Integrate multiple independently trained models, and combine their prediction results through strategies such as voting and weighted averaging to form the final classification decision.

[0040] 5.2 Region-wide classification prediction: Apply the finally selected and trained single model or integrated model to the feature data of the entire study area to generate a preliminary fine classification map of aquatic vegetation at the pixel level or object level.

[0041] 5.3 Uncertainty assessment: When generating the classification results, use the probability output of the model itself or the prediction differences between integrated models to calculate the classification uncertainty or confidence of each classification unit, and generate an uncertainty thematic map to provide a basis for the reliability analysis and application of the results.

[0042] 5.4 Post-classification processing: As needed, the preliminary classification results can be post-processed and optimized to remove small fragmented patches or to correct or mark low-confidence regions in combination with the uncertainty assessment results.

[0043] Step 6, Dynamic monitoring and analysis of aquatic vegetation 6.1 Change detection: By comparing the fine classification result maps of different years or key periods, methods such as change matrices, image difference methods, and post-classification comparison are used to generate maps of the transfer of aquatic vegetation types and maps of area increase and decrease.

[0044] 6.2 Trend analysis: Analyze the change trends of the areas and spatial distribution centers of gravity of various types of aquatic vegetation in a long time series (such as year by year).

[0045] 6.3 Driving force analysis: Combine environmental factors (water quality, water level, air temperature, etc.) and socioeconomic data to analyze the driving factors for changes in aquatic vegetation.

[0046] Step 7, Result output and verification: Generate thematic maps of the fine classification of aquatic vegetation for the specified year or period, area statistical tables for each category, thematic maps of dynamic changes, change statistical reports, and uncertainty maps, etc.

[0047] The present invention adopts deep fusion of multi-source heterogeneous data: breaking through the limitations of a single sensor, organically integrating optical satellite data, radar satellite data, and other auxiliary information to achieve complementary information and significantly improve the identification ability for complex water environments and different vegetation types.

[0048] The present invention adopts collaborative utilization of spatio-temporal features: fully exploiting the time series information of remote sensing images, combining the phenological laws of vegetation and the temporal dynamic characteristics of spectra / backscattering to effectively distinguish vegetation types with similar spectra but different phenologies and to capture intra-annual and inter-annual changes.

[0049] The present invention is driven by an intelligent decision-making model: adopting advanced machine learning or deep learning models to replace traditional thresholds or simple classifiers, capable of automatically learning the complex relationships between high-dimensional, non-linear features and vegetation types, achieving higher-precision and more robust automated classification, and having the ability to process big data and conduct uncertainty assessment.

[0050] The present invention adopts integration of refined classification and dynamic monitoring: not only can it distinguish the main vegetation life forms (emergent - hygrophytic, submerged, floating-leaved), but also can develop towards more refined sub-class or community-level classification, and apply the high-precision classification results to dynamic monitoring to achieve accurate understanding of the spatio-temporal evolution process of aquatic vegetation.

[0051] The present invention has a high level of automation and intelligence: The entire process design aims to improve the degree of automation and reduce manual intervention, especially in the feature extraction and classification decision-making links, to meet the needs of large-scale and long-time-series operational monitoring.

[0052] For the parts not described in detail in the present invention, reference can be made to the prior art or the well-known technology in the art. This embodiment does not make any limitations thereto and will not be described in detail herein.

[0053] The embodiments of the present invention have been described above in conjunction with the accompanying drawings. However, the present invention is not limited to the above specific embodiments. The above specific embodiments are merely illustrative rather than restrictive. Under the inspiration of the present invention, those of ordinary skill in the art can also make many forms without departing from the spirit of the present invention and the scope protected by the claims, and all of them fall within the protection scope of the present invention.

Claims

1. A method for fine classification and dynamic monitoring of aquatic vegetation, characterized in that, The method extracts spatiotemporal information including vegetation index and phenological characteristics by fusing multi-source remote sensing data, uses machine learning or deep learning models to achieve high-precision classification of emergent-wetland, submerged, and floating-leaf vegetation, and monitors dynamic changes through change detection and trend analysis.

