A method for dynamic monitoring of forest resources and automatic database update based on multi-source remote sensing data

Abstract

This invention discloses a method for dynamic monitoring of forest resources and automatic database updating based on multi-source remote sensing data, belonging to the field of forestry informatization. This method addresses the inconsistencies in resolution, imaging time, and band coverage of multi-source remote sensing data by dynamically generating multi-scale grids to achieve spatial alignment and fusion of data. Utilizing temporal residual discriminative convolutional networks and sparse attention mechanisms, it intelligently detects changes in forest resources, achieving efficient and automatic attribute inversion. The system also includes automatic differential database updates and multi-version management to ensure information synchronization and reliability. Combined with external climate and fire data, it provides alarms for abnormal changes and suggestions for auxiliary intervention, forming an intelligent closed loop from monitoring to management. This method improves the automation, intelligence, and precision of forest resource monitoring and is suitable for large-scale, dynamic forest resource management scenarios.

Description

Technical Field

[0001] This invention belongs to the field of forestry informatization, and more specifically relates to a method for dynamic monitoring of forest resources and automatic database updating based on multi-source remote sensing data. Background Technology

[0002] With increasing awareness of global climate change and ecological environmental protection, the dynamic monitoring and efficient management of forest resources have become a key focus for forestry research and management departments. Traditional forest resource survey methods mainly rely on manual ground surveys and remote sensing image analysis from single data sources, which suffer from problems such as long data update cycles, limited spatial resolution, insufficient coverage, and difficulty in meeting the needs of large-scale, real-time monitoring. In recent years, with the rapid development of various remote sensing technologies such as satellites, aerial drones, and ground observation, multi-source remote sensing data with different spatial resolutions, temporal resolutions, and spectral ranges can be acquired, providing a rich data foundation for comprehensive perception and refined management of spatiotemporal changes in forest resources. However, multi-source remote sensing data vary significantly in resolution, imaging time, and spectral coverage, leading to numerous challenges in data spatiotemporal fusion, information complementarity, and unified interpretation.

[0003] In complex and ever-changing forest ecosystems, data from a single source often fails to accurately reflect the dynamic changes in forest resources. Furthermore, existing data processing and updating mechanisms generally rely on manual intervention, resulting in low levels of automation and issues such as data redundancy, information lag, and attribute inconsistencies. Simultaneously, with the continuous changes in forest resources and frequent updates to database content, achieving large-scale, automated, and accurate change detection and efficient data synchronization management has become a pressing technical challenge. In addition, traditional anomaly alerts and management recommendations largely rely on static threshold judgments, failing to fully integrate external environmental factors for intelligent decision support, thus limiting the practical application value of monitoring results. Therefore, there is an urgent need for a technical approach that can fully integrate multi-source remote sensing data, possessing efficient spatiotemporal alignment, intelligent change detection, dynamic attribute inversion, automatic database updates, and intelligent anomaly alerts. This would enhance the informatization and intelligence level of dynamic forest resource monitoring, better supporting scientific decision-making for forest protection and sustainable management. Summary of the Invention

[0004] This invention aims to address the technical bottlenecks in existing dynamic monitoring of forest resources, such as the difficulty in fusing multi-source remote sensing data, insufficient spatiotemporal alignment and feature complementarity, low levels of automation and intelligence in change detection, inadequate database update efficiency and consistency, and slow response to anomaly alarms and auxiliary decisions. The goal is to achieve high-precision, automated, and dynamic monitoring of forest resources and efficient synchronous database updates, thereby enhancing the informatization, intelligence, and scientific decision-making capabilities of forest management.

[0005] To achieve the above objectives, the present invention employs the following technical solution: the method comprises:

[0006] Heterogeneous data source fusion and spatiotemporal alignment address the inconsistencies in resolution, imaging time, and band coverage of multi-source remote sensing data. Based on the spatiotemporal labels of any two images, a multi-scale grid is dynamically generated to spatially stretch and register data from different sources.

