Coastal wetland vegetation degradation dynamic monitoring method based on multi-modal transfer learning

Abstract

The invention discloses a coastal wetland vegetation degradation dynamic monitoring method based on multi-modal transfer learning, and belongs to the technical field of ecological environment monitoring, and the method comprises the steps: data acquisition: collecting coastal wetland region data through a plurality of sensors, feature extraction and fusion: preprocessing multi-source data, extracting space, time series and physical parameter features, and carrying out feature fusion; and fusing the data features of different modalities. According to the method, all-around and high-precision perception of the coastal wetland ecological environment is realized, the spatial-temporal resolution of data acquisition is improved, the response capability to complex environmental factors is enhanced, the generalization capability and prediction precision of the model are effectively improved, key environmental factors such as soil salinity can be incorporated into an evaluation system, and the evaluation accuracy is improved. The scientificity and pertinence of ecological risk identification are further improved, the efficiency and response speed of coastal wetland ecological management are improved, and powerful support is provided for formulating scientific and reasonable ecological protection and restoration measures.

Description

Technical Field

[0001] The present invention belongs to the technical field of ecological environment monitoring, and specifically provides a dynamic monitoring method for coastal wetland vegetation degradation based on multimodal transfer learning. Background Art

[0002] Coastal wetlands, as a key transition zone between terrestrial and marine ecosystems, have important ecological functions and biodiversity value. Their vegetation plays an irreplaceable role in maintaining ecosystem stability, preventing coastal erosion, regulating climate and carbon sequestration. However, affected by multiple factors such as global climate change, sea level rise, human interference and increased soil salinization, coastal wetland vegetation is facing increasingly serious degradation problems, manifested in phenomena such as decreased vegetation coverage, changes in species composition, disrupted growth cycles and declining ecosystem functions.

[0003] Existing coastal wetland vegetation monitoring methods mainly rely on manual surveys and single remote sensing image analysis. Manual surveys are costly and inefficient, making it difficult to achieve large-scale, high-frequency dynamic monitoring. In addition, methods based on a single remote sensing data source are limited by factors such as spatial resolution, temporal coverage, and cloud cover, making it difficult to fully reflect the health status of vegetation and its changing trends. In addition, the monitoring process fails to fully consider the comprehensive impact of key environmental factors such as soil salinity, temperature and humidity in northern wetlands on vegetation degradation. Summary of the Invention

[0004] The purpose of the present invention is to provide a dynamic monitoring method for coastal wetland vegetation degradation based on multimodal transfer learning to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: comprising the following steps:

[0006] S1. Data acquisition: Collect data on coastal wetland areas through a variety of sensors;

[0007] S2. Feature extraction and fusion: Preprocess multi-source data and extract spatial, temporal and physical parameter features, and fuse data features from different modalities;

[0008] S3. Intelligent Prediction and Assessment: Predict vegetation degradation status, make adaptive adjustments for target coastal wetlands, propose corresponding ecological protection suggestions based on the ecological environment, and monitor and assess soil salinity;

[0009] S4. Dynamic monitoring: Update data regularly, continuously monitor changes in vegetation health, and adjust protection strategies in a timely manner;

[0010] As a further preferred embodiment of the present technical solution: the S1 data acquisition: clarifying the monitoring objectives, such as vegetation coverage, health status, and degradation trends, determining the types of data to be collected, such as multispectral images, radar data, meteorological data, and soil moisture, delineating monitoring grids in the target wetlands, deploying satellite remote sensing receiving stations, drone aerial survey points, and ground sensor nodes in each grid, acquiring regular hyperspectral images through satellite platforms, performing high-resolution imaging with drones, and installing temperature, humidity, soil conductivity, and wind speed sensors on the ground to monitor the ground soil environment;

[0011] As a further preferred embodiment of the present technical solution: S1, using the API interface to download the latest images from the satellite operator, regularly checking and recording the data of the ground sensors to ensure continuity and accuracy, converting data from different sources into a consistent format for subsequent processing, removing noise interference, correcting errors, improving data reliability, and ensuring that the data has the same geographic coordinate system;

[0012] As a further preferred embodiment of the present technical solution: the S2 feature extraction and fusion: using remote sensing images to extract vegetation index to characterize vegetation coverage and health status, extracting texture features, edge information and surface spatial structure information of ground object distribution pattern, and combining with geographic information system to analyze the spatial distribution pattern of vegetation;

