Digital forestry big data management method

Abstract

The application relates to the technical field of forestry data management, and discloses a digital forestry big data management method, which comprises the following steps: cleaning and standardizing historical multi-source forestry data and event records of a target region; spatiotemporally aligning the two, and constructing a supervised learning data set with event type labels; adopting an association rule mining method to extract frequent feature combinations and the association between the feature combinations and event types from the data set, form formal mapping rules, and construct a queryable association relationship database; after the same processing is performed on real-time multi-source forestry data flowing in, the mapping rules in the association relationship database are quickly matched, corresponding event types, confidence and other real-time early warning information are automatically generated, and end-to-end intelligent mapping from multi-source forestry data to event identification is completed; and the application realizes precise event-driven forestry big data management in different forestry scenes.

Description

Technical Field

[0001] This invention relates to the field of forestry data management technology, and more specifically to a digital forestry big data management method. Background Technology

[0002] Current management of forestry big data (covering forestry data, garden plant communities, and urban green space system data) focuses primarily on data storage, retrieval, and simple statistics, lacking the ability to deeply mine and analyze the forestry events behind the data. Existing technologies often rely on human experience or single data thresholds for event judgment, which has the following drawbacks: First, the value of historical data was not fully utilized, making it impossible to improve the accuracy of event judgments through historical correlation patterns. Second, the lack of standardized storage and management of the relationship between events and data makes it difficult to reuse the judgment logic of forestry events in different forestry scenarios (such as forestry scenarios, garden plant community scenarios, and urban green space system scenarios). Third, the mapping relationship between multi-source data and events is vague, making it impossible to effectively integrate multi-dimensional data to achieve accurate identification of complex events. This severely restricts the ability to detect and handle forestry events early in different forestry scenarios, making it difficult to meet the needs of precise and intelligent forestry management. Summary of the Invention

[0003] To address the aforementioned shortcomings in existing technologies, this invention provides a digital forestry big data management method. By constructing a database of correlations between feature data and events based on historical data mining, it achieves end-to-end intelligent mapping from real-time multi-source forestry data input to event type output. This solves problems in existing technologies such as the difficulty in quantifying and applying historical experience, the shallow mining of correlations between multi-source data and events, and the low efficiency of real-time event identification, ultimately achieving precise event-driven forestry big data management.

[0004] To achieve the above-mentioned objectives, the technical solution adopted by this invention is as follows: A digital forestry big data management method includes the following steps: Acquire multi-source forestry raw data of the target area within a preset historical period, perform spatiotemporal benchmark unification, missing value filling, outlier removal, and format unification to generate a cleaned historical multi-source forestry dataset; Based on the cleaned historical multi-source forestry dataset, standardization processing is performed to generate a standardized feature dataset. Obtain forestry event records for the target area within a preset historical period, extract the event type, occurrence time, and occurrence location for each event, and generate a structured list of event records; After aligning the structured list of event records with the standardized feature dataset in time and space, feature data extraction and event type labeling are performed to generate a supervised learning dataset with event type labels. Based on a supervised learning dataset with event type labels, an association rule mining method is used to obtain frequently occurring feature data and event type combinations, generate candidate association rules, and generate a set of formalized mapping rules through rule filtering and formalization. Based on formalized mapping rules, a database of the association between feature data and events is constructed; The current multi-source forestry data of the target area is obtained, and after spatiotemporal benchmark unification, missing value imputation, outlier removal, format unification and standardization processing, a standardized current feature dataset is generated. Using the standardized current feature dataset as query input, the database of associations between feature data and events is invoked. By executing mapping rule matching, real-time event information corresponding to the current multi-source forestry data is generated.

