A natural resource investigation intelligent monitoring and early warning method and system for land engineering

Abstract

The application discloses a kind of natural resource survey intelligent monitoring early warning method and system of land engineering, by constructing multi-source collaborative monitoring network, the natural resource multi-source monitoring data in land engineering area is collected;The natural resource multi-source monitoring data is sequentially denoised, space-time alignment and standardized, and standardization monitoring data is obtained;Dynamic early warning model is constructed, the high-order fusion feature is input into dynamic early warning model, and the evaluation index of natural resource risk level is output;According to evaluation index, graded early warning information is generated and pushed to disposal terminal, while the corresponding disposal process is started.The early warning accuracy is high, response is quick, effectively avoids early warning lag.Establishment graded early warning and closed-loop disposal mechanism, realize early warning information accurate push and differential treatment, substantially improve risk disposal efficiency.

Description

Technical Field

[0001] This invention relates to the field of natural resource monitoring technology, and in particular to an intelligent monitoring and early warning method and system for natural resource surveys in land engineering. Background Technology

[0002] Current land engineering natural resource monitoring relies heavily on traditional manual inspections and single monitoring methods, resulting in limited monitoring scope and difficulty in achieving full-dimensional coverage. Multi-source data lacks effective collaboration, leading to problems such as noise interference, spatiotemporal misalignment, and inconsistent dimensions, resulting in low data utilization. Feature extraction methods are simplistic and fail to fully explore the spatial, temporal, and target characteristics of the data. Furthermore, feature weight allocation depends on human experience, and early warning models are mostly static threshold-based, making it difficult to capture the dynamic evolution of risks, resulting in delayed early warnings and low accuracy. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an intelligent monitoring and early warning method and system for natural resource surveys in land engineering.

[0004] To achieve the above objectives, the technical solution of the present invention further includes the following steps in the above-mentioned intelligent monitoring and early warning method for natural resource surveys in land engineering: A multi-source collaborative monitoring network is constructed to collect multi-source monitoring data of natural resources within the land engineering area; the multi-source monitoring data of natural resources are then subjected to noise reduction, spatiotemporal alignment, and standardization to obtain standardized monitoring data. Preliminary features of standardized monitoring data were extracted using CNN convolutional neural network, statistical analysis methods and YOLOv8 target detection algorithm respectively. The importance weights of each feature were adaptively assigned based on the information gain ratio algorithm. The preliminary features were then aggregated and transformed by GCN graph convolutional network to obtain high-order fused features. A dynamic early warning model is constructed based on the Transformer-LSTM network. The high-order fusion features are input into the dynamic early warning model, and the assessment index of natural resource risk level is output. Based on the assessment index, a graded early warning information is generated and pushed to the handling terminal, and the corresponding handling process is initiated at the same time.

[0005] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the construction of a multi-source collaborative monitoring network to collect multi-source monitoring data of natural resources within the land engineering area; and the sequential denoising, spatiotemporal alignment, and standardization of the multi-source monitoring data of natural resources to obtain standardized monitoring data, including: Noise reduction was performed on UAV and satellite imagery data from multi-source natural resource monitoring data using a combination of Gaussian filtering and median filtering to eliminate salt-and-pepper noise and Gaussian noise while retaining effective feature information. Wavelet analysis was used to decompose the data signals and remove high-frequency noise components from the time-series soil and hydrological data collected by sensors to obtain noise-reduced monitoring data. The timestamps of all noise reduction monitoring data are calibrated, and data from different acquisition frequencies are uniformly interpolated to one record per hour; the location data of UAVs, satellite images and monitoring stations are transformed using the WGS84 coordinate system as the spatial reference, and the image data is matched with the location of the monitoring stations through georegistration technology to obtain aligned monitoring data; The min-max standardization method is used to map the aligned monitoring data of different dimensions and magnitudes to the [0,1] interval to obtain standardized monitoring data.

[0006] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the preliminary features of standardized monitoring data are extracted using CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 target detection algorithm, respectively. The importance weights of each feature are adaptively assigned based on the information gain ratio algorithm. The preliminary features are then aggregated and transformed using a GCN graph convolutional network to obtain higher-order fused features, including: Spatial features of UAV and satellite imagery spatial data in standardized monitoring data are extracted layer by layer using the convolutional and pooling layers of a CNN convolutional neural network, including surface texture, vegetation distribution outline and terrain undulation features, and output spatial feature vectors. Statistical indicators of soil moisture, groundwater level and soil nutrient data are calculated using statistical analysis methods, including mean, variance, range, trend slope and coefficient of variation, to obtain the numerical variation characteristics and stability characteristics of the data, and output numerical feature vectors. This paper uses the YOLOv8 target detection algorithm to detect abnormal targets in image data, extracts the location, area and shape features of the targets, and outputs the target feature vector.