2. The fine classification and dynamic monitoring method of aquatic vegetation according to claim 1, characterized in that The steps of the method include: S1. Acquisition and preprocessing of multi-source time series remote sensing data; S2, construction of multi-source time series feature engineering; S3, sample selection and annotation; S4, fine classification based on intelligent decision-making model; S5, classification decision and post-processing optimization; S6. Dynamic monitoring and analysis of aquatic vegetation; S7. Result output and verification.

3. The fine classification and dynamic monitoring method for aquatic vegetation according to claim 2, characterized in that, The contents of step S1 include: S11. Data source selection: According to the monitoring objectives and regional characteristics, select and obtain a variety of remote sensing data covering the study area and with complete time series; S12. Preprocessing: Standardize the preprocessing of remote sensing data from each source to ensure data quality and spatial consistency.

4. The fine classification and dynamic monitoring method for aquatic vegetation according to claim 2, characterized in that The content of step S2 includes: S21, optical index calculation: calculate vegetation index, water index and texture characteristics; S22, radar feature extraction: extract SAR backscattering coefficient, polarization decomposition parameters, interference coherence and texture features; S23, filtering, smoothing and data extraction: filtering and smoothing the preprocessed long time series data to extract key phenological parameters, time series statistics and time series change trends; S24, feature fusion and extraction: fuse the features from different data sources and at different times, and extract features at the pixel level or feature level to obtain spatial context features.

5. The fine classification and dynamic monitoring method of aquatic vegetation according to claim 2, characterized in that, The content of step S3 includes: S31. Sample data source: Combine multiple approaches to obtain ground truth information; S32, determine the sampling strategy and mark it; S33. Dataset division: The total labeled sample set is randomly divided into three independent subsets: training set, validation set, and test set according to a certain ratio.

6. The fine classification and dynamic monitoring method for aquatic vegetation according to claim 2, characterized in that, The content of step S4 includes: S41. Feature selection and dimensionality reduction: Use feature selection methods to screen out the feature subset that contributes most to classification, and use dimensionality reduction technology to reduce data redundancy to improve model training efficiency and generalization ability; S42, Model training: Select a suitable machine learning model or deep learning model according to the classification target and feature characteristics, and train the selected model using ground truth samples combined with the optimized feature set; S43, parameter tuning: During the model training process, the cross-validation method is used, combined with grid search and random search, to adaptively adjust the key hyperparameters of the model to obtain the optimal model performance; S44. Model Validation: Use an independent validation dataset to conduct rigorous performance evaluation of the trained model and calculate accuracy metrics.

7. The fine classification and dynamic monitoring method of aquatic vegetation according to claim 2, characterized in that The content of step S5 includes: S51, model integration: Integrate multiple independently trained models, synthesize their prediction results through voting and weighted average strategies, and form the final classification decision; S52, whole-area classification prediction: Apply the final selected and trained single model or integrated model to the feature data of the entire study area to generate a preliminary aquatic vegetation fine classification map at the pixel level or object level; S53. Uncertainty assessment: While generating the classification results, use the probability output of the model itself or the prediction differences between integrated models to calculate the classification uncertainty or confidence of each classification unit, and generate an uncertainty thematic map to provide a basis for the reliability analysis and application of the results; S54. Post-classification processing: Post-process and optimize the preliminary classification results as needed, remove small patches with too small an area, or correct or mark low-confidence regions in combination with the uncertainty assessment results.

8. The fine classification and dynamic monitoring method for aquatic vegetation according to claim 3, characterized in that, The preprocessing in step S12 includes radiometric calibration, atmospheric correction, geometric precise correction and registration, noise suppression, cloud / shadow detection and masking, and water body boundary extraction.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the content of the method for fine classification and dynamic monitoring of aquatic vegetation according to any one of claims 1-8.

10. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor executes the computer program to implement the content of the method for fine classification and dynamic monitoring of aquatic vegetation according to any one of claims 1-8.

Images