[0005] The present invention has the following beneficial effects: 1. The digital forestry big data management method proposed in this invention systematically mines the correlation between historical multi-source forestry data and events, combines standardized management mapping knowledge, and deeply integrates multi-source forestry data to achieve precise event-driven forestry big data management in different forestry scenarios. It also realizes a paradigm shift from experience-driven to data and knowledge-driven approaches, improving the accuracy and foresight of event identification. By systematically mining the strong correlation rules between multi-source feature combinations and forestry events in historical multi-source forestry data, implicit and vague human experience is transformed into explicit and quantifiable executable knowledge, i.e., mapping rules. This overcomes the false alarms and missed alarms caused by traditional methods relying on single thresholds or subjective experience, making event judgments (such as fire and pest warnings) more reliable, more accurate, and possessing a certain predictive and early warning capability. 2. This invention also realizes the systematic management and efficient application of event judgment logic for forestry big data; it stores and manages the massive amount of association rules mined in a structured manner in the form of graph databases, forming a queryable, scalable, and interpretable forestry event knowledge graph, which not only realizes the management of forestry big data, but also realizes the prediction of events from forestry big data. 3. This invention realizes precise and intelligent forestry big data management; when new multi-source forestry data flows in, it automatically executes a standardized process consistent with historical forestry data, and performs rapid matching with the mapping rules in the database in real time, directly outputting possible event types, confidence levels and location information. This transforms the traditional inefficient mode that relies on manual inspection and post-event reporting into an all-weather, automated, near real-time intelligent monitoring and early warning mode, providing key technical support for early detection and early handling of fire prevention, pest and disease control and other work, and meeting the core needs of precise and intelligent forestry management. 4. This invention also proposes an innovative architecture that combines the construction of a database through historical multi-source forestry data mining with the application of a real-time forestry data matching database. This systematically solves the three core problems in forestry event identification: difficulty in quantifying experience, inability to reuse knowledge, and serious lag in response. It realizes the intelligent upgrade of forestry big data management from passive storage and statistics to proactive event-driven management. Attached Figure Description

[0006] Figure 1 This is a flowchart illustrating a digital forestry big data management method proposed in this invention. Detailed Implementation

[0007] The specific embodiments of the present invention are described below to enable those skilled in the art to understand the present invention. However, it should be understood that the present invention is not limited to the scope of the specific embodiments. For those skilled in the art, various changes are obvious as long as they are within the spirit and scope of the present invention as defined and determined by the appended claims. All inventions utilizing the concept of the present invention are protected.

[0008] like Figure 1 As shown, a digital forestry big data management method includes the following steps: Obtain multi-source forestry raw data of the target area within a preset historical period, and perform spatiotemporal benchmark unification, missing value filling, outlier removal, and format unification to generate a cleaned historical multi-source forestry dataset.

[0009] Specifically, multi-source forestry raw data includes meteorological data, remote sensing image data, ground monitoring data, and human activity data.

[0010] In this embodiment, the process of historical multi-source data collection and cleaning is as follows: 1. Set input data: The target area includes multi-source forestry raw data from a preset historical period, including meteorological data (such as temperature, precipitation, and humidity), remote sensing imagery data (such as NDVI index and surface temperature), ground monitoring data (such as soil moisture and tree growth status in forestry scenarios; for garden plant community scenarios, soil moisture and growth status of garden trees or plants can be collected), and human activity data (such as logging and afforestation records in forestry scenarios; for garden plant community scenarios, afforestation records of garden trees can be collected). Data formats include CSV, NetCDF, GeoTIFF, and JSON. The target area refers to the target area under different forestry scenarios.

[0011] 2. Data cleaning: 1) Data quality check: Verify data integrity, temporal continuity, and spatial coverage; record missing and abnormal situations. 2) Spatiotemporal benchmark unification: Convert the timestamps of all multi-source forestry raw data to UTC standard time, format ISO 8601; project the geographic data in the multi-source forestry raw data to the WGS84 coordinate system; at the same time, resample the data with different spatial resolutions in the multi-source forestry raw data to a unified grid; finally generate historical multi-source forestry data with unified spatiotemporal benchmark. 3) Missing value imputation: Based on historical multi-source forestry data with unified spatiotemporal benchmarks, spatiotemporal interpolation is used. Linear interpolation or seasonal decomposition interpolation is used for missing values ​​in the time series, and kriging interpolation or inverse distance weighted interpolation is used for missing values ​​in the spatial series. Finally, historical multi-source forestry data with missing values ​​imputed is generated. 4) Outlier removal: Based on the historical multi-source forestry data after missing value imputation, the mean and standard deviation of each feature data in the historical multi-source forestry data are calculated using the 3σ principle based on statistical distribution. Data that exceed the mean ± 3 times the standard deviation are regarded as outliers and removed, and finally the historical multi-source forestry data with outliers removed are generated. 5) Unified Format: Based on historical multi-source forestry data with outliers removed, the data is converted into a unified data structure, where each row represents a spatiotemporal unit and each column represents a feature variable, generating a cleaned historical multi-source forestry dataset. The data format of this cleaned historical multi-source forestry dataset is a structured table or array, containing spatiotemporal representations (time, latitude and longitude) and feature data values. At the same time, a data quality report is output, recording detailed logs of the cleaning process.

