Method and system for collecting geological heritage information in geological parks based on multi-source data

Abstract

The present invention discloses a method and system for collecting geological heritage information in a geopark based on multi-source data. The method comprises obtaining GIS surface images, geological heritage environmental parameters, and geological heritage update data, updating drone flight paths and executing drone flight strategies to obtain low-altitude aerial images, performing data matching and data correction, determining a data fusion region based on the multi-source corrected data, constructing a geological heritage multi-source data fusion model based on the multi-source corrected data, inputting the multi-source perception data of the geopark to be collected into the geological heritage multi-source data fusion model to obtain geological heritage collection information, and determining a collection plan based on the data fusion region and collection time. The method not only improves the efficiency and accuracy of geological heritage information collection in a geopark based on multi-source data, but also has good interpretability and can be directly applied to a geological heritage information collection system in a geopark based on multi-source data.

Description

Technical Field

[0001] The present invention relates to the field of data collection, and in particular to a method and system for collecting geological heritage information in a geological park based on multi-source data. Background Art

[0002] As an important place to display the history of earth's evolution, geological relics and natural landscapes, geological parks have important scientific, educational and tourism values. The improvement of geological relic information collection requirements and the emergence of new technologies have also promoted the development of geological relic information collection methods, which plays an important role in improving data collection efficiency and accuracy.

[0003] Traditional methods for collecting geological relic information mainly rely on manual surveys and single data sources, which have the following problems: low data collection efficiency, and the dynamic changes in geological relic information require the system to have incremental learning capabilities to adapt to the addition of new data and the updating of old data, and the existing methods do not update data in a timely manner; the diversity and complexity of geological relics make it difficult for a single data source to fully reflect their true status; there are spatiotemporal inconsistencies and modal differences between different data sources, making it difficult to effectively match and fuse these data; multimodal data are modally unbalanced and feature-laden, making it difficult to ensure the accuracy and reliability of the fusion results. Therefore, the present invention integrates multi-source perception data and designs a geological park geological relic information collection method and system based on multi-source data through dynamic incremental learning, spatiotemporal attention mechanism and cross-modal fusion technology. This overcomes the shortcomings of traditional geological relic information collection technology, comprehensively collects and efficiently manages geological relic information, and provides strong support for scientific research and tourism development in geological parks. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for collecting geological heritage information in a geological park based on multi-source data.

[0005] To achieve the above object, the present invention is implemented according to the following technical solutions:

[0006] The present invention comprises the following steps:

[0007] Construct a multi-source perception network to obtain multi-source perception data of the Geopark;

[0008] performing multi-source data matching on the multi-source perception data, correcting the multi-source perception data to obtain multi-source corrected data, and determining a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region;

[0009] Constructing a geological heritage multi-source data fusion model based on the multi-source corrected data;

[0010] The multi-source perception data of the geological park to be collected is input into the multi-source data fusion model of the geological relics to obtain the geological relics collection information, and the collection plan is determined according to the data fusion area and collection time.

[0011] Furthermore, the method for constructing the multi-source perception network to obtain multi-source perception data of the geological park includes:

[0012] Obtaining GIS surface images through satellite remote sensing technology, and generating a hierarchical navigation map of geological relics based on the GIS surface images and corresponding geographic coordinates; the hierarchical navigation map of geological relics is divided into levels according to continents, countries, administrative regions and park projects;

[0013] Ground sensors are set up at representative monitoring points within the Geopark to collect geological heritage environmental parameters, and the geological heritage environmental parameters are allocated to the corresponding levels of the geological heritage layered navigation map according to the geographical coordinates of the ground sensors; the representative monitoring points are determined by a clustering algorithm;

[0014] Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on user spatiotemporal information, and review user uploaded data to obtain geological heritage update data; the tourist trajectory includes the coordinates and elevation information of all tourists when uploading data;

[0015] Update the drone's flight path based on the tourist trajectory and the drone's original flight path, and execute the drone's flight strategy based on the updated flight path to obtain low-altitude aerial images;

[0016] The geopark multi-source perception data is composed of GIS surface images, geological heritage environmental parameters, geological heritage update data and low-altitude aerial images and transmitted to the transit database;

[0017] The method for constructing the multi-source data fusion model of geological relics includes:

[0018] The updated data of geological relics, the environmental correction parameters of geological relics and the corrected images of macro-geological relics were combined according to geographic coordinates and aligned in time series to form a multi-source data set. The multi-source data were divided into training set and validation set in a ratio of 7:3 using random forest.

[0019] The geological heritage multi-source data fusion model includes a dynamic incremental learning module, a spatiotemporal attention module, and a cross-modal fusion module;

[0020] The dynamic incremental learning module uses elastic weight solidification to dynamically adjust model parameters and retain important weights, and adopts a memory playback mechanism to store old data and conduct joint training with new data;

[0021] The spatiotemporal attention module uses a Transformer encoder to capture the temporal dependencies of environmental parameters and determine the temporal attention weights. It uses CBAM to focus on the relic feature map and macroscopic geological relics to correct the key areas of the image. The outputs of temporal attention and spatial attention are weightedly fused to generate spatiotemporal features.