[0007] Intelligent detection of forest resource changes: Based on time-series data, a residual discriminative convolutional network is constructed to automatically learn the spatial and structural characteristics of forest changes and introduce a sparse attention mechanism.

[0008] Cross-source dynamic attribute inversion and rule-adaptive correction: By utilizing the fusion of multi-source remote sensing features, a deep generative model-based inversion network for forest stock volume, tree height, and age class indices is designed.

[0009] The forest resource database is automatically updated and version managed. By using the change detection results of previous and subsequent time phases, it automatically generates change difference files and only writes the newly added or changed information into the database, thus improving update efficiency.

[0010] Anomaly alerts and decision support generate various intervention suggestions, achieving a closed loop from monitoring to management.

[0011] In one approach, the heterogeneous data source fusion and spatiotemporal alignment specifically include:

[0012] First, for any two input remote sensing images, a multi-scale spatial grid is dynamically generated based on the spatiotemporal labels of these two images;

[0013] Calculate the minimum common area of ​​spatial coverage between the two, and select the main reference resolution based on the resolution difference to generate a scale set.

[0014] After completing the spatial grid division, a temporal consistency discriminator is designed for remote sensing data obtained at different times. For the same grid cell, if there is no corresponding observation value at a certain time point, feature interpolation is required. The discriminator introduces an interpolation function based on temporal weighting.

[0015] After fusing features from multiple sources and time periods, a feature embedding spatial mapping method is adopted.

[0016] In one approach, the intelligent detection of forest resource changes specifically includes:

[0017] First, for the multi-source remote sensing feature sequence that has been fused and aligned, the sequence is input into the residual discriminative convolutional network; the front end of the network consists of multiple one-dimensional or two-dimensional convolutional layers stacked together, and residual connections are introduced to directly add the output of the first layer to the convolution result of the second layer.

[0018] In the process of convolutional feature extraction, a sparse attention mechanism is introduced, specifically, sparse attention weights are applied to the convolutional output.

[0019] In this way, the network can adaptively focus on locations where structural changes are suspected, automatically ignore unusable pixels or interference areas, and enhance its ability to detect forest changes.

[0020] To address the detailed distinction of change types, a multi-label voting decision was designed. Specifically, the change probability vector of each unit is first obtained at the pixel level.

[0021] In one approach, the cross-source dynamic attribute inversion and rule adaptive correction specifically includes: first, inputting the embedded features that have undergone the aforementioned multi-source spatiotemporal fusion and change detection into a deep generative inversion network;

[0022] To ensure the model's generalization ability in specific regions, a data-driven rule self-learning module was designed. Specifically, the model's prediction residuals are first subjected to spatial statistics and anomaly detection. If a systematic bias is found in a certain region, the rule optimization process is activated. New correction rules are generated by combining Bayesian inference with decision tree self-learning algorithms.

[0023] To achieve seamless adaptation to new remote sensing data sources, the model integrates a meta-learning scheme within the feature embedding layer; when a new data source is added, it automatically updates the feature mapping parameters and the generation network weights through a small-sample rapid adaptation mechanism.

[0024] In one approach, the automatic update and version management of the forest resource database specifically includes:

[0025] First, based on the forest resource change detection results of the two time phases, the attribute feature vector of each spatial object is extracted;

[0026] By calculating the difference in changes to object attributes, the system automatically determines the type of change and the validity of the data. Based on the significance of each indicator in the difference vector, difference profiles are generated only for objects that have actually been updated.

[0027] Before the data is written to the database, the system enables the data consistency verification module, whose core is a multi-source conflict detection algorithm. For each changed object, the system checks the set of attribute values ​​generated by different sources and uses a weighted consistency scoring function.

[0028] If the threshold is not met, an exception handling process is triggered, including manual review prompts or attribute correction based on regression models, to ensure the reliability and consistency of the data entering the database.

[0029] To support spatiotemporal evolution analysis and rapid backtracking of forest resources, a multi-version time-series index is designed for each spatial object. The set of all version features of the object in the time series is set, and a unique version number is generated for each update. The database adopts a key-value structure with versioning, and an index table is constructed at the same time.