[0013] As a further preferred embodiment of the present technical solution: S2 performs trend analysis on long-term remote sensing data series to identify vegetation growth cycles, degradation inflection points, and recovery capabilities, calculates monthly averages, seasonal variation amplitudes, and interannual fluctuation statistical characteristics, reveals the dynamic change patterns of vegetation, and analyzes the temporal evolution characteristics of ground sensor data, and the changes in temperature, humidity, and salinity with seasons or extreme climate events;

[0014] As a further preferred embodiment of the present technical solution: S2, establishing a unified data feature representation framework, integrating the spatial features, time series features of remote sensing images and the physical parameter features of ground sensors, constructing a comprehensive feature vector by using weighted fusion, principal component analysis and feature splicing according to the characteristics of different modal data, designing a multi-input structure to process different modal data respectively, and fusing them at the middle layer to achieve cross-modal information complementarity;

[0015] As a further preferred embodiment of the present technical solution: the S3 intelligent prediction and evaluation: uses a convolutional neural network, a recurrent neural network or a multimodal fusion model with good spatial or temporal modeling capabilities, and the model output is a vegetation degradation level classification result of "healthy", "mildly degraded", "moderately degraded" and "severely degraded". The fused multimodal data is input into the optimized model, and the development trend of vegetation degradation in the future is predicted based on time series analysis;

[0016] As a further preferred embodiment of this technical solution: S3 utilizes a ground sensor network to collect soil electrical conductivity data in real time, combines remote sensing inversion technology to obtain a large-scale soil salinity distribution, analyzes the spatial correlation between vegetation degradation and soil salinity, identifies degradation hotspots caused by salinization, introduces salt indicators as key input variables into the prediction model, improves the accuracy of degradation assessment, divides areas into different risk levels, assists in formulating governance strategies, and automatically matches the corresponding ecological protection and restoration strategy library based on the vegetation degradation level and soil salinity assessment results output by the model;

[0017] As a further preferred embodiment of the present technical solution: the S4 dynamic monitoring: formulate a clear data update frequency plan, obtain remote sensing images on a weekly basis, update ground sensor data daily, conduct regular drone flight missions, collect high-resolution images of key areas, generate new vegetation health distribution maps, identify areas with degradation trend changes, identify areas with sudden changes in vegetation cover and abnormal salinity increases by comparing historical data with current monitoring results, set a threshold warning mechanism, and when the degradation level in a certain area increases or the salinity exceeds the standard, the system automatically triggers a warning prompt;

[0018] As a further preferred embodiment of this technical solution: S4, based on the latest monitoring results, evaluates the effectiveness of existing protection measures, and proposes targeted adjustment suggestions for areas with intensified degradation, such as increasing the frequency of water replenishment, changing plant varieties and implementing enclosures, and generates periodic ecological health reports, which include changes in vegetation status, degradation trends, analysis of key areas and countermeasures, and derives key indicators for decision-making reference.

[0019] Compared with the prior art, the present invention has the following beneficial effects:

[0020] 1. In the present invention, by constructing an integrated "sky-air-ground" multi-source sensor collaborative acquisition system, a comprehensive and high-precision perception of the coastal wetland ecological environment is achieved, which not only improves the spatiotemporal resolution of data acquisition, but also enhances the response capability to complex environmental factors, thereby providing solid data support for the accurate assessment of vegetation health status. By integrating information from different sources through weighted fusion, principal component analysis and feature splicing, a more representative comprehensive feature vector is constructed, which effectively improves the generalization ability and prediction accuracy of the model.

[0021] 2. In the present invention, through accurate classification of vegetation degradation levels and prediction of future trends, the system can identify degradation hotspots and incorporate key environmental factors such as soil salinity into the assessment system, further improving the scientific nature and pertinence of ecological risk identification, and realizing intelligent and automated monitoring mode. It not only improves the efficiency and response speed of coastal wetland ecological management, but also provides strong support for the formulation of scientific and reasonable ecological protection and restoration measures, and has good prospects for promotion and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 The operating process of the coastal wetland vegetation degradation dynamic monitoring method based on multimodal transfer learning of the present invention Figure 1 ;

[0023] Figure 2 The operating process of the coastal wetland vegetation degradation dynamic monitoring method based on multimodal transfer learning of the present invention Figure 2 ;