[0007] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the preliminary features of standardized monitoring data are extracted using CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 target detection algorithm, respectively. The importance weights of each feature are adaptively assigned based on the information gain ratio algorithm. The preliminary features are then aggregated and transformed using a GCN graph convolutional network to obtain higher-order fused features, including: The information gain ratio algorithm is used to calculate the information gain ratio of each preliminary feature for natural resource risk assessment. Using spatial feature vectors, numerical feature vectors, and target feature vectors as inputs, and historical risk data as labels, the information gain of each feature is calculated, and then divided by the entropy value of the feature to obtain the information gain ratio. After normalization, the weight coefficients of each feature are obtained, and the sum of the weight coefficients is 1, thus obtaining the preliminary features with weight coefficients.

[0008] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the preliminary features of standardized monitoring data are extracted using CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 target detection algorithm, respectively. The importance weights of each feature are adaptively assigned based on the information gain ratio algorithm. The preliminary features are then aggregated and transformed using a GCN graph convolutional network to obtain higher-order fused features, including: Preliminary features with weighted coefficients are input into the GCN graph convolutional network. A feature association graph is constructed using monitoring stations as nodes and the spatial correlation of monitoring areas as edges. By using graph convolution operations to aggregate, transform, and fuse the initial features, the intrinsic relationships between different features are explored, and a high-order fused feature vector is output.

[0009] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the step of constructing a dynamic early warning model based on a Transformer-LSTM network, inputting the high-order fusion features into the dynamic early warning model, and outputting an assessment index for the natural resource risk level includes: The Transformer-LSTM network structure includes an input layer, an encoding layer, a decoding layer, and an output layer. The input layer receives high-order fused feature vectors and temporal information. The encoding layer consists of a Transformer encoder and an LSTM layer. The Transformer encoder captures long-distance correlations between different features through a self-attention mechanism, while the LSTM layer learns the temporal variation patterns of features and captures the dynamic evolution trend of natural resource risks. The decoding layer decodes the encoded features and outputs risk assessment-related parameters. The output layer uses a fully connected layer combined with a sigmoid activation function to output an assessment index for natural resource risk levels between 0 and 100.

[0010] Furthermore, in the aforementioned intelligent monitoring and early warning method for natural resource surveys in land engineering, the step of generating tiered early warning information based on the assessment index and pushing it to the processing terminal, while simultaneously initiating the corresponding processing procedure, includes: The assessment index generates tiered early warning information, which includes the warning level, warning area, abnormal indicators, risk description, and warning time.

[0011] Furthermore, in a land engineering natural resource survey intelligent monitoring and early warning system, the natural resource survey intelligent monitoring and early warning system includes the following modules: The monitoring data acquisition module is used to construct a multi-source collaborative monitoring network and collect multi-source monitoring data of natural resources within the land engineering area; the multi-source monitoring data of natural resources is sequentially subjected to noise reduction, spatiotemporal alignment and standardization to obtain standardized monitoring data; The feature extraction and fusion module is used to extract preliminary features from standardized monitoring data using CNN convolutional neural network, statistical analysis methods and YOLOv8 target detection algorithm respectively. It adaptively assigns importance weights to each feature based on information gain ratio algorithm, and aggregates and transforms the preliminary features through GCN graph convolutional network to obtain high-order fused features. The early warning model building module is used to construct a dynamic early warning model based on the Transformer-LSTM network. The high-order fusion features are input into the dynamic early warning model, and the assessment index of the natural resource risk level is output. The intelligent monitoring and early warning module is used to generate graded early warning information based on the assessment index and push it to the handling terminal, while simultaneously initiating the corresponding handling process.

[0012] Furthermore, in a land engineering natural resource survey intelligent monitoring and early warning system, the feature extraction and fusion module includes the following sub-modules: The extraction submodule is used to extract spatial features from the UAV and satellite imagery spatial data in the standardized monitoring data layer by layer using the convolutional and pooling layers of the CNN convolutional neural network, including surface texture, vegetation distribution outline and terrain undulation features, and output spatial feature vectors. The analysis submodule is used to calculate statistical indicators of soil moisture, groundwater level and soil nutrient numerical data through statistical analysis methods, including mean, variance, range, trend slope and coefficient of variation, to obtain the numerical change characteristics and stability characteristics of the data, and output numerical feature vectors. The detection submodule is used to detect abnormal targets in image data based on the YOLOv8 target detection algorithm, extract the location, area and shape features of the target, and output the target feature vector.