[0012] In summary, this step addresses the data quality issues faced by multi-source heterogeneous forestry big data by acquiring and cleaning historical multi-source forestry raw data. Through a unified spatiotemporal benchmark, missing value processing, and outlier removal, raw data from different sources, formats, and qualities are integrated into a high-quality, spatiotemporally consistent, and structurally standardized dataset, providing a reliable and comparable data foundation for all subsequent analyses.

[0013] Based on the cleaned historical multi-source forestry dataset, standardization processing is performed to generate a standardized feature dataset.

[0014] In this embodiment, the standardization process for historical multi-source forestry datasets is as follows: 1. Set input data: Cleaned historical multi-source forestry dataset; 2. Standardize the input data: 1) Feature Data Extraction: Feature data was extracted from the cleaned historical multi-source dataset and divided into numerical feature data and categorical feature data. Numerical feature data refers to attributes that can quantitatively describe forestry-related objects or environments, have a continuous value range, and can be directly used for mathematical calculations. Examples include, in forestry scenarios, daily average temperature (°C), daily precipitation (mm), relative humidity (%), and wind speed (m / s) from meteorological data; normalized difference vegetation index (NDVI), surface temperature (K), vegetation cover (%), and leaf area index (LAI) from remote sensing image data; and soil moisture (%), tree diameter at breast height (cm), and tree height (m) from ground monitoring data. Examples of human activity data include monthly logging area (hectares) and number of afforested trees (trees) in a specific region. Similarly, categorical feature data refers to discrete category attributes used to describe forestry-related objects or events. These attributes have no continuous value range, can only be classified into a pre-defined finite set of categories, and cannot be directly used for mathematical operations (they need to be converted into numerical form through encoding before being analyzed). For example, in forestry scenarios, vegetation-related data includes tree species (pine, cypress, etc.) and vegetation community types (coniferous forest, broadleaf forest, etc.); soil-related data includes soil types (red soil, black soil, etc.) and soil texture (sandy soil, clay, etc.); and human activity-related data includes activity types (legal logging, illegal logging, etc.) and tool types (machetes, excavators, etc.).

[0015] 2) Normalize the numerical feature data, specifically by using the minimum and maximum normalization method. Based on the minimum and maximum values ​​of each numerical feature data, map each numerical feature data to the 0-1 interval to generate a normalized subset of numerical feature data. 3) Encode the categorical feature data. Specifically, for each categorical feature data, such as soil type and vegetation type, perform unique heat encoding, that is, convert it into numerical form to generate an encoded subset of categorical feature data. For example, in the forestry scenario, vegetation types include pine, cypress and birch. After unique heat encoding, the unique heat encoding for pine is [1, 0, 0]; the unique heat encoding for cypress is [0, 1, 0], etc.

[0016] 4) Finally, the normalized numerical feature data subset and the encoded categorical feature data subset are also converted into a unified data structure, that is, each row represents a spatiotemporal unit and each column represents a feature variable (i.e., normalized numerical feature data or encoded categorical feature data), generating a standardized feature dataset.

[0017] In summary, this step solves the problem that different feature data cannot be directly correlated due to differences in scale and type by generating a standardized feature dataset. Through normalization and one-hot encoding, all feature data are transformed to a unified numerical scale (0-1 interval) and structured format, enabling multi-dimensional feature data such as meteorological, remote sensing, and human activities to be processed and analyzed fairly and effectively by the algorithm.

[0018] Obtain forestry event records for the target area within a preset historical period, extract the event type, occurrence time, and occurrence location for each event, and generate a structured list of event records.