[0022] The cross-modal fusion module uses a cross-modal attention mechanism to align features of different modalities and a gated multimodal fusion mechanism to dynamically adjust the contribution of each modality to output a unified fusion feature. The gated multimodal fusion mechanism includes a gating unit and a feature fusion module. The gating unit generates different modal weights based on Softmax, and the feature fusion module integrates the weighted modal features. The unified fusion feature includes a point cloud map of the spatial distribution of geological relics, a time series map of the evolution of geological relics, a feature description of the geological relics, and a feature vector of the geological relics.

[0023] The output layer extracts the multi-source data index associated with the unified fusion feature, and finally outputs the geological relic collection information; the geological relic collection information includes the unified fusion feature and the corresponding multi-source data index;

[0024] The loss functions include cross entropy loss function and cosine similarity loss function. The LAMB optimizer is used to adjust the model hyperparameters and improve the convergence speed, and a verification machine is used to verify the model performance. The cross entropy loss function is used to classify geological relics into predefined categories. The cosine similarity loss function ensures the consistency of multimodal features after fusion.

[0025] Furthermore, the method for obtaining the updated data of the geological relics includes:

[0026] Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on spatiotemporal information, process user uploaded data to obtain audit data, and retrieve the audit data based on user identity to obtain geological heritage update data; the user identity includes researchers and tourists; the spatiotemporal information is obtained from the EXIF ​​data of user-uploaded images;

[0027] The specific method of verifying the user identity and spatiotemporal information is as follows: determining the user identity according to the user code, and filtering out the spatiotemporal information and user uploaded data other than the target geological park according to the spatiotemporal information;

[0028] The specific method for processing user-uploaded data to obtain audit data is as follows: using a language model to process the text information uploaded by the user to obtain descriptive features, using a lightweight image model to process the relic image uploaded by the user to obtain image features and relic feature images, fusing the descriptive features with the image features to obtain relic features, and combining the relic features and relic feature images to form audit data; the language model includes a TF-IDF method and a mutual information method; the lightweight image model includes a CNN convolutional neural network, a PCA principal component analysis method, and an activation maximization method;

[0029] The specific method of retrieving the audit data according to the user identity to obtain the geological heritage update data is as follows: when the user identity is a researcher, the audit data is directly determined to be the geological heritage update data and the audit result is fed back to the user; when the user identity is a tourist, the audit data is compared and searched with the platform database; if there is no duplicate data, the audit data is determined to be the geological heritage update data and the audit result is fed back to the user; if the data is duplicated, the geological heritage update data is not generated and the audit result is fed back to the user; the platform database is composed of historically generated geological heritage update data;

[0030] The specific method for comparing and retrieving the audit data with the platform database is as follows: setting a distance threshold, calculating the upload distance between the audit data and the platform data based on the geographic coordinates, screening the platform data with an upload distance less than the distance threshold as fuzzy matching data, calculating the comprehensive similarity between the audit data and the fuzzy matching data, and determining whether the data is duplicated based on the comprehensive similarity; the comprehensive similarity is the sum of the cosine similarity and Jaccard similarity weighted values ​​of the two sets of data relic feature vectors.

[0031] Furthermore, the method for updating the flight path of the drone includes:

[0032] The original flight path of the drone is converted into UTM coordinates, and the DBSCAN clustering is used to process the tourist trajectory to form hot spots. The GIS surface image is divided into different pixel segments. For the geological heritage areas, the statistical method is used to set the regional weight according to the importance of the geological heritage;

[0033] Determine the objective function for updating the UAV flight path, expressed as:

[0034]

[0035]

[0036]

[0037]

[0038] in The objective function for updating the UAV flight path, is the path cost weight, is the path cost, is the number of waypoints, For the The UTM coordinates of the waypoints, is the Euclidean distance between adjacent waypoints, is the steering angle penalty weight, For the The heading angle of the waypoint, Penalize weights for high variance, For the The flight altitude of each waypoint, is the coverage weight, is the coverage, The area of ​​tourist hotspots verified by drones, is the total area of ​​tourist hotspots, is the blind area coverage weight coefficient, To add new coverage to the historical blind area, is the total area of ​​historical blind spots, is the risk factor weight, is the risk factor, For the The area weight of the block area, For drones The dwell time of the block area, For the path to The minimum distance to the center of the block area, is the smoothing factor, is the time weight, For time consistency, For the The flight speed of the segment path, For the Flight time of the path segment;

[0039] The original flight path of the drone is used as the initial path of the ant colony, and the adaptive ant colony algorithm is used to update the pheromone concentration:

[0040]

[0041]