[0013] Furthermore, in a land engineering natural resource survey intelligent monitoring and early warning system, the feature extraction and fusion module includes the following sub-modules: The calculation submodule is used to calculate the information gain ratio of each preliminary feature to the natural resource risk assessment using the information gain ratio algorithm. The adjustment submodule is used to take spatial feature vectors, numerical feature vectors, and target feature vectors as inputs, use historical risk data as labels, calculate the information gain of each feature, divide it by the entropy value of the feature to obtain the information gain ratio, and obtain the weight coefficient of each feature through normalization. The sum of the weight coefficients is 1, resulting in the preliminary features with weight coefficients.

[0014] Its beneficial effects lie in solving the problems of narrow monitoring scope and single data in traditional monitoring by monitoring the entire area, all elements, and all weather conditions, ensuring the comprehensiveness and real-time nature of monitoring data. Through targeted noise reduction, spatiotemporal alignment, and standardized preprocessing, data quality is effectively improved, laying a solid foundation for subsequent feature extraction and early warning analysis, and reducing errors caused by data interference. Thirdly, multi-algorithm fusion is used to extract multi-dimensional preliminary features, combined with adaptive weight allocation based on information gain ratio, and efficient feature aggregation is achieved through GCN, solving the problems of incomplete feature extraction and subjective weight allocation, and improving feature representativeness. A dynamic early warning model is built based on Transformer-LSTM, taking into account both long-distance dependence and temporal change capture, resulting in high early warning accuracy and rapid response, effectively avoiding early warning lag. A hierarchical early warning and closed-loop handling mechanism is established to achieve precise push and differentiated handling of early warning information, significantly improving the efficiency of risk handling. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 This is a schematic diagram of the first embodiment of a land engineering natural resource survey intelligent monitoring and early warning method according to an embodiment of the present invention; Figure 2 This is a schematic diagram of a second embodiment of an intelligent monitoring and early warning method for natural resource surveys in land engineering, as described in this invention. Figure 3 This is a schematic diagram of the first embodiment of an intelligent monitoring and early warning system for natural resource surveys in land engineering, as described in this invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0018] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms "one," "an," and "this" used herein may also include the plural forms. It should be further understood that the terminology used in this specification includes the presence of features, integers, steps, operations, elements, and / or components, but does not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0019] The present invention will now be described in detail with reference to the accompanying drawings. Figure 1As shown, a smart monitoring and early warning method for natural resource surveys in land engineering includes the following steps: Step 101: Construct a multi-source collaborative monitoring network and collect multi-source monitoring data of natural resources within the land engineering area; perform noise reduction, spatiotemporal alignment and standardization on the multi-source monitoring data of natural resources in sequence to obtain standardized monitoring data; Specifically, in this embodiment, the UAV and satellite image data in the multi-source monitoring data of natural resources are denoised using a combination of Gaussian filtering and median filtering to eliminate salt-and-pepper noise and Gaussian noise in the images and retain effective feature information; wavelet analysis algorithm is used to decompose the data signal and remove high-frequency noise components from the time series data of soil and hydrology collected by the sensor to obtain denoised monitoring data. The timestamps of all noise reduction monitoring data are calibrated, and data from different acquisition frequencies are uniformly interpolated to one record per hour; the location data of UAVs, satellite images and monitoring stations are transformed using the WGS84 coordinate system as the spatial reference, and the image data is matched with the location of the monitoring stations through georegistration technology to obtain aligned monitoring data; The min-max standardization method is used to map the aligned monitoring data of different dimensions and magnitudes to the [0,1] interval to obtain standardized monitoring data.

[0020] Combining the topography, natural resource types (such as soil, hydrology, vegetation, geology, and construction scope) of the land engineering area, an integrated multi-source collaborative monitoring network (air-space-ground-underground) is constructed to achieve full-area, full-element, and all-weather monitoring coverage. Airborne monitoring utilizes UAVs equipped with hyperspectral cameras and thermal infrared sensors to collect visual and spectral data on vegetation cover, surface soil moisture, and surface disturbance within the area. Flight altitude is adjusted to 50-200 meters depending on the size of the monitoring area, with a flight frequency of 1-2 times per week. Space-based monitoring relies on high-resolution satellites, such as Gaofen-1 and Gaofen-6, to acquire macro-level data on land use types, vegetation indices, and water distribution at the regional scale, with data updates every 3-7 days. Ground-based monitoring involves a grid-based layout of monitoring stations within the engineering area. Equipment such as soil moisture sensors, groundwater level monitors, soil nutrient sensors, and geological settlement meters are deployed to collect real-time micro-data such as soil moisture content, groundwater level depth, soil nitrogen, phosphorus, and potassium content, and surface subsidence. Data collection is conducted every 1-2 hours, increasing to every 30 minutes during extreme weather events such as heavy rain or typhoons. Underground monitoring targets groundwater resources and underground soil and rock masses, employing borehole monitoring combined with ground-penetrating radar to collect data on underground soil and rock density, groundwater flow velocity, and direction. A comprehensive survey is conducted quarterly, with supplementary surveys monthly for key areas. All monitoring data is uniformly aggregated at the data acquisition terminal, synchronously recording metadata such as collection time, location, latitude and longitude, and monitoring equipment number to ensure data traceability.