[0030] In one solution, the anomaly alarm and auxiliary decision-making includes: constructing an "intelligent alarm and auxiliary intervention suggestion module": anomalies are automatically classified and linked with external climate, fire and other data to trigger alarms from relevant departments, and various intervention suggestions are generated using the output confidence interval to achieve a closed loop from monitoring to management.

[0031] Beneficial effects of this invention:

[0032] This invention innovatively integrates and aligns multi-source remote sensing data, overcoming traditional data barriers caused by differences in resolution, imaging time series, and band coverage, achieving high-precision, all-weather, dynamic monitoring of forest resources. The introduction of residual discriminative convolutional networks and sparse attention mechanisms significantly improves the automation and intelligence of forest change detection, effectively reducing manual intervention and misjudgment rates. Utilizing an attribute inversion network based on a deep generative model, key indicators such as forest stock volume, tree height, and age class can be dynamically and accurately inverted, greatly improving the timeliness and reliability of attribute information updates. The automatic database update mechanism only writes changes that are significant, and combined with version control, it not only improves data synchronization and query efficiency but also ensures the traceability and consistency of historical information. The intelligent alarm and auxiliary decision-making module can automatically classify and respond to abnormal changes, and, combined with external environmental data, intelligently generate intervention suggestions, significantly enhancing the scientific rigor and practicality of forest resource management. This invention significantly improves the informatization, automation, and intelligence of forest resource monitoring and management, providing solid support for ecological protection and sustainable management. Attached Figure Description

[0033] Figure 1 This is a flowchart of the method of the present invention. Detailed Implementation

[0034] To facilitate understanding of the present invention, a more complete description will be given below with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete.

[0035] Unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as understood by one of ordinary skill in the art to which this invention pertains. The terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the invention. To facilitate understanding, the invention will now be described more fully with reference to the accompanying drawings. Typical embodiments of the invention are shown in the drawings. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to make the disclosure of the invention more thorough and complete.

[0036] like Figure 1 As shown, a method for dynamic monitoring of forest resources and automatic database updating based on multi-source remote sensing data includes the following steps:

[0037] Step 1: Heterogeneous Data Source Fusion and Spatiotemporal Alignment

[0038] To address the inconsistencies in resolution, imaging time, and band coverage among multi-source remote sensing data (such as optical satellite, SAR radar, and lidar), an "adaptive multi-scale spatiotemporal grid fusion algorithm" is proposed. Based on the spatiotemporal labels of any two images, a multi-scale grid is dynamically generated, spatially stretching and registering data from different sources. A temporal consistency discriminator is designed to perform feature interpolation on data points with overlapping time phases, ensuring consistent temporal features within the same spatial unit. A feature embedding spatial mapping method is used to normalize the data from each source into a unified multi-dimensional feature vector, eliminating inter-source noise and redundancy.

[0039] First, for any two input remote sensing images, denoted as I... A and I B Its corresponding spatial resolution is R A R B The imaging time is T A T B The algorithm dynamically generates a multi-scale spatial grid based on the spatiotemporal labels of the two images. Specifically, it calculates the minimum common area Ω of the spatial coverage of the two images and selects the primary reference resolution R based on the resolution difference. * =min(R) A ,R B Based on this, a scale set {G} is generated on Ω. k}(where each G) k The grid division represents a scale (k is the scale index), which allows for finer grid refinement in high-resolution regions and larger step sizes in low-resolution regions, enabling adaptive spatial registration.

[0040] After completing the spatial grid division, a temporal consistency discriminant was designed for remote sensing data acquired at different times. For the same grid cell g∈G...k If there is no corresponding observation at a certain time point, feature interpolation is required. Let... This indicates that unit g is at time t. i The observation characteristics, then for the missing time t j The discriminator introduces a time-weighted interpolation function F. interp ,like:

[0041]

[0042] Where the weight w i,j =exp(-α|t j -t i |), where α is the temporal decay coefficient, thereby achieving smooth interpolation of temporally adjacent observations and ensuring that each spatial unit has consistent characteristics at each time point.