[0042] in For the Update time and The pheromone concentration between For the Update time and The pheromone concentration between is the pheromone volatility intensity, is the number of ants in the ant colony, For the The ant in the Update time and The pheromone concentration increment between , also represents the adaptive weight, For elite ants The path weight of For elite ants In the Update time and The pheromone concentration increment between 、 The minimum and maximum values ​​of the ant paths that reach the end point in this iteration, is the pheromone intensity coefficient, For the Updated at the time The path length of the ants, is the coverage adaptive weight, is the number of unverified trajectory points, is the number of all trajectory points, is the cost adaptive weight, ;

[0043] Adopting an elite strategy to retain the top three optimal paths in each generation and exponentially increase the pheromone concentration of the optimal path , using Bayesian optimization parameter combination; the parameter combination includes 、 、 、 、 and ;

[0044] The pheromone concentration is updated multiple times, and the path with the highest pheromone concentration is selected as the optimal path. The objective function value is calculated based on the optimal path until the objective function is minimized and the output UAV flight path is stopped.

[0045] Furthermore, the method for determining the data fusion area includes:

[0046] The multi-source correction data includes macroscopic geological heritage correction images, geological heritage environment correction parameters and geological heritage update data;

[0047] Based on the geographic coordinate matching of GIS surface images and low-altitude aerial images in the transfer database, the matching results are imported into ArcGIS software for image synthesis to output macro-geological relic images. The tourist trajectory and the macro-geological relic image are located according to the geographic coordinates. The elevation information and time information of the tourist trajectory are used to correct the corresponding information of the macro-geological relic image to obtain the macro-geological relic corrected image. The macro-geological relic corrected image collected this time is compared with the last time to determine the image change area, and the geographic coordinate range corresponding to the image change area is set as the first fusion area.

[0048] Match the representative monitoring point in the transit database that is closest to the geographical coordinates of the tourist trajectory point, use the relic characteristics of the tourist trajectory point to correct the corresponding data of the geological relic environmental parameters of the representative monitoring point to obtain the geological relic environmental correction parameters, and set the geographical coordinates corresponding to the geological relic update data as the second fusion area.

[0049] Furthermore, the method for determining the acquisition plan includes:

[0050] Determine the change degree of geological relic information based on the deviation between the fusion area and the geological relic collection information. When the change degree of geological relic information is greater than the change threshold or the collection interval is greater than the set collection period, store the geological relic collection information in the local database and back it up to the cloud database. Otherwise, back up the geological relic collection information to the cloud database.

[0051] The change degree of the geological relic information is calculated as follows:

[0052]

[0053] in is the degree of change of geological heritage information, 、 is the fusion region weight, is the number of the first fusion regions, is the number of the second fusion regions, For the The area of ​​the first fusion region, For the The geological heritage collection information deviation of the first fusion area, No. The area of ​​the second fusion region, For the The geological heritage collection information deviation of the second fusion area.

[0054] Secondly, the geological park geological heritage information collection system based on multi-source data includes:

[0055] Perception network module: used to build a multi-source perception network to obtain multi-source perception data of the geological park; the multi-source perception network includes satellite remote sensing technology to obtain GIS surface images, ground sensors to collect geological heritage environmental parameters, geological heritage information update platform to obtain geological heritage update data, and drones to obtain low-altitude aerial images;

[0056] Platform module: used to verify user identity and spatiotemporal information, generate tourist trajectories based on the user's spatiotemporal information, and review user uploaded data to obtain updated data of the geological relics;

[0057] Drone module: used to update the drone flight path based on the tourist trajectory and the original flight path of the drone, and execute the drone flight strategy based on the updated flight path to obtain low-altitude aerial images;

[0058] A data processing module is configured to perform multi-source data matching on the multi-source perception data, correct the multi-source perception data to obtain multi-source corrected data, and determine a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region;

[0059] Model module: used to construct a geological heritage multi-source data fusion model based on the multi-source correction data, and input the multi-source perception data of the geological park to be collected into the geological heritage multi-source data fusion model to obtain geological heritage collection information;

[0060] Management module: used to view and manage the multi-source perception data and the geological heritage collection information, and determine the collection plan according to the data fusion area and collection time.

[0061] The beneficial effects of the present invention are:

[0062] The present invention is a method and system for collecting geological heritage information in geological parks based on multi-source data. Compared with the prior art, the present invention has the following technical effects:

[0063] The present invention can improve the data preprocessing capability and enhance the model adaptability in the geological heritage information collection of geological parks by constructing a multi-source perception network, setting up a platform, updating the flight path, data matching, data correction and building a model. It can realize the precise matching and intelligent fusion of multi-source heterogeneous data, thereby improving the efficiency and accuracy of geological heritage information collection in geological parks, optimizing the geological heritage information collection technology of geological parks, greatly saving resources, improving work efficiency, realizing the collection of geological heritage information in geological parks, and providing strong support for scientific research and tourism development of geological parks. It can adapt to different geological heritage information collection systems based on multi-source data and the terminal collection needs of geological heritage information of geological parks based on multi-source data of different users, and has a certain universality. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] Figure 1 The present invention is a flowchart of the steps of the method for collecting geological heritage information of geological parks based on multi-source data. DETAILED DESCRIPTION

[0065] The present invention will be further described below through specific examples. The illustrative examples and descriptions of the present invention are used to explain the present invention but are not intended to limit the present invention.