[0019] In this embodiment, the structured processing of historical event records is as follows: 1. Set input data: Forestry event records of the target area within a preset historical period, including forestry department reports, satellite hotspot monitoring, news reports, etc., usually in unstructured or semi-structured text or tables; 2. Perform structured processing on the input data: 1) Multi-source event fusion: Spatiotemporal matching and merging of the same event records from different sources to generate a unique event identifier (ID); 2) Extract event information: Use natural language processing technology to extract key information from unstructured or semi-structured text or tables, including: event type (such as fire, pest, and illegal logging events in forestry scenarios), time of occurrence (accurate to day or hour and converted to UTC standard time), and location of occurrence (latitude and longitude coordinates and converted to latitude and longitude coordinates in WGS84 coordinate system). 3) Structured organization: Organize each event into a structured record, including fields: event ID, event type, occurrence time, and occurrence location (longitude and latitude). Finally, a structured list of event records is generated in a table (such as CSV) or GeoJSON format, with each row corresponding to one event. Event statistics reports can also be output, including event type distribution, spatiotemporal distribution, etc.

[0020] In summary, this step transforms unstructured, scattered event reports (such as text records) into machine-readable structured data by generating a structured list of event records. This step enables the digital and standardized management of forestry events, making the events themselves data objects that can be accurately analyzed in spatiotemporal relation with feature datasets, which is a prerequisite for building supervised learning datasets.

[0021] After aligning the structured list of event records with the standardized feature dataset in both time and space, feature data extraction and event type labeling are performed to generate a supervised learning dataset with event type labels.

[0022] Specifically, the process of spatiotemporally aligning the structured list of event records with the standardized feature dataset, extracting feature data, and labeling event types to generate a supervised learning dataset with event type labels is as follows: Based on the time and location of each event, a time window and spatial range for the event are set.

[0023] In this embodiment, the time window for the event is set to 30 days before the event, and the spatial range for the event is set to a 5-kilometer buffer zone around the event point. Therefore, this step, by clearly defining the time window and spatial range, that is, combining the 30 days before the event with the 5-kilometer buffer zone, ensures that subsequent feature extraction is limited to both time and spatial range, thus avoiding features that are unrelated to the event.

[0024] The feature data of each event within the corresponding time window and spatial range are extracted from the standardized feature dataset to generate a positive sample feature matrix. Based on the positive sample feature matrix, the corresponding event type is labeled for each positive sample.

[0025] Simultaneously, random sampling is performed from spatiotemporal locations where no events occur to generate negative samples, which are then labeled as having no events.

[0026] Oversampling is performed on positive samples. When the ratio of positive samples to negative samples reaches a preset balance threshold, oversampling stops, and balanced positive and negative samples are generated.

[0027] In this embodiment, the preset balance threshold is 1:2, meaning the ratio of positive samples to negative samples is 1:2. The purpose of oversampling is to balance positive and negative samples while avoiding undersampling to prevent the loss of negative sample information.

[0028] The balanced positive and negative samples are converted into a fixed format, including sample identifiers, feature data vectors, event type labels, and spatiotemporal reference information, to obtain a supervised learning dataset with event type labels.

[0029] In summary, this step successfully established the correlation between historical events and corresponding spatiotemporal environmental features by generating a supervised learning dataset with event type labels. By extracting features centered on events (positive samples) and sampling from event-free areas (negative samples), and balancing the sample ratio, a high-quality labeled dataset was generated. This dataset can also be directly used for training machine learning models, thereby enabling the trained models to identify event information in real-time multi-source forestry data. Therefore, this step is a crucial one in transforming the business problem (event identification) into a data-solvable problem (classification / association analysis).

[0030] Based on a supervised learning dataset with event type labels, an association rule mining method is used to obtain frequently occurring feature data and event type combinations to generate candidate association rules. Through rule filtering and formalization, a set of formalized mapping rules is generated.

[0031] Specifically, based on a supervised learning dataset with event type labels, an association rule mining method is used to obtain frequently occurring feature data and event type combinations, generate candidate association rules, and generate a set of formalized mapping rules through rule filtering and formalization. Set the minimum support threshold and the minimum confidence threshold.