[0043] After fusing features from multiple sources and time phases, a feature embedding spatial mapping method is employed to eliminate inter-source noise and redundancy. Specifically, for each spatial unit g, a multi-source feature set is used... By designing a nonlinear embedding function Φ(·), the original features are projected onto a unified embedding space:

[0044] z g =Φ(X g )=σ(W·concat(X g )+b)

[0045] Where W and b are trainable parameters, σ(·) is the activation function (ReLU), and concat(X) g () indicates that multi-source features are spliced ​​by channel. This mapping process not only achieves feature normalization, but also optimizes parameters through the training process, so that the data from each source can highlight information related to forest changes in the embedding space, reduce irrelevant noise and redundant interference, and achieve efficient spatiotemporal fusion and unified expression of cross-source remote sensing data, providing standardized and high-quality input for subsequent forest resource status identification and dynamic monitoring.

[0046] Step 2: Intelligent Detection of Forest Resource Changes

[0047] Based on time-series data, a residual discriminative convolutional network is constructed to automatically learn the spatial and structural characteristics of forest change. A sparse attention mechanism is introduced to focus on suspected change areas and automatically ignore interference from clouds, fog, and image noise. Combining pixel-level and object-level information, a multi-label voting decision-making method is used to distinguish different change types such as forest growth, degradation, logging, and fire.

[0048] In realizing intelligent detection of forest resource changes, this invention proposes a "discriminative forest structure change sensing network," whose core innovations include a residual discriminative convolutional network, a sparse attention mechanism, and a multi-label voting discrimination module. The specific implementation process is as follows:

[0049] First, for the fused and aligned multi-source remote sensing feature sequences, let the feature representation of a spatial unit g at N consecutive time points be: The sequence is input into a Residual Discriminative Convolutional Network (RDCN). The network front-end consists of multiple stacked one-dimensional or two-dimensional convolutional layers (depending on the data structure), where each convolutional layer is C. l Its output is in The input is taken as the concatenated temporal features, and σ(·) is the nonlinear activation function. The core innovation lies in introducing residual connections, which directly add the output of the (l-1)th layer to the convolution result of the lth layer, i.e.:

[0050]

[0051] This residual structure helps capture small changes in forest structure over time and space, while suppressing the gradient vanishing problem in deep networks.

[0052] In the convolutional feature extraction process, a sparse attention mechanism is introduced to effectively focus on potentially changing regions. Specifically, for the convolutional output... ( L Apply sparse attention weights to the last layer. Let the attention score for each unit be...

[0053]

[0054] Where q g and k g These are the query and key vectors for cell g, respectively, where d is the dimension normalization coefficient, and m is the value of m. g This is a mask generated by the independent noise discrimination branch, used to block out cloud / fog interference or anomalous noise regions. The final sparse attention-weighted output is...

[0055]

[0056] In this way, the network can adaptively focus on locations where structural changes are suspected, automatically ignoring unusable pixels or interference areas, thus enhancing its ability to detect forest changes.

[0057] To finely differentiate between change types, the algorithm incorporates a multi-label voting decision module. Specifically, it first obtains the change probability vector for each unit at the pixel level. Where K represents the number of change types, such as forest growth, degradation, logging, and fire. Subsequently, pixels spatially belonging to the same object (e.g., a forest compartment or subcompartment) are aggregated, and the object-level discrimination result is obtained using the following weighted voting method:

[0058]

[0059] Where O represents a spatial object, w g The confidence weights for each pixel are assigned. Finally, the main label is selected by setting a threshold or sorting to achieve fine recognition of different types of changes, while retaining multi-label information to support complex or mixed boundary situations.

[0060] Discriminative forest structure change sensing networks, through the organic combination of residual learning, attention focusing, and object-level multi-label discrimination, can achieve high-precision and highly interpretable automatic detection and classification of dynamic changes in forest resources, providing a solid intelligent analysis foundation for dynamic monitoring systems of forest resources.