[0066] The method and system for collecting geological heritage information in a geological park based on multi-source data of the present invention include the following steps:

[0067] like Figure 1 As shown, in this embodiment, the following steps are included:

[0068] Construct a multi-source perception network to obtain multi-source perception data of the Geopark;

[0069] performing multi-source data matching on the multi-source perception data, correcting the multi-source perception data to obtain multi-source corrected data, and determining a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region;

[0070] Constructing a geological heritage multi-source data fusion model based on the multi-source corrected data;

[0071] Input the multi-source sensing data of the geological park to be collected into the multi-source data fusion model of the geological heritage to obtain the geological heritage collection information, and determine the collection plan according to the data fusion area and collection time;

[0072] The method for constructing the multi-source data fusion model of geological relics includes:

[0073] The updated data of geological relics, the environmental correction parameters of geological relics and the corrected images of macro-geological relics were combined according to geographic coordinates and aligned in time series to form a multi-source data set. The multi-source data were divided into training set and validation set in a ratio of 7:3 using random forest.

[0074] The geological heritage multi-source data fusion model includes a dynamic incremental learning module, a spatiotemporal attention module, and a cross-modal fusion module;

[0075] The dynamic incremental learning module uses an elastic weight solidification method to dynamically adjust model parameters and retain important weights, and adopts a memory playback mechanism to store old data and jointly train with new data. The spatiotemporal attention module uses a Transformer encoder to capture the temporal dependencies of environmental parameters and determine the temporal attention weights. It uses CBAM to focus on the relic feature map and the key areas of the macroscopic geological relic correction image, and performs a weighted fusion of the outputs of temporal attention and spatial attention to generate spatiotemporal features.

[0076] The cross-modal fusion module uses a cross-modal attention mechanism to align features of different modalities and a gated multimodal fusion mechanism to dynamically adjust the contribution of each modality to output a unified fusion feature. The gated multimodal fusion mechanism includes a gating unit and a feature fusion module. The gating unit generates different modal weights based on Softmax, and the feature fusion module integrates the weighted modal features. The unified fusion feature includes a point cloud map of the spatial distribution of geological relics, a time series map of the evolution of geological relics, a feature description of the geological relics, and a feature vector of the geological relics.

[0077] The output layer extracts the multi-source data index associated with the unified fusion feature, and finally outputs the geological relic collection information; the geological relic collection information includes the unified fusion feature and the corresponding multi-source data index;

[0078] The loss functions include cross entropy loss function and cosine similarity loss function. The LAMB optimizer is used to adjust the model hyperparameters and improve the convergence speed, and a verification machine is used to verify the model performance. The cross entropy loss function is used to classify geological relics into predefined categories. The cosine similarity loss function ensures the consistency of multimodal features after fusion.

[0079] In this embodiment, the method for constructing the multi-source perception network to obtain multi-source perception data of the geopark includes:

[0080] Obtaining GIS surface images through satellite remote sensing technology, and generating a hierarchical navigation map of geological relics based on the GIS surface images and corresponding geographic coordinates; the hierarchical navigation map of geological relics is divided into levels according to continents, countries, administrative regions and park projects;

[0081] Ground sensors are set up at representative monitoring points within the Geopark to collect geoheritage environmental parameters, and the geoheritage environmental parameters are assigned to the corresponding levels of the geoheritage hierarchical navigation map according to the geographical coordinates of the ground sensors; the representative monitoring points are determined by a clustering algorithm; the geoheritage environmental parameters include temperature, humidity, vibration, magnetic field and noise;

[0082] Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on user spatiotemporal information, and review user uploaded data to obtain geological heritage update data; the tourist trajectory includes the coordinates and elevation information of all tourists when uploading data;

[0083] Update the drone's flight path based on the tourist trajectory and the drone's original flight path, and execute the drone's flight strategy based on the updated flight path to obtain low-altitude aerial images;

[0084] The geopark multi-source perception data is composed of GIS surface images, geological heritage environmental parameters, geological heritage update data and low-altitude aerial images and transmitted to the transit database;

[0085] In the actual assessment, taking a geological park with an area of ​​100 square kilometers as an example, the GIS surface image obtained by satellite remote sensing technology has a resolution of 1m / pixel and is divided into m=10,000 areas;

[0086] Twenty representative monitoring points and locations in the geological park were determined through a clustering algorithm. Ground sensors were set up at the monitoring points to collect the environmental parameters of the geological relics. Monitoring data were obtained by taking the data of two monitoring points as examples: monitoring point 1 (temperature 25°C, humidity 60%, vibration 0.5m / s², magnetic field 45μT, noise 40dB), and monitoring point 2 (temperature 24°C, humidity 62%, vibration 0.4m / s², magnetic field 43μT, noise 42dB).