[0024] Figure 3 The operating process of the coastal wetland vegetation degradation dynamic monitoring method based on multimodal transfer learning of the present invention Figure 3 . DETAILED DESCRIPTION

[0025] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0026] Example

[0027] See also Figure 1-Figure 3 As shown, the present invention provides a technical solution: comprising the following steps:

[0028] S1. Data acquisition: Collect data on coastal wetland areas through a variety of sensors;

[0029] S2. Feature extraction and fusion: Preprocess multi-source data and extract spatial, temporal and physical parameter features, and fuse data features from different modalities;

[0030] S3. Intelligent Prediction and Assessment: Predict vegetation degradation status, make adaptive adjustments for target coastal wetlands, propose corresponding ecological protection suggestions based on the ecological environment, and monitor and assess soil salinity;

[0031] S4. Dynamic monitoring: Update data regularly, continuously monitor changes in vegetation health, and adjust protection strategies in a timely manner;

[0032] In this embodiment, specifically: S1 data acquisition: clarify the monitoring target, determine the type of data to be collected, delineate a monitoring grid in the target wetland, deploy satellite remote sensing receiving stations, drone aerial survey points and ground sensor nodes in each grid, obtain regular hyperspectral images through satellite platforms, perform high-resolution imaging with drones, and install ground-mounted temperature, humidity, soil conductivity and wind speed sensors to monitor the ground soil environment;

[0033] In this embodiment, specifically: S1 uses an API interface to download the latest images from the satellite operator, regularly checks and records the data of ground sensors, converts data from different sources into a consistent format, removes noise interference, corrects errors, and ensures that the data has the same geographic coordinate system;

[0034] In this embodiment, specifically: the S2 feature extraction and fusion: using remote sensing images to extract vegetation indices to characterize vegetation coverage and health status, extracting texture features, edge information and surface spatial structure information of ground feature distribution patterns, and combining with geographic information systems to analyze the spatial distribution pattern of vegetation;

[0035] In this embodiment, specifically: S2 performs trend analysis on long-term remote sensing data series to identify vegetation growth cycles, degradation inflection points, and recovery capabilities, calculates monthly averages, seasonal variation amplitudes, and interannual fluctuation statistical characteristics, reveals the dynamic change patterns of vegetation, and analyzes the temporal evolution characteristics of ground sensor data, and the changes in temperature, humidity, and salinity with seasons or extreme climate events;

[0036] In this embodiment, specifically: S2 establishes a unified data feature representation framework, integrates the spatial features, time series features of remote sensing images, and physical parameter features of ground sensors, and constructs a comprehensive feature vector based on the characteristics of different modal data using weighted fusion, principal component analysis, and feature splicing. A multi-input structure is designed to process different modal data separately, and fusion is performed in the middle layer;

[0037] In this embodiment, specifically: the S3 intelligent prediction and evaluation uses a convolutional neural network, a recurrent neural network, or a multimodal fusion model. The model output is a classification result of vegetation degradation level, which is "healthy," "mildly degraded," "moderately degraded," and "severely degraded." The fused multimodal data is input into the optimized model, and based on time series analysis, the development trend of vegetation degradation in the future is predicted.

[0038] In this embodiment, specifically: S3 utilizes a ground sensor network to collect soil electrical conductivity data in real time, combines remote sensing inversion technology to obtain a large-scale soil salinity distribution, analyzes the spatial correlation between vegetation degradation and soil salinity, identifies degradation hotspots caused by salinization, introduces salinity indicators as key input variables into the prediction model, and automatically matches the corresponding ecological protection and restoration strategy library based on the vegetation degradation level and soil salinity assessment results output by the model;

[0039] In this embodiment, specifically: S4 dynamic monitoring: formulate a clear data update frequency plan, obtain remote sensing images on a weekly basis, update ground sensor data daily, and conduct regular drone flight missions to collect high-resolution images of key areas. By comparing historical data with current monitoring results, areas with sudden changes in vegetation cover and abnormally increased salinity can be identified. When the degradation level in a certain area increases or the salinity exceeds the standard, the system automatically triggers an early warning prompt;

[0040] In this embodiment, specifically: S4, based on the latest monitoring results, evaluates the effectiveness of existing protection measures, and proposes targeted adjustment suggestions for areas with intensified degradation, such as increasing the frequency of water replenishment, changing plant varieties, and implementing enclosures, and generates periodic ecological health reports, which include changes in vegetation status, degradation trends, key area analysis, and countermeasures, and derives key indicators for decision-making reference.