[0061] Step 3: Cross-source dynamic attribute inversion and rule adaptive correction

[0062] By leveraging multi-source remote sensing feature fusion, an inversion network for forest stock volume, tree height, and age class based on a deep generative model was designed. To address the potential model failure in specific regions, a data-driven rule self-learning module was designed to automatically capture abnormal patterns and optimize inference rules. The model supports the automatic incorporation of novel remote sensing data sources into the inversion model without manual parameter tuning.

[0063] A "Dynamic Forest Parameter Inversion-Rule Coupled Inference Model" is proposed, aiming to accurately and efficiently invert key attributes such as forest stock volume, tree height, and age class by fusing multi-source remote sensing features, while simultaneously achieving automatic rule correction for anomaly areas and seamless adaptation to new data sources. The specific implementation process is as follows:

[0064] First, the embedded features z, which have undergone the aforementioned multi-source spatiotemporal fusion and change detection, are... g The input is fed into a deep generative inversion network. This network essentially employs a Conditional Variational Autoencoder (CVAE) structure; for each spatial unit g, the network uses its fused features z... g As conditional inputs, this generates the posterior probability distribution of forest attributes. The forest volume (denoted as v) is used as the input. g Tree height (h) g ) and age class (a g For example, the model is trained using the following loss function:

[0065]

[0066] Where y g =(v g ,h g ,a g () represents the actual observed attributes. The attribute q generated for the network φ and p θ These are the posterior and prior distributions, D. KL This represents the Kullback–Leibler divergence. By optimizing the above loss, the network can adaptively learn the complex mapping relationship between multi-source features and forest dynamic attributes, achieving synergistic inversion of multiple indicators.

[0067] To ensure the model's generalization ability in special regions (such as extreme terrain and areas with strong human interference), the algorithm incorporates a data-driven rule-based self-learning module. Specifically, it first processes the model's prediction residuals... Spatial statistics and anomaly detection are performed. If a systematic deviation is found within a certain region R, the rule optimization process is activated. This process generates a new corrected rule R(z) by combining Bayesian inference with a decision tree self-learning algorithm. g ), in the form of:

[0068]

[0069] Where f corr (z g The error correction function is automatically learned and can optimize parameters in real time according to abnormal patterns, thereby significantly improving the applicability of the model in special situations.

[0070] Furthermore, to achieve seamless adaptation to new remote sensing data sources, the model integrates a meta-learning scheme within the feature embedding layer. When a new data source (such as a new type of optical satellite) is added, the model automatically updates the feature mapping parameter Φ′(·) and the generation network weights through a small-sample rapid adaptation mechanism, as shown in the following formula:

[0071]

[0072] Where η is the meta-learning rate, L new This is a mini-batch loss mechanism based on new data sources. This mechanism allows for the introduction of new remote sensing features without manual parameter tuning, enabling the model to continuously evolve and maintain its adaptive capabilities.

[0073] By coupling deep generative models with automated rule optimization, the problems of accuracy, robustness and scalability of dynamic inversion of attributes in multi-source data forests are systematically solved, providing an intelligent and flexible technical foundation for large-scale dynamic monitoring.

[0074] Step 4: Automatic Updates and Version Management of the Forest Resource Database

[0075] By utilizing the change detection results from different time phases, a change differential profile is automatically generated, and only newly added or changed information is written to the database, improving update efficiency. Before writing to the database, a data consistency verification module is enabled to resolve potential conflicts introduced by multi-source data. A multi-version time-series index for forest resource spatial objects is designed to enable historical data traceability, analysis, and rapid backtracking. In the automatic update and version management stage of the forest resource database, this invention adopts an "intelligent differential knowledge writing and multi-version consistency management method" to achieve efficient and reliable storage and historical traceability of forest resource monitoring results. The specific implementation is as follows:

[0076] First, the system is based on two consecutive time phases t. pre and t cur Based on the forest resource change monitoring results, for each spatial object O (such as forest compartment, sub-compartment, etc.), its attribute feature vectors are extracted as follows: and By calculating the difference in changes of object attributes The system automatically determines the change type and data validity based on the significance of each indicator in the difference vector (e.g., whether it exceeds a minimum threshold). Differential archives are generated only for objects that have been actually updated. The differential data structure is as follows: The tags contain additional knowledge such as the type of change and the source of the change, which allows only necessary information to be written, greatly improving the efficiency of database updates and storage utilization.