[0087] In this embodiment, the method for obtaining the updated data of the geological relics includes:

[0088] Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on spatiotemporal information, process user uploaded data to obtain audit data, and retrieve the audit data based on user identity to obtain geological heritage update data; the user identity includes researchers and tourists; the spatiotemporal information is obtained from the EXIF ​​data of user-uploaded images, including geographic coordinates, coordinate elevation and time information;

[0089] The specific method of verifying the user identity and spatiotemporal information is as follows: determining the user identity according to the user code, and filtering out the spatiotemporal information and user uploaded data other than the target geological park according to the spatiotemporal information;

[0090] The specific method for processing user-uploaded data to obtain audit data is as follows: using a language model to process the text information uploaded by the user to obtain descriptive features, using a lightweight image model to process the relic images uploaded by the user to obtain image features and relic feature images, fusing the descriptive features with the image features to obtain relic features, and combining the relic features and the relic feature images to form audit data; the language model includes a TF-IDF method and a mutual information method, the TF-IDF method performs feature extraction on the text information, and the mutual information method performs feature selection on the extracted features to obtain descriptive features; the lightweight image model includes a CNN convolutional neural network, a PCA principal component analysis method, and an activation maximization method, the CNN convolutional neural network performs feature extraction and classification on the relic images uploaded by the user, the PCA principal component analysis method performs feature selection and dimensionality reduction on the extracted features to obtain image features, and the activation maximization method generates a relic feature image based on the extracted features;

[0091] The specific method of retrieving the audit data according to the user identity to obtain the geological heritage update data is as follows: when the user identity is a researcher, the audit data is directly determined to be the geological heritage update data and the audit result is fed back to the user; when the user identity is a tourist, the audit data is compared and searched with the platform database; if there is no duplicate data, the audit data is determined to be the geological heritage update data and the audit result is fed back to the user; if the data is duplicated, the geological heritage update data is not generated and the audit result is fed back to the user; the platform database is composed of historically generated geological heritage update data;

[0092] The specific method for comparing and retrieving the audit data with the platform database is as follows: setting a distance threshold, calculating the upload distance between the audit data and the platform data based on geographic coordinates, screening the platform data whose upload distance is less than the distance threshold as fuzzy matching data, calculating the comprehensive similarity between the audit data and the fuzzy matching data, and determining whether the data is duplicated based on the comprehensive similarity; the comprehensive similarity is the sum of the cosine similarity and the Jaccard similarity weighted values ​​of the two sets of data relic feature vectors;

[0093] In the actual evaluation, the geological relic information update platform received 50 user-uploaded data within a week, of which 10 were uploaded by researchers (the 10 data were directly uploaded after being processed by the language model and lightweight image model and fed back to the researchers), and 40 were uploaded by tourists (the 40 data were processed by the language model and lightweight image model and compared with the platform database for retrieval. 15 data were determined to be non-duplicate and directly uploaded, and 25 data were determined to be duplicates and not uploaded, and fed back to the researchers). Taking one of the tourist-uploaded data as an example: its EXIF ​​data shows that the shooting location coordinates are (10.5, 20.3, 150) and the time is 10:00 on May 1, 2023. The uploaded text information is processed by the TF-IDF method and the mutual information method to obtain a description feature vector of [0.2, 0.3, 0.1]. The relic image is processed by the lightweight image model to obtain an image feature vector of [0.4, 0.2, 0.3]. The fused relic feature is [0.3, 0.25, 0.2]. The relic feature image is composed of pixel matrix data.

[0094] The distance threshold is set to 100m, and the upload distance between the audit data and the platform data is calculated to select 5 platform data that meet the distance requirements. The comprehensive similarity between the audit data and the 5 groups of fuzzy matching data is calculated to be 0.77, 0.59, 0.32, 0.6, and 0.15. The comprehensive similarity of the five groups of data is less than the comprehensive similarity threshold of 0.90. In this case, there is no duplicate data in the audit data, and it is determined to be the updated data of the geological relic and fed back to the tourists and uploaded to the transit database.

[0095] In this embodiment, the method for updating the flight path of the drone includes:

[0096] The original flight path of the drone is converted into UTM coordinates, and the DBSCAN clustering is used to process the tourist trajectory to form hot spots. The GIS surface image is divided into different pixel segments. For the geological heritage areas, the statistical method is used to set the regional weight according to the importance of the geological heritage;

[0097] Determine the objective function for updating the UAV flight path, expressed as:

[0098]

[0099]

[0100]

[0101]

[0102] in The objective function for updating the UAV flight path, is the path cost weight, is the path cost, is the number of waypoints, For the The UTM coordinates of the waypoints, is the Euclidean distance between adjacent waypoints, is the steering angle penalty weight, For the The heading angle of the waypoint, Penalize weights for high variance, For the The flight altitude of each waypoint, is the coverage weight, is the coverage, The area of ​​tourist hotspots verified by drones, is the total area of ​​tourist hotspots, is the blind area coverage weight coefficient, which is 0.8. To add new coverage to the historical blind area, is the total area of ​​historical blind spots, is the risk factor weight, is the risk factor, For the The area weight of the block area, For drones The dwell time of the block area, For the path to The minimum distance to the center of the block area, is the smoothing factor, is the time weight, For time consistency, For the The flight speed of the segment path, For the Flight time of the path segment;