[0032] Continuous feature data in the supervised learning dataset with event type labels are divided into equal-frequency or equal-width bins. N Each discrete interval is assigned a discrete category label to generate discrete feature data.

[0033] In this embodiment, continuous feature data is generally divided into 5 to 10 discrete intervals, i.e., N∈[5,10]; for example, based on the equal-width binning method, temperature feature data can be divided into [0℃~10℃), [10℃~20℃), ..., [40℃~50℃].

[0034] Discretized feature data is combined with event types to form a transaction set.

[0035] Mine all frequent itemsets with support no less than the minimum support threshold from the transaction set, and filter out frequent itemsets containing event type labels.

[0036] Based on frequent itemsets containing event types, candidate association rules are generated. The antecedent of the candidate association rule is discretized feature data, and the consequent is the event type. It can be represented as discretized feature data → event type.

[0037] Calculate the confidence level of each candidate association rule to filter out candidate association rules that are not lower than the minimum confidence threshold, and use them as valid mapping rules.

[0038] The effective mapping rules are converted into a standard structural form, generating formalized mapping rules, namely:

[0039] in, Represents formalized mapping rules, This represents a unique identifier for the mapping rule, serving as the number of the selected candidate association rules. It is the unique number for each mapping rule, used to distinguish the association between different combinations of feature data and event types. This represents the set of discretized feature data used in the determination, and the discretized feature data includes one or more of the following: meteorological discrete features, remote sensing discrete features, ground monitoring discrete features, and human activity discrete features. Represents a set The set of judgment conditions corresponding to the feature data in the data, such as in a forestry scenario, when the set Given {high rainfall, loose tree species, high NDVI coverage}, the set of criteria is as follows: Let {Rainfall (standardized) ∈ [0.7, 1] (corresponding to the discrete category "Rainfall_High"), tree species number = pine (unique thermal code is [1, 0, 0]), NDVI ∈ [0.7, 1] (corresponding to the discrete category "NDVI_High Coverage")}, An identifier indicating the type of associated event, such as original event type = no event. Text identification or numerical encoding can be used, that is =No event or If the original event type is forest fire, then =Fire incident (text label) or (Digital coding, and a unified mapping rule needs to be established in advance for digital coding. For example, in the forestry scenario, 0 represents no event, 1 represents a fire event, 2 represents a pest and disease event, and 3 represents an illegal logging event, etc.) Indicates the support of the mapping rule. This indicates the confidence level of the mapping rule.

[0040] In summary, this step automatically mines strong association rules between feature data patterns and event types from historical experience (supervised learning datasets) by generating a set of formalized mapping rules; by using association rule mining (Apriori algorithm), it can discover interpretable knowledge that when certain feature combinations (such as high temperature, drought, specific vegetation) occur frequently, a certain type of event (such as fire) also occurs frequently; by setting support and confidence thresholds, reliable and stable rules can be selected and formalized into a standard structure, realizing the transformation from data to knowledge, replacing the traditional crude judgment method that relies on human experience to set a single threshold.

[0041] The process of constructing a database of associations between feature data and events based on formal mapping rules is as follows: The formalized mapping rules are used as mapping rule nodes to extract the set of formalized mapping rules. The discretized feature data is used as discretized feature data nodes to extract the identifiers from the formalized mapping rules. The event type is used as an event type node.

[0042] Configure structured attributes for mapping rule nodes, discretized feature data nodes, and event type nodes, specifically: Discretized feature data nodes contain a unique identifier and a feature name.

[0043] The event type node contains a unique identifier and an event type name.

[0044] The mapping rule node contains a unique identifier, support, and confidence.

[0045] Connect the mapping rule nodes with the discretized feature data nodes to generate edges that associate rules with features; connect the mapping rule nodes with the event type nodes to generate edges that associate rules with events; and finally obtain a database of the association between feature data and events.

[0046] In summary, this step systematically manages and stores the mined, discrete mapping rules by constructing a database of relationships between feature data and events. By adopting a graph database structure, rules, features, and events are treated as nodes, and their relationships are treated as edges, forming a visible, queryable, and scalable knowledge graph. This solves the problem of rules being difficult to reuse and manage, and provides an efficient and structured query foundation for real-time rule matching.