[0077] Following this, before the data is officially written to the database, the system activates the data consistency verification module. Its core is a multi-source conflict detection algorithm, which checks the set of attribute values ​​generated by different sources (such as optical, radar, and lidar inversion results) for each changed object O. Using a weighted consistency scoring function:

[0078]

[0079] in Let S be the source pair confidence level, δ be the consistency decision threshold, C be the total number of source pair combinations, and I(·) be the indicator function. If S O If the data falls below the preset threshold, an exception handling process is triggered, including manual review and prompts or attribute correction based on a regression model, thereby ensuring the reliability and consistency of the data entering the database.

[0080] Finally, to support spatiotemporal evolution analysis and rapid backtracking of forest resources, the system designs a multi-version time-series index for each spatial object O. Let object O be in the time sequence {t1, t2, ..., t...} n The set of all version features of} is Generate a unique version number for each update. The database uses a versioned key-value structure. Simultaneously build the index table:

[0081] Among them, tag i Record the type and source information of this change. This enables efficient source tracing, version comparison, and rollback operations of the object's historical status, making the evolution trajectory of forest resources clear and reproducible, supporting scientific management and post-event source tracing.

[0082] Through differential knowledge writing, automatic consistency verification, and multi-version time-series indexing, the system achieves high efficiency in automatic database updates and intelligent data lifecycle management, providing a solid information foundation and scalable support for a large-scale dynamic monitoring platform for forest resources.

[0083] Step 5: Anomaly Alarms and Decision Support

[0084] Based on change monitoring and automatic database updates, an "intelligent alarm and auxiliary intervention suggestion module" is constructed: abnormal changes are automatically classified (such as sudden large-scale degradation, illegal logging, etc.), and linked with external climate, fire, and other data to trigger alarms from relevant departments. Utilizing the confidence intervals output by the model, various intervention suggestions are generated, achieving a closed loop from monitoring to management.

[0085] First, the system automatically aggregates the latest real-time forest resource change archives and historical data, continuously tracking the dynamic status of each spatial object. When phenomena such as large-scale forest degradation or suspected illegal logging are detected, the module automatically classifies the anomalies based on multi-dimensional characteristics such as change type, affected area, and impact level. For example, for sudden events where the change area exceeds a set threshold, the system marks them as high priority; while sporadic, slow, normal periodic changes are classified as low priority, achieving intelligent screening and urgency ranking of anomalies.

[0086] To enhance the scientific rigor and timeliness of alerts, the system also integrates in real-time with external data sources such as meteorology, fire monitoring, and pest and disease data. Based on anomalies in the external environment (such as recent high temperatures, drought, or increased forest fire risk), the system dynamically adjusts thresholds and grading standards to accurately determine the causes of changes and improve its sensitivity to warnings of extreme events (such as natural disasters coupled with cross-risks like illegal activities). When trigger conditions are met, the module automatically pushes tiered alert information to relevant departments (such as forestry authorities and emergency response teams), including rich details such as the type of change, scope of impact, suspected causes, spatial location, and historical evolution trajectory. It also supports access to multiple notification channels (such as SMS, email, and platform push notifications) to ensure efficient transmission of critical information.

[0087] Building upon anomaly identification and alerting, the system also provides various auxiliary intervention suggestions for management decisions based on the model's output confidence intervals and multi-objective results. For each anomaly, the module automatically generates a list of intervention plans by comprehensively considering confidence level, trends, historical management measures, and external environmental factors. For example, for areas suspected of being illegally logged with a high confidence level, it recommends promptly increasing patrol forces or deploying monitoring cameras; for large-scale natural degradation accompanied by extreme weather, the system suggests implementing emergency replanting, strengthening water supply security, or initiating post-disaster recovery legal procedures. All intervention suggestions include their basis and priority, facilitating managers to quickly formulate response measures based on the actual situation.