[0103] The original flight path of the drone is used as the initial path of the ant colony, and the adaptive ant colony algorithm is used to update the pheromone concentration:

[0104]

[0105]

[0106] in For the Update time and The pheromone concentration between For the Update time and The pheromone concentration between is the pheromone volatility intensity, is the number of ants in the ant colony, For the The ant in the Update time and The pheromone concentration increment between , also represents the adaptive weight, For elite ants The path weight of For elite ants In the Update time and The pheromone concentration increment between 、 The minimum and maximum values ​​of the ant paths that reach the end point in this iteration, is the pheromone intensity coefficient, For the Updated at the time The path length of the ants, is the coverage adaptive weight, is the number of unverified trajectory points, is the number of all trajectory points, is the cost adaptive weight, ;

[0107] Adopting an elite strategy to retain the top three optimal paths in each generation and exponentially increase the pheromone concentration of the optimal path , using Bayesian optimization parameter combination; the parameter combination includes 、 、 、 、 and ;

[0108] Update the pheromone concentration multiple times, select the path with the highest pheromone concentration as the optimal path, calculate the objective function value based on the optimal path, and stop updating the output drone flight path when the objective function is minimized;

[0109] In the actual evaluation, the UTM plane coordinates of the 12 waypoints of the original drone path are {(50,80), (120,150), (200,210), (300,320), (400,450), (500,530), (600,680), (700,750), (800,820), (900,900), (1000,1050), (1100,1180)}. DBSCAN clustering was used to process the tourist trajectories to form five hotspots with areas of 8, 12, 10, 15, and 9 square kilometers, respectively;

[0110] Take path cost weight , coverage weight 5. Risk factor weight , time weight , steering angle penalty weight , height change penalty weight , blind spot coverage weight coefficient , pheromone volatility intensity , pheromone intensity coefficient , smoothing factor ; Calculate the objective function of the original path of the UAV , where: path cost , coverage ( 、 、 、 , unit square kilometer), risk coefficient , time consistency ;

[0111] After several iterations, the path with the highest pheromone concentration was selected as the optimal path, and the new waypoint coordinates were {(45,75), (110,140), (190,220), (290,310), (390,440), (490,520), (590,670), (690,740), (790,820), (890,895), (1000,1040), (1090,1170)}.

[0112] In this embodiment, the method for determining the data fusion area includes:

[0113] The multi-source correction data includes macroscopic geological heritage correction images, geological heritage environment correction parameters and geological heritage update data;

[0114] Based on the geographic coordinate matching of GIS surface images and low-altitude aerial images in the transfer database, the matching results are imported into ArcGIS software for image synthesis to output macro-geological relic images. The tourist trajectory and the macro-geological relic image are located according to the geographic coordinates. The elevation information and time information of the tourist trajectory are used to correct the corresponding information of the macro-geological relic image to obtain the macro-geological relic corrected image. The macro-geological relic corrected image collected this time is compared with the last time to determine the image change area, and the geographic coordinate range corresponding to the image change area is set as the first fusion area.

[0115] Match the representative monitoring point in the transit database that is closest to the geographical coordinates of the tourist trajectory point, use the heritage characteristics of the tourist trajectory point to correct the corresponding data of the geological heritage environmental parameters of the representative monitoring point to obtain the geological heritage environmental correction parameters, and set the geographical coordinates corresponding to the geological heritage update data as the second fusion area;

[0116] In the actual evaluation, taking the multi-source sensor data of a geological park with an area of ​​100 square kilometers within a week as an example, a comparison of the macroscopic geological relic correction map collected this time and the previous time revealed that the image of an area with a geographic coordinate range of (X:100-200, Y:300-400) and an area of ​​10,000 square meters had changed. This area was set as the first fusion area.

[0117] Taking the tourist trajectory point (10.5, 20.3, 150) as an example, the geological heritage environmental parameters of the nearest representative monitoring point are matched to correct the temperature to 25.5°C and the humidity to 61%. The 5m area centered on the tourist trajectory point (10.5, 20.3, 150) is set as the second fusion area.