[0021] Targeted algorithms are employed to address the noise characteristics of different data types. For UAV and satellite imagery data, a combination of Gaussian filtering and median filtering is used to eliminate salt-and-pepper noise and Gaussian noise while retaining effective feature information. For time-series data on soil and hydrology collected by sensors, wavelet analysis is used to decompose the data signal and remove high-frequency noise components, restoring the true trend of data changes. Ground-penetrating radar data undergoes denoising to eliminate clutter interference and highlight abnormal signals in underground soil and rock. After denoising, the data undergoes validity verification, and outliers are removed. Using the 3σ principle, data exceeding ±3 standard deviations of the mean are marked as outliers, and missing data are supplemented using linear interpolation. Using Beijing time as a unified time reference, the timestamps of all monitoring data are calibrated, and data from different acquisition frequencies are uniformly interpolated to one record per hour to ensure consistency in the time dimension. Using the WGS84 coordinate system as a spatial reference, coordinate transformation is performed on the location data of UAVs, satellite imagery, and monitoring stations. Georegistration technology is used to accurately match image data with monitoring station locations, ensuring a one-to-one spatial correspondence between different types of data in the same monitoring area and eliminating spatial offset errors. The min-max normalization method is used to map monitoring data of different dimensions and magnitudes to the [0,1] interval, thereby eliminating the impact of dimensional differences on subsequent feature extraction.

[0022] Step 102: Use CNN convolutional neural network, statistical analysis method and YOLOv8 target detection algorithm to extract preliminary features of standardized monitoring data respectively. Based on the information gain ratio algorithm, adaptively allocate the importance weight of each feature. Use GCN graph convolutional network to aggregate and transform the preliminary features to obtain high-order fused features. Specifically, in this embodiment, the convolutional and pooling layers of a CNN convolutional neural network are used to extract spatial features from the UAV and satellite imagery spatial data in the standardized monitoring data layer by layer. These features include surface texture, vegetation distribution outline, and terrain undulation features, and the output spatial feature vector is then generated. Statistical analysis methods are used to calculate statistical indicators of soil moisture, groundwater level, and soil nutrient numerical data, including mean, variance, range, trend slope, and coefficient of variation, to obtain the numerical variation and stability characteristics of the data, and the output numerical feature vector is then generated. Based on the YOLOv8 target detection algorithm, abnormal targets in the image data are detected, and the location, area, and morphological features of the targets are extracted, and the target feature vector is generated.

[0023] The information gain ratio algorithm is used to calculate the information gain ratio of each preliminary feature for natural resource risk assessment. Spatial feature vector, numerical feature vector and target feature vector are used as inputs, and historical risk data are used as labels to calculate the information gain of each feature. The information gain ratio is obtained by dividing by the entropy value of the feature. The weight coefficients of each feature are obtained through normalization. The sum of the weight coefficients is 1, and the preliminary features with weight coefficients are obtained.

[0024] The preliminary features with weighted coefficients are input into the GCN graph convolutional network. The monitoring stations are used as nodes and the spatial correlation of the monitoring area is used as edges to construct a feature correlation graph. The preliminary features are aggregated, transformed and fused through graph convolution operations to explore the intrinsic correlation between different features and output a high-order fused feature vector.

[0025] CNN (Convolutional Neural Network): For spatial data such as UAV and satellite imagery, it extracts spatial features layer by layer through convolutional and pooling layers, including surface texture, vegetation distribution outlines, and terrain undulation features, outputting spatial feature vectors. Statistical Analysis Methods: For time-series numerical data such as soil moisture, groundwater level, and soil nutrients, it calculates statistical indicators such as mean, variance, range, trend slope, and coefficient of variation to extract numerical variation and stability features, outputting numerical feature vectors. YOLOv8 Object Detection Algorithm: For abnormal targets in imagery data, such as surface subsidence, water pollution, areas of withered vegetation, and areas disturbed by illegal construction, it performs target detection, extracting features such as target location, area, and morphology, outputting target feature vectors. These three algorithms output preliminary feature vectors, forming a preliminary feature set. An information gain ratio (IGR) algorithm is employed to calculate the IRGR of each preliminary feature for natural resource risk assessment. A higher IRGR indicates a greater contribution of the feature to risk identification, and the importance weights of each preliminary feature are adaptively assigned accordingly. Specifically, the preliminary feature set is used as input, historical risk data is used as labels, the IRGR of each feature is calculated, and then divided by the entropy value of that feature to obtain the IRGR. Normalization is then applied to obtain the weight coefficients of each feature, with the sum of the weight coefficients being 1. This ensures the rationality and objectivity of the weight allocation and avoids biases caused by manually setting weights. The preliminary features with weighted coefficients are input into the GCN graph convolutional network. The monitoring stations are used as nodes and the spatial correlation of the monitoring area is used as edges to construct a feature correlation graph. The preliminary features are aggregated, transformed and fused through graph convolution operations to explore the intrinsic correlation between different features, such as the correlation between soil moisture and vegetation coverage, and the correlation between land subsidence and groundwater level. Feature redundancy is eliminated, effective features are strengthened, and finally, a high-order fused feature vector with unified dimensions and strong correlation is output.