[0047] The process of using the standardized current feature dataset as query input, calling the database of associations between feature data and events, and generating real-time event information corresponding to the current multi-source forestry data by performing mapping rule matching is as follows: The continuous feature data in the standardized current feature dataset are divided into groups using either equal-frequency or equal-width binning methods. N Each discrete interval is assigned a discrete category label, generating a discretized current feature dataset.

[0048] Call the database of relationships between feature data and events, and load the mapping rule nodes, discretized feature data nodes, event type nodes, edges of rules associated with features, and edges of rules associated with events.

[0049] Match the discretized current feature dataset with the set of mapping rules in the feature data and event association database. Filter out the set of discretized current feature datasets Mapping rules for all feature items, generating mapping rules for successful matches.

[0050] Extract the event type nodes associated with the successfully matched mapping rules, integrate the support and confidence of the mapping rule with the spatiotemporal information of the current multi-source forestry data, and generate real-time event information corresponding to the current multi-source forestry data. The real-time event information includes a unique event representation, event type name, confidence level, spatiotemporal occurrence (UTC time, WGS84 coordinate system latitude and longitude), and a unique identifier (ID) of the successfully matched mapping rule.

[0051] In summary, this step extracts real-time event information corresponding to current multi-source forestry data. By generating a standardized current feature dataset, it performs preprocessing and standardization on real-time incoming forestry data in a manner completely consistent with historical forestry data. This ensures that real-time forestry data and the historical data upon which the constructed knowledge base relies are in the same feature space and data standard, guaranteeing the fairness and accuracy of subsequent rule matching and avoiding misjudgments caused by inconsistent data processing. Simultaneously, it enables real-time, automatic identification and early warning of forestry events. By rapidly matching the real-time processed feature data with rules in the relational database, it can immediately output the possible event types and their confidence levels. This significantly improves the timeliness and intelligence of event detection, transforming traditional post-event reporting into pre-event data-driven early warning, providing crucial decision support for precise forestry management and emergency response.

[0052] In addition, the relational database can be dynamically updated. For example, every preset period (e.g., 3 months), new historical multi-source forestry data and new forestry event records can be collected, the above operations can be repeated, new mapping rules can be mined and added to the relational database, and outdated rules with support below the minimum support threshold can be deleted to achieve iterative optimization of the relational database.

[0053] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0054] Those skilled in the art will recognize that the embodiments described herein are intended to help the reader understand the principles of the invention, and should be understood that the scope of protection of the invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific modifications and combinations based on the technical teachings disclosed in this invention without departing from the spirit of the invention, and these modifications and combinations are still within the scope of protection of this invention.

Claims

1. A digital forestry big data management method, characterized in that, Includes the following steps: Acquire multi-source forestry raw data of the target area within a preset historical period, perform spatiotemporal benchmark unification, missing value filling, outlier removal, and format unification to generate a cleaned historical multi-source forestry dataset; Based on the cleaned historical multi-source forestry dataset, standardization processing is performed to generate a standardized feature dataset. Obtain forestry event records for the target area within a preset historical period, extract the event type, occurrence time, and occurrence location for each event, and generate a structured list of event records; After spatiotemporally aligning the structured list of event records with the standardized feature dataset, feature data extraction and event type labeling are performed to generate a supervised learning dataset with event type labels, specifically: Based on the time and location of each event, a time window and spatial range for the event are set. Extract feature data for each event within the corresponding time window and spatial range from the standardized feature dataset, generate a positive sample feature matrix, and label each positive sample with the corresponding event type based on the positive sample feature matrix; Simultaneously, random sampling is performed from spatiotemporal locations where no events have occurred to generate negative samples, which are then labeled as having no events. Oversampling is performed on positive samples. When the ratio of positive samples to negative samples reaches a preset balance threshold, oversampling is stopped, and balanced positive and negative samples are generated. The balanced positive and negative samples are converted into a fixed format, including sample identifiers, feature data vectors, event type labels, and spatiotemporal reference information, to obtain a supervised learning dataset with event type labels. Based on a supervised learning dataset with event type labels, an association rule mining method is used to obtain frequently occurring feature data and event type combinations, generate candidate association rules, and generate a set of formalized mapping rules through rule filtering and formalization. Based on formalized mapping rules, a database of the association between feature data and events is constructed; The current multi-source forestry data of the target area is obtained, and after spatiotemporal benchmark unification, missing value imputation, outlier removal, format unification and standardization processing, a standardized current feature dataset is generated. Using the standardized current feature dataset as query input, the database of associations between feature data and events is invoked. By executing mapping rule matching, real-time event information corresponding to the current multi-source forestry data is generated.