[0118] The elastic weight solidification method uses EWC and playback buffer, and the expression is:

[0119]

[0120]

[0121]

[0122] in For the The Fisher information matrix value of the parameter, For old task datasets Medium sample expectations, For the model in parameters Next pair of samples Output Label The maximum likelihood of For the The parameters of the model, is the likelihood pair parameter The gradient, is the EWC regularization loss term, is the regularization strength, For the The parameters of the old model, For the The playback buffer at the moment, is the buffer update function, For the data samples, For samples The eigenvector of is the total number of buffer samples;

[0123] The spatiotemporal attention module introduces a spatiotemporal attenuation term and uses the spatiotemporal attenuation term to modify the attention score expression as follows:

[0124]

[0125]

[0126] in is the spatiotemporal attenuation term, is the geographical distance, is the time difference, 、 、 、 is the adjustable attenuation parameter, is the corrected attention score, For satellite characteristics, For drone features;

[0127] The multi-source perception data of the geological park to be collected is input into the multi-source data fusion model of the geological relics to obtain the geological relics collection information: the spatial distribution point cloud map of the geological relics, the time series map of the geological relics evolution, the characteristic description of the geological relics and the characteristic vector of the geological relics; taking the characteristic description of some geological relics as an example: tourist 001zzy found a columnar rock with a height of 15 meters and a diameter of about 3 meters at (10.5, 20.3, 150). The surface of the rock is gray-black with a honeycomb texture. There is a fossil with a complete clover outline on the bottom surface of the rock. The corresponding geological relic characteristic vector [regional weight, morphological uniqueness weight, texture and color feature weight, geological structure feature weight, paleontological information weight] is [0.4, 0.35, 0.3, 0.2, 0.18].

[0128] In this embodiment, the method for determining the acquisition plan includes:

[0129] Determine the change degree of geological relic information based on the deviation between the fusion area and the geological relic collection information. When the change degree of geological relic information is greater than the change threshold or the collection interval is greater than the set collection period, store the geological relic collection information in the local database and back it up to the cloud database. Otherwise, back up the geological relic collection information to the cloud database.

[0130] The change degree of the geological relic information is calculated as follows:

[0131]

[0132] in is the degree of change of geological heritage information, 、 is the fusion region weight, is the number of the first fusion regions, is the number of the second fusion regions, For the The area of ​​the first fusion region, For the The geological heritage collection information deviation of the first fusion area, No. The area of ​​the second fusion region, For the The deviation of geological heritage collection information in the second fusion area;

[0133] In the actual evaluation, there are 4 first fusion regions in this acquisition, and the corresponding regional areas and information deviations are (10000, 0.2), (8000, 0.18), (12000, 0.22), and (9000, 0.15), respectively. There are 8 first fusion regions, and the corresponding regional areas and information deviations are (78.5, 0.8), (12.56, 0.88), (12.56, 0.75), (50.24, 0.66), (78.5, 0.58), (50.24, 0.92), (28.26, 0.82), and (28.26, 0.79), respectively. The fusion region weights are: 、 , calculate the change degree of geological heritage information , which is greater than the change threshold of 3000, the geological heritage collection information will be stored in the local database and backed up to the cloud database.

[0134] Secondly, the geological park geological heritage information collection system based on multi-source data includes:

[0135] Perception network module: used to build a multi-source perception network to obtain multi-source perception data of the geological park; the multi-source perception network includes satellite remote sensing technology to obtain GIS surface images, ground sensors to collect geological heritage environmental parameters, geological heritage information update platform to obtain geological heritage update data, and drones to obtain low-altitude aerial images;

[0136] Platform module: used to verify user identity and spatiotemporal information, generate tourist trajectories based on the user's spatiotemporal information, and review user uploaded data to obtain updated data of the geological relics;

[0137] Drone module: used to update the drone flight path based on the tourist trajectory and the original flight path of the drone, and execute the drone flight strategy based on the updated flight path to obtain low-altitude aerial images;

[0138] A data processing module is configured to perform multi-source data matching on the multi-source perception data, correct the multi-source perception data to obtain multi-source corrected data, and determine a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region;

[0139] Model module: used to construct a geological heritage multi-source data fusion model based on the multi-source correction data, and input the multi-source perception data of the geological park to be collected into the geological heritage multi-source data fusion model to obtain geological heritage collection information;

[0140] Management module: used to view and manage the multi-source perception data and the geological heritage collection information, and determine the collection plan according to the data fusion area and collection time.