[0026] Step 103: Construct a dynamic early warning model based on the Transformer-LSTM network, input high-order fusion features into the dynamic early warning model, and output the assessment index of natural resource risk level; Specifically, in this embodiment, the Transformer-LSTM network structure includes an input layer, an encoding layer, a decoding layer, and an output layer; The input layer receives high-order fused feature vectors and temporal information. The encoding layer consists of a Transformer encoder and an LSTM layer. The Transformer encoder captures long-distance correlations between different features through a self-attention mechanism, while the LSTM layer learns the temporal variation patterns of features and captures the dynamic evolution trend of natural resource risks. The decoding layer decodes the encoded features and outputs risk assessment-related parameters. The output layer uses a fully connected layer combined with a sigmoid activation function to output an assessment index for natural resource risk levels between 0 and 100.

[0027] The Transformer-LSTM network combines the long-range dependency capture capability of the Transformer with the temporal feature learning capability of the LSTM. The model structure consists of an input layer, an encoder layer, a decoder layer, and an output layer. The input layer receives high-order fused feature vectors and temporal information. The encoder layer comprises a Transformer encoder and an LSTM layer. The Transformer encoder captures long-range correlations between different features through a self-attention mechanism, while the LSTM layer learns the temporal variation patterns of features, capturing the dynamic evolution trend of natural resource risks. The decoder layer decodes the encoded features and outputs risk assessment parameters. The output layer uses a fully connected layer combined with a sigmoid activation function to output a natural resource risk level assessment index between 0 and 100; a higher index indicates a higher risk level.

[0028] A training dataset was constructed using historical monitoring data and historical risk event data, and divided into training, validation, and test sets in a 7:2:1 ratio. The loss function was the error between the predicted value of the evaluation index and the actual risk level, using the mean squared error loss function. The Adam optimizer was used to adjust model parameters, and the model was iteratively trained until the prediction accuracy reached over 90% and the loss function converged. Simultaneously, a dropout layer was introduced to prevent overfitting, and the model was periodically validated using a validation set. Based on the validation results, parameters such as the number of network layers and neurons were adjusted to optimize model performance.

[0029] The obtained high-order fusion feature vector is input into the trained dynamic early warning model. The model combines the temporal changes of real-time monitoring data to dynamically output the natural resource risk level assessment index, and at the same time outputs the confidence level of the assessment result. A confidence level of ≥85% is a valid assessment result, and a confidence level of less than 85% triggers the data re-collection and feature extraction process.

[0030] Step 104: Generate tiered early warning information based on the assessment index and push it to the handling terminal, while simultaneously initiating the corresponding handling process. Specifically, in this embodiment, a graded early warning information is generated based on the evaluation index, including the early warning level, early warning area, abnormal indicators, risk description, and early warning time.

[0031] Based on the range of assessment indices, natural resource risks are divided into four levels with clearly defined grading standards: Blue Alert, Low Risk: Assessment index below 30, indicating stable natural resource status with no significant risk; Yellow Alert, Moderate Risk: Assessment index 30-50, indicating slight anomalies in natural resources with potential risks; Orange Alert, Higher Risk: Assessment index 50-80, indicating significant anomalies in natural resources, with risks gradually escalating, potentially affecting normal land engineering construction or the surrounding ecological environment; Red Alert, High Risk: Assessment index above 80, indicating serious anomalies in natural resources, potentially triggering major risks such as geological disasters and ecological damage, requiring immediate action.

[0032] Based on the warning level, targeted warning information is automatically generated, including the warning level, warning area, precise latitude and longitude range, abnormal indicators such as excessively low soil moisture, excessively rapid land subsidence, risk description, and warning time. Warning information is pushed simultaneously to multiple terminals, including the mobile terminals of land engineering management personnel, the large screen of the project monitoring center, and the terminals of regional natural resources management departments. At the same time, relevant responsible persons are notified via SMS, app push notifications, and voice reminders to ensure rapid dissemination of warning information.