[0041] Working principle or structural principle: In order to achieve all-round perception of the coastal wetland environment, the system works together through a variety of sensor networks, including satellite remote sensing platforms, drone aerial photography equipment and ground sensor nodes, etc., to build an integrated "sky, air and ground" data collection architecture, delineate standard monitoring grids within the target wetland range, and deploy corresponding observation points in each grid. The satellite remote sensing receiving station regularly acquires hyperspectral images for macro-scale vegetation cover analysis, and the drone performs periodic flight missions to provide high-resolution images of local areas and enhance the ability to capture details. The ground sensor nodes collect key environmental parameters such as temperature, humidity, wind speed and soil conductivity in real time to ensure accurate grasp of the wetland microclimate and soil conditions. The collected data must undergo pre-processing operations such as unified format conversion, noise removal, and error correction to ensure data consistency and availability. At the same time, standardized registration of the geographic coordinate system is also carried out to ensure that data from different sources have a good spatial correspondence, laying the foundation for subsequent modeling and analysis;

[0042] After completing the preliminary data preparation, the system enters the feature extraction and fusion stage, extracting vegetation indices from remote sensing images to quantify the growth status and coverage of vegetation. Combining image texture analysis, edge detection and other means, it extracts surface structure information to assist in identifying vegetation distribution patterns and their changing trends. It conducts in-depth mining of time series data to analyze the growth cycle, degradation inflection point and recovery potential of vegetation with seasonal changes. By calculating statistical characteristics such as monthly averages, interannual fluctuations, and seasonal amplitudes, it reveals the dynamic evolution of vegetation status. At the same time, time series data collected by ground sensors are also included in the analysis framework to further characterize temperature and humidity. In order to more effectively integrate information from different modalities, the system establishes a unified feature representation framework to more effectively integrate information from different modalities. Based on the characteristics of different data types, it adopts weighted fusion, principal component analysis, or feature splicing to organically integrate the spatial features of remote sensing images, the change characteristics of time series, and the physical parameter characteristics of ground sensors to construct a highly representative comprehensive feature vector. In addition, it designs a multi-input neural network structure to process different types of data streams respectively and perform information fusion in the middle layer, thereby achieving cross-modal information complementarity and synergy, and improving the expressiveness of the overall model.

[0043] Based on feature fusion, the system introduces transfer learning technology to construct an intelligent prediction model for vegetation degradation status. The model can be adaptively adjusted according to the specific ecological and environmental characteristics of the target wetland. The bottom network is frozen to retain its general feature extraction capabilities, the top network structure is redesigned and some parameters are fine-tuned to better adapt to the current task requirements. Domain adaptation technology is introduced to narrow the data distribution differences between the source domain and the target domain, and improve the migration effect of the model. The model outputs the vegetation degradation level classification results, which are divided into "healthy", "mildly degraded", "moderately degraded" and "severely degraded", and predicts the degradation development trend in the future through time series modeling. In this process, the system pays special attention to the impact of soil salt content, using ground sensor networks and remote sensing inversion technology to obtain large-scale soil salt distribution, analyze its spatial correlation with vegetation degradation, and identify degradation hotspots caused by salinization;

[0044] The system sets a clear data update frequency. Remote sensing images are acquired weekly, ground sensor data is uploaded daily, and drones are flown regularly to collect images of key areas to ensure the timeliness and completeness of monitoring information. The system compares historical data with current monitoring results to identify areas of sudden changes such as a sudden drop in vegetation cover and an abnormal increase in salinity, and sets a threshold warning mechanism. Once the degradation level in a certain area is found to have increased or the salinity exceeds the standard, the system immediately triggers an early warning prompt and notifies management personnel to intervene in time. The system also has strategy optimization capabilities. Based on the latest monitoring results, the system can evaluate the effectiveness of existing protection measures and propose adjustment suggestions for areas with intensified degradation, such as increasing the frequency of water replenishment, changing plant varieties, and implementing fencing projects. At the same time, the system will regularly generate ecological health reports covering changes in vegetation status, degradation trends, key area analysis and countermeasures, and support the export of key indicators for decision-making reference, forming a closed-loop management system of "monitoring, evaluation, intervention, and re-monitoring."