[0088] Through the above-mentioned intelligent alarm and auxiliary decision-making design throughout the entire process, a complete link from automatic perception, scientific classification, cross-departmental collaboration to management closed loop has been realized, providing efficient and intelligent technical support for forest resource protection, risk control and emergency response.

[0089] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0090] It should be understood that the above detailed description of the technical solutions of the present invention with reference to preferred embodiments is illustrative and not restrictive. Those skilled in the art can modify the technical solutions described in the embodiments or make equivalent substitutions for some of the technical features based on reading this specification; however, these modifications or substitutions do not cause the essence of the corresponding technical solutions to depart from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for dynamic monitoring of forest resources and automatic database updating based on multi-source remote sensing data, characterized in that: The method includes: Heterogeneous data source fusion and spatiotemporal alignment are used to acquire multi-source remote sensing images. Based on the spatiotemporal labels of any two input images, a multi-scale spatial grid is dynamically generated. The minimum common area of ​​spatial coverage between the two images is calculated, and a master reference resolution is selected based on the resolution difference to generate a scale set. After completing the spatial grid division, a temporal consistency discriminator is designed for remote sensing data acquired at different times. For the same grid cell, if there is no corresponding observation value at a certain time point, feature interpolation is required. The discriminator introduces an interpolation function based on temporal weighting. After fusing features from multiple sources and multiple time phases, a feature embedding spatial mapping method is adopted. Intelligent forest resource change detection takes a multi-source remote sensing feature sequence that has been fused and aligned, and inputs the sequence into a residual discriminative convolutional network. The front end of this network consists of multiple stacked one-dimensional or two-dimensional convolutional layers. Residual connections are introduced to directly add the output of the first layer to the convolution result of the second layer. During the convolutional feature extraction process, a sparse attention mechanism is introduced to apply sparse attention weights to the convolutional output. For detailed differentiation of change types, a multi-label voting decision is designed. Specifically, the change probability vector of each unit is first obtained at the pixel level. Cross-source dynamic attribute inversion and rule adaptive correction: The embedded features, which have undergone the aforementioned multi-source spatiotemporal fusion and change detection, are input into the forest volume, tree height, and age class index inversion network based on a deep generative model; spatial statistics and anomaly detection are performed on the model prediction residuals; if a systematic deviation is found in a certain region, the rule optimization process is activated; new correction rules are generated by combining Bayesian inference and decision tree self-learning algorithms. The forest resource database features automatic updates and version management. Utilizing change detection results from previous and subsequent time phases, it automatically generates change difference profiles, writing only newly added or changed information to the database, thus improving update efficiency. Based on the combined forest resource change detection results from two different time phases, it extracts the attribute feature vector for each spatial object. By calculating the change difference of object attributes, it automatically determines the change type and data validity. Based on the significance of each indicator in the difference vector, it generates difference profiles only for objects with actual updates. Before formally writing to the database, the system activates a data consistency verification module, whose core is a multi-source conflict detection algorithm. For each changed object, it checks the attribute value sets generated by different sources, using a weighted consistency scoring function. If the values ​​do not meet the preset threshold, an exception handling process is triggered, including manual review prompts or attribute correction based on a regression model, thereby ensuring the reliability and consistency of the data entering the database. To support spatiotemporal evolution analysis and rapid backtracking of forest resources, a multi-version time-series index is designed for each spatial object. This index sets all version feature sets of the object in the time series and generates a unique version number for each update. The database uses a versioned key-value structure and simultaneously constructs an index table. Anomaly alerts and decision support generate various intervention suggestions, achieving a closed loop from monitoring to management.

2. The method for dynamic monitoring of forest resources and automatic database update based on multi-source remote sensing data according to claim 1, characterized in that: The aforementioned abnormal alarm and auxiliary decision-making includes: constructing an "intelligent alarm and auxiliary intervention suggestion module": abnormal changes are automatically classified and linked with external climate and fire data to trigger alarms from relevant departments. Using the output confidence interval, various intervention suggestions are generated to achieve a closed loop from monitoring to management.

Images