[0033] Based on different warning levels, differentiated response procedures are initiated: Blue Warning: A routine patrol mechanism is activated, and the monitoring frequency is increased from once per hour to once every 30 minutes to continuously track the status of natural resources; Yellow Warning: A warning verification process is initiated, with staff arranged to conduct on-site verification of abnormalities, analyze the causes of the abnormalities, and take targeted prevention and control measures, such as soil rehydration and vegetation maintenance, with daily verification reports submitted; Orange Warning: An emergency prevention and control process is activated, suspending land engineering construction in the relevant areas, organizing professional teams to conduct a comprehensive investigation, formulating specific response plans, and monitoring the effectiveness of the response in real time until the risk is reduced to yellow warning level or below; Red Warning: A major emergency response is immediately activated, evacuating surrounding personnel and construction equipment, sealing off the warning area, reporting to the superior natural resources management department and emergency management department, organizing experts for on-site assessment, implementing emergency response measures, such as geological reinforcement and water body treatment, and tracking risk changes throughout the process until the risk is eliminated. After the response is completed, monitoring data is updated, re-entered into the model for evaluation, forming a closed-loop management system of monitoring-warning-response-review.

[0034] Its beneficial effects lie in solving the problems of narrow monitoring scope and single data in traditional monitoring by monitoring the entire area, all elements, and all weather conditions, ensuring the comprehensiveness and real-time nature of monitoring data. Through targeted noise reduction, spatiotemporal alignment, and standardized preprocessing, data quality is effectively improved, laying a solid foundation for subsequent feature extraction and early warning analysis, and reducing errors caused by data interference. Thirdly, multi-algorithm fusion is used to extract multi-dimensional preliminary features, combined with adaptive weight allocation based on information gain ratio, and efficient feature aggregation is achieved through GCN, solving the problems of incomplete feature extraction and subjective weight allocation, and improving feature representativeness. A dynamic early warning model is built based on Transformer-LSTM, taking into account both long-distance dependence and temporal change capture, resulting in high early warning accuracy and rapid response, effectively avoiding early warning lag. A hierarchical early warning and closed-loop handling mechanism is established to achieve precise push and differentiated handling of early warning information, significantly improving the efficiency of risk handling.

[0035] Please see Figure 2 In a method for intelligent monitoring and early warning of natural resources in land engineering, a multi-source collaborative monitoring network is constructed to collect multi-source monitoring data of natural resources within the land engineering area. The multi-source monitoring data is then subjected to noise reduction, spatiotemporal alignment, and standardization to obtain standardized monitoring data. This process includes the following steps: Step 201: Denoising the UAV and satellite image data in the multi-source monitoring data of natural resources is performed by combining Gaussian filtering and median filtering to eliminate salt-and-pepper noise and Gaussian noise in the images and retain effective feature information; Wavelet analysis algorithm is used to decompose the data signal and remove high-frequency noise components from the time series data of soil and hydrology collected by the sensor to obtain noise-reduced monitoring data. Step 202: Calibrate the timestamps of all noise reduction monitoring data, and uniformly interpolate the data from different acquisition frequencies to one record per hour; use the WGS84 coordinate system as the spatial reference to perform coordinate transformation on the location data of the UAV, satellite imagery and monitoring stations, and use georeferencing technology to match the image data with the location of the monitoring stations to obtain aligned monitoring data; Step 203: Using the min-max standardization method, the aligned monitoring data of different dimensions and magnitudes are mapped to the [0,1] interval to obtain standardized monitoring data.

[0036] The above describes an embodiment of the intelligent monitoring and early warning method for natural resource surveys in land engineering according to the present invention. Please refer to [link / reference]. Figure 3 In a land engineering natural resource survey intelligent monitoring and early warning system, the natural resource survey intelligent monitoring and early warning system includes the following modules: The monitoring data acquisition module is used to construct a multi-source collaborative monitoring network and collect multi-source monitoring data of natural resources within the land engineering area; the multi-source monitoring data of natural resources are then subjected to noise reduction, spatiotemporal alignment and standardization to obtain standardized monitoring data. The feature extraction and fusion module is used to extract preliminary features from standardized monitoring data using CNN convolutional neural network, statistical analysis methods and YOLOv8 target detection algorithm respectively. It adaptively assigns importance weights to each feature based on information gain ratio algorithm, and aggregates and transforms the preliminary features through GCN graph convolutional network to obtain high-order fused features. The early warning model building module is used to build a dynamic early warning model based on the Transformer-LSTM network. It inputs high-order fusion features into the dynamic early warning model and outputs an assessment index of the natural resource risk level. The intelligent monitoring and early warning module is used to generate graded early warning information based on the assessment index and push it to the handling terminal, while simultaneously initiating the corresponding handling process.