[0045] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A dynamic monitoring method for coastal wetland vegetation degradation based on multimodal transfer learning, characterized by: The following steps are involved: S1. Data acquisition: Collect data on coastal wetland areas through a variety of sensors; S2. Feature extraction and fusion: Preprocess multi-source data and extract spatial, temporal and physical parameter features, and fuse data features from different modalities; S3. Intelligent Prediction and Assessment: Predict vegetation degradation status, make adaptive adjustments for target coastal wetlands, propose corresponding ecological protection suggestions based on the ecological environment, and monitor and assess soil salinity; S4. Dynamic monitoring: Update data regularly, continuously monitor changes in vegetation health, and adjust protection strategies in a timely manner.

2. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 1 is characterized by: The S1 data acquisition is as follows: clarify the monitoring objectives, determine the types of data that need to be collected, delineate monitoring grids in the target wetlands, deploy satellite remote sensing receiving stations, drone aerial survey points and ground sensor nodes in each grid, obtain regular hyperspectral images through satellite platforms, perform high-resolution imaging with drones, and install temperature, humidity, soil conductivity and wind speed sensors on the ground to monitor the ground soil environment.

3. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 2 is characterized by: The S1 uses the API interface to download the latest images from satellite operators, regularly checks and records the data of ground sensors, converts data from different sources into a consistent format, removes noise interference, corrects errors, and ensures that the data has the same geographic coordinate system.

4. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 1 is characterized by: The S2 feature extraction and fusion: remote sensing images are used to extract vegetation indices to characterize vegetation coverage and health status, and texture features, edge information, and surface spatial structure information of object distribution patterns are extracted. Combined with geographic information systems, the spatial distribution pattern of vegetation is analyzed.

5. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 4 is characterized by: The S2 performs trend analysis on long-term remote sensing data series, identifies vegetation growth cycles, degradation inflection points, and recovery capabilities, calculates monthly averages, seasonal variation amplitudes, and interannual fluctuation statistical characteristics, reveals the dynamic patterns of vegetation changes, analyzes the temporal evolution characteristics of ground sensor data, and the changes in temperature, humidity, and salinity with seasons or extreme climate events.

6. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 4 is characterized by: The S2 establishes a unified data feature representation framework, integrates the spatial features, time series features of remote sensing images and the physical parameter features of ground sensors, and constructs a comprehensive feature vector based on the characteristics of different modal data using weighted fusion, principal component analysis and feature splicing. A multi-input structure is designed to process different modal data separately and perform fusion in the middle layer.

7. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 1, characterized in that: The S3 intelligent prediction and evaluation uses a convolutional neural network, a recurrent neural network, or a multimodal fusion model. The model output is a classification result of vegetation degradation levels, including "healthy", "mildly degraded", "moderately degraded", and "severely degraded". The fused multimodal data is input into the optimized model, and based on time series analysis, the development trend of vegetation degradation in the future is predicted.

8. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 7, characterized in that: The S3 utilizes a ground sensor network to collect soil electrical conductivity data in real time, combines it with remote sensing inversion technology to obtain a large-scale soil salinity distribution, analyzes the spatial correlation between vegetation degradation and soil salinity, identifies degradation hotspots caused by salinization, and introduces salt indicators as key input variables into the prediction model. Based on the vegetation degradation level and soil salinity assessment results output by the model, the corresponding ecological protection and restoration strategy library is automatically matched.

9. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 1, characterized in that: The S4 dynamic monitoring system: formulates a clear data update frequency plan, acquires remote sensing images on a weekly basis, updates ground sensor data daily, and conducts regular drone flight missions to collect high-resolution images of key areas. By comparing historical data with current monitoring results, it identifies areas with sudden changes in vegetation coverage and abnormal increases in salinity. When the degradation level in a certain area increases or the salinity exceeds the standard, the system automatically triggers an early warning.

10. The method for dynamic monitoring of coastal wetland vegetation degradation based on multimodal transfer learning according to claim 9, characterized in that: S4, based on the latest monitoring results, evaluates the effectiveness of existing protection measures and proposes targeted adjustment suggestions for areas with intensified degradation, such as increasing the frequency of water replenishment, changing plant varieties and implementing enclosures. It also generates periodic ecological health reports that include changes in vegetation status, degradation trends, analysis of key areas and countermeasures, and derives key indicators for decision-making reference.

Images