2. The digital forestry big data management method according to claim 1, characterized in that, Multi-source forestry raw data includes meteorological data, remote sensing image data, ground monitoring data, and human activity data.

3. The digital forestry big data management method according to claim 2, characterized in that, The types of incidents include fire incidents, pest and disease incidents, and illegal logging incidents.

4. The digital forestry big data management method according to claim 3, characterized in that, Based on a supervised learning dataset with event type labels, an association rule mining method is used to obtain frequently occurring feature data and event type combinations, generate candidate association rules, and generate a set of formalized mapping rules through rule filtering and formalization. Set the minimum support threshold and the minimum confidence threshold; Continuous feature data in the supervised learning dataset with event type labels are divided into equal-frequency or equal-width bins. N Each discrete interval is assigned a discrete category label to generate discrete feature data. Discretized feature data and event types are combined into a transaction set; Mine all frequent itemsets with support no less than the minimum support threshold from the transaction set, and filter out frequent itemsets containing event type labels; Based on frequent itemsets containing event types, candidate association rules are generated, where the antecedent of the candidate association rule is discretized feature data and the consequent is the event type. Calculate the confidence score of each candidate association rule, and select candidate association rules that are not lower than the minimum confidence score threshold as valid mapping rules; The effective mapping rules are converted into a standard structural form, generating formalized mapping rules, namely: in, Represents formalized mapping rules, A unique identifier representing a mapping rule. This represents the set of discretized feature data that participate in the decision-making process. Represents a set The set of judgment conditions corresponding to the feature data in the data. An identifier indicating the associated event type. Indicates the support of the mapping rule. This indicates the confidence level of the mapping rule.

5. The digital forestry big data management method according to claim 4, characterized in that, The process of constructing a database of associations between feature data and events based on formal mapping rules is as follows: The formalized mapping rules are used as mapping rule nodes to extract the set of formalized mapping rules. The discretized feature data is used as discretized feature data nodes to extract the identifiers from the formalized mapping rules. The event type is used as an event type node; Configure structured attributes for mapping rule nodes, discretized feature data nodes, and event type nodes, specifically: Discretized feature data nodes contain a unique identifier and a feature name; The event type node contains a unique identifier and an event type name; Mapping rule nodes include a unique identifier, support, and confidence level; Connect the mapping rule nodes with the discretized feature data nodes to generate edges that associate rules with features; connect the mapping rule nodes with the event type nodes to generate edges that associate rules with events; and finally obtain a database of the association between feature data and events.

6. The digital forestry big data management method according to claim 5, characterized in that, The process of using the standardized current feature dataset as query input, calling the database of associations between feature data and events, and generating real-time event information corresponding to the current multi-source forestry data by performing mapping rule matching is as follows: The continuous feature data in the standardized current feature dataset are divided into groups using either equal-frequency or equal-width binning methods. N Each discrete interval is assigned a discrete category label to generate a discretized current feature dataset. Call the database of relationships between feature data and events, and load the mapping rule nodes, discretized feature data nodes, event type nodes, edges of rules associated with features, and edges of rules associated with events; Match the discretized current feature dataset with the set of mapping rules in the feature data and event association database. Filter out the set of discretized current feature datasets Mapping rules for all feature items, generating mapping rules for successful matches; Extract the event type nodes associated with the successfully matched mapping rules, integrate the support and confidence of the mapping rule with the spatiotemporal information of the current multi-source forestry data, and generate real-time event information corresponding to the current multi-source forestry data.

Images