[0037] Its beneficial effects lie in solving the problems of narrow monitoring scope and single data in traditional monitoring by monitoring the entire area, all elements, and all weather conditions, ensuring the comprehensiveness and real-time nature of monitoring data. Through targeted noise reduction, spatiotemporal alignment, and standardized preprocessing, data quality is effectively improved, laying a solid foundation for subsequent feature extraction and early warning analysis, and reducing errors caused by data interference. Thirdly, multi-algorithm fusion is used to extract multi-dimensional preliminary features, combined with adaptive weight allocation based on information gain ratio, and efficient feature aggregation is achieved through GCN, solving the problems of incomplete feature extraction and subjective weight allocation, and improving feature representativeness. A dynamic early warning model is built based on Transformer-LSTM, taking into account both long-distance dependence and temporal change capture, resulting in high early warning accuracy and rapid response, effectively avoiding early warning lag. A hierarchical early warning and closed-loop handling mechanism is established to achieve precise push and differentiated handling of early warning information, significantly improving the efficiency of risk handling.

[0038] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended technical solutions and equivalents.

Claims

1. A method for intelligent monitoring and early warning of natural resource surveys in land engineering, characterized in that, The intelligent monitoring and early warning method for natural resource surveys includes the following steps: A multi-source collaborative monitoring network is constructed to collect multi-source monitoring data of natural resources within the land engineering area; the multi-source monitoring data of natural resources are then subjected to noise reduction, spatiotemporal alignment, and standardization to obtain standardized monitoring data. Preliminary features of standardized monitoring data were extracted using CNN convolutional neural network, statistical analysis methods and YOLOv8 target detection algorithm respectively. The importance weights of each feature were adaptively assigned based on the information gain ratio algorithm. The preliminary features were then aggregated and transformed by GCN graph convolutional network to obtain high-order fused features. A dynamic early warning model is constructed based on the Transformer-LSTM network. The high-order fusion features are input into the dynamic early warning model, and the assessment index of natural resource risk level is output. Based on the assessment index, a graded early warning information is generated and pushed to the handling terminal, and the corresponding handling process is initiated at the same time.

2. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The aforementioned multi-source collaborative monitoring network is constructed to collect multi-source monitoring data of natural resources within the land engineering area; The multi-source monitoring data of natural resources are sequentially subjected to noise reduction, spatiotemporal alignment, and standardization to obtain standardized monitoring data, including: Noise reduction was performed on UAV and satellite imagery data from multi-source natural resource monitoring data using a combination of Gaussian filtering and median filtering to eliminate salt-and-pepper noise and Gaussian noise while retaining effective feature information. Wavelet analysis was used to decompose the data signals and remove high-frequency noise components from the time-series soil and hydrological data collected by sensors to obtain noise-reduced monitoring data. The timestamps of all noise reduction monitoring data are calibrated, and data from different acquisition frequencies are uniformly interpolated to one record per hour; the location data of UAVs, satellite images and monitoring stations are transformed using the WGS84 coordinate system as the spatial reference, and the image data is matched with the location of the monitoring stations through georegistration technology to obtain aligned monitoring data; The min-max standardization method is used to map the aligned monitoring data of different dimensions and magnitudes to the [0,1] interval to obtain standardized monitoring data.

3. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The method utilizes CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 object detection algorithm to extract preliminary features from standardized monitoring data. Based on the information gain ratio algorithm, importance weights are adaptively assigned to each feature. A GCN graph convolutional network is then used to aggregate and transform the preliminary features to obtain higher-order fused features, including: Spatial features of UAV and satellite imagery spatial data in standardized monitoring data are extracted layer by layer using the convolutional and pooling layers of a CNN convolutional neural network, including surface texture, vegetation distribution outline and terrain undulation features, and output spatial feature vectors. Statistical indicators of soil moisture, groundwater level and soil nutrient data are calculated using statistical analysis methods, including mean, variance, range, trend slope and coefficient of variation, to obtain the numerical variation characteristics and stability characteristics of the data, and output numerical feature vectors. This paper uses the YOLOv8 target detection algorithm to detect abnormal targets in image data, extracts the location, area and shape features of the targets, and outputs the target feature vector.

4. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The method utilizes CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 object detection algorithm to extract preliminary features from standardized monitoring data. Based on the information gain ratio algorithm, importance weights are adaptively assigned to each feature. A GCN graph convolutional network is then used to aggregate and transform the preliminary features to obtain higher-order fused features, including: The information gain ratio algorithm is used to calculate the information gain ratio of each preliminary feature for natural resource risk assessment. Using spatial feature vectors, numerical feature vectors, and target feature vectors as inputs, and historical risk data as labels, the information gain of each feature is calculated, and then divided by the entropy value of the feature to obtain the information gain ratio. After normalization, the weight coefficients of each feature are obtained, and the sum of the weight coefficients is 1, thus obtaining the preliminary features with weight coefficients.

5. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The method utilizes CNN convolutional neural networks, statistical analysis methods, and the YOLOv8 object detection algorithm to extract preliminary features from standardized monitoring data. Based on the information gain ratio algorithm, importance weights are adaptively assigned to each feature. A GCN graph convolutional network is then used to aggregate and transform the preliminary features to obtain higher-order fused features, including: Preliminary features with weighted coefficients are input into the GCN graph convolutional network. A feature association graph is constructed using monitoring stations as nodes and the spatial correlation of monitoring areas as edges. By using graph convolution operations to aggregate, transform, and fuse the initial features, the intrinsic relationships between different features are explored, and a high-order fused feature vector is output.

6. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The dynamic early warning model constructed based on the Transformer-LSTM network inputs the high-order fusion features into the dynamic early warning model and outputs an assessment index of the natural resource risk level, including: The Transformer-LSTM network structure includes an input layer, an encoding layer, a decoding layer, and an output layer. The input layer receives high-order fused feature vectors and temporal information. The encoding layer consists of a Transformer encoder and an LSTM layer. The Transformer encoder captures long-distance correlations between different features through a self-attention mechanism, while the LSTM layer learns the temporal variation patterns of features and captures the dynamic evolution trend of natural resource risks. The decoding layer decodes the encoded features and outputs risk assessment-related parameters. The output layer uses a fully connected layer combined with a sigmoid activation function to output an assessment index for natural resource risk levels between 0 and 100.

7. The intelligent monitoring and early warning method for natural resource surveys in land engineering as described in claim 1, characterized in that, The process of generating tiered early warning information based on the assessment index and pushing it to the handling terminal, while simultaneously initiating the corresponding handling procedure, includes: The assessment index generates tiered early warning information, which includes the warning level, warning area, abnormal indicators, risk description, and warning time.

8. A smart monitoring and early warning system for natural resource surveys in land engineering, characterized in that, The intelligent monitoring and early warning system for natural resource surveys includes the following modules: The monitoring data acquisition module is used to construct a multi-source collaborative monitoring network and collect multi-source monitoring data of natural resources within the land engineering area; the multi-source monitoring data of natural resources is sequentially subjected to noise reduction, spatiotemporal alignment and standardization to obtain standardized monitoring data; The feature extraction and fusion module is used to extract preliminary features from standardized monitoring data using CNN convolutional neural network, statistical analysis methods and YOLOv8 target detection algorithm respectively. It adaptively assigns importance weights to each feature based on information gain ratio algorithm, and aggregates and transforms the preliminary features through GCN graph convolutional network to obtain high-order fused features. The early warning model building module is used to construct a dynamic early warning model based on the Transformer-LSTM network. The high-order fusion features are input into the dynamic early warning model, and the assessment index of the natural resource risk level is output. The intelligent monitoring and early warning module is used to generate graded early warning information based on the assessment index and push it to the handling terminal, while simultaneously initiating the corresponding handling process.

9. The intelligent monitoring and early warning system for natural resource surveys in land engineering as described in claim 8, characterized in that, The feature extraction and fusion module includes the following sub-modules: The extraction submodule is used to extract spatial features from the UAV and satellite imagery spatial data in the standardized monitoring data layer by layer using the convolutional and pooling layers of the CNN convolutional neural network, including surface texture, vegetation distribution outline and terrain undulation features, and output spatial feature vectors. The analysis submodule is used to calculate statistical indicators of soil moisture, groundwater level and soil nutrient numerical data through statistical analysis methods, including mean, variance, range, trend slope and coefficient of variation, to obtain the numerical change characteristics and stability characteristics of the data, and output numerical feature vectors. The detection submodule is used to detect abnormal targets in image data based on the YOLOv8 target detection algorithm, extract the location, area and shape features of the target, and output the target feature vector.

10. The intelligent monitoring and early warning system for natural resource surveys in land engineering as described in claim 8, characterized in that, The feature extraction and fusion module includes the following sub-modules: The calculation submodule is used to calculate the information gain ratio of each preliminary feature to the natural resource risk assessment using the information gain ratio algorithm. The adjustment submodule is used to take spatial feature vectors, numerical feature vectors, and target feature vectors as inputs, use historical risk data as labels, calculate the information gain of each feature, divide it by the entropy value of the feature to obtain the information gain ratio, and obtain the weight coefficient of each feature through normalization. The sum of the weight coefficients is 1, resulting in the preliminary features with weight coefficients.

Images