[0141] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for collecting geological heritage information in a geological park based on multi-source data, characterized in that: The following steps are involved: S1. Construct a multi-source perception network to obtain multi-source perception data of the Geopark; S2. Perform multi-source data matching on the multi-source perception data, correct the multi-source perception data to obtain multi-source corrected data, and determine a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region; S3, constructing a geological heritage multi-source data fusion model based on the multi-source corrected data; S4, inputting the multi-source sensing data of the geological park to be collected into the multi-source data fusion model of the geological heritage to obtain geological heritage collection information, and determining the collection plan according to the data fusion area and collection time; The method for constructing the multi-source data fusion model of geological relics includes: The updated data of geological relics, the environmental correction parameters of geological relics and the corrected images of macro-geological relics were combined according to geographic coordinates and aligned in time series to form a multi-source data set. The multi-source data were divided into training set and validation set in a ratio of 7:3 using random forest. The geological heritage multi-source data fusion model includes a dynamic incremental learning module, a spatiotemporal attention module, and a cross-modal fusion module; The dynamic incremental learning module uses elastic weight solidification to dynamically adjust model parameters and retain important weights, and adopts a memory playback mechanism to store old data and conduct joint training with new data; The spatiotemporal attention module uses a Transformer encoder to capture the temporal dependencies of environmental parameters and determine the temporal attention weights. It uses CBAM to focus on the relic feature map and macroscopic geological relics to correct the key areas of the image. The outputs of temporal attention and spatial attention are weightedly fused to generate spatiotemporal features. The cross-modal fusion module uses a cross-modal attention mechanism to align features of different modalities and a gated multimodal fusion mechanism to dynamically adjust the contribution of each modality to output a unified fusion feature. The gated multimodal fusion mechanism includes a gating unit and a feature fusion module. The gating unit generates different modal weights based on Softmax, and the feature fusion module integrates the weighted modal features. The unified fusion feature includes a point cloud map of the spatial distribution of geological relics, a time series map of the evolution of geological relics, a feature description of the geological relics, and a feature vector of the geological relics. The output layer extracts the multi-source data index associated with the unified fusion feature, and finally outputs the geological relic collection information; the geological relic collection information includes the unified fusion feature and the corresponding multi-source data index; The loss functions include cross entropy loss function and cosine similarity loss function. The LAMB optimizer is used to adjust the model hyperparameters and improve the convergence speed, and a verification machine is used to verify the model performance. The cross entropy loss function is used to classify geological relics into predefined categories. The cosine similarity loss function ensures the consistency of multimodal features after fusion.

2. The method for collecting geological heritage information in a geological park based on multi-source data according to claim 1, characterized in that: The method for constructing the multi-source perception network to obtain multi-source perception data of the geological park includes: Obtaining GIS surface images through satellite remote sensing technology, and generating a hierarchical navigation map of geological relics based on the GIS surface images and corresponding geographic coordinates; the hierarchical navigation map of geological relics is divided into levels according to continents, countries, administrative regions and park projects; Ground sensors are set up at representative monitoring points within the Geopark to collect geological heritage environmental parameters, and the geological heritage environmental parameters are allocated to the corresponding levels of the geological heritage layered navigation map according to the geographical coordinates of the ground sensors; the representative monitoring points are determined by a clustering algorithm; Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on user spatiotemporal information, and review user uploaded data to obtain geological heritage update data; the tourist trajectory includes the coordinates and elevation information of all tourists when uploading data; Update the drone's flight path based on the tourist trajectory and the drone's original flight path, and execute the drone's flight strategy based on the updated flight path to obtain low-altitude aerial images; The multi-source perception data of the geological park is composed of GIS surface images, geological heritage environmental parameters, geological heritage update data and low-altitude aerial images and transmitted to the transit database.

3. The method for collecting geological heritage information in a geological park based on multi-source data according to claim 2, characterized in that: The method for obtaining the updated data of the geological relics comprises: Build a geological heritage information update platform, verify user identity and spatiotemporal information, generate tourist trajectories based on spatiotemporal information, process user uploaded data to obtain audit data, and retrieve the audit data based on user identity to obtain geological heritage update data; the user identity includes researchers and tourists; the spatiotemporal information is obtained from the EXIF ​​data of user-uploaded images; The specific method of verifying the user identity and spatiotemporal information is as follows: determining the user identity according to the user code, and filtering out the spatiotemporal information and user uploaded data other than the target geological park according to the spatiotemporal information; The specific method for processing user-uploaded data to obtain audit data is as follows: using a language model to process the text information uploaded by the user to obtain descriptive features, using a lightweight image model to process the relic image uploaded by the user to obtain image features and relic feature images, fusing the descriptive features with the image features to obtain relic features, and combining the relic features and relic feature images to form audit data; the language model includes a TF-IDF method and a mutual information method; the lightweight image model includes a CNN convolutional neural network, a PCA principal component analysis method, and an activation maximization method; The specific method of retrieving the audit data according to the user identity to obtain the geological heritage update data is as follows: when the user identity is a researcher, the audit data is directly determined to be the geological heritage update data and the audit result is fed back to the user; when the user identity is a tourist, the audit data is compared and searched with the platform database; if there is no duplicate data, the audit data is determined to be the geological heritage update data and the audit result is fed back to the user; if the data is duplicated, the geological heritage update data is not generated and the audit result is fed back to the user; the platform database is composed of historically generated geological heritage update data; The specific method for comparing and retrieving the audit data with the platform database is as follows: setting a distance threshold, calculating the upload distance between the audit data and the platform data based on the geographic coordinates, screening the platform data with an upload distance less than the distance threshold as fuzzy matching data, calculating the comprehensive similarity between the audit data and the fuzzy matching data, and determining whether the data is duplicated based on the comprehensive similarity; the comprehensive similarity is the sum of the cosine similarity and Jaccard similarity weighted values ​​of the two sets of data relic feature vectors.

4. The method for collecting geological heritage information in a geological park based on multi-source data according to claim 2, characterized in that: The method for updating the flight path of the drone comprises: The original flight path of the drone is converted into UTM coordinates, and the DBSCAN clustering is used to process the tourist trajectory to form hot spots. The GIS surface image is divided into different pixel segments. For the geological heritage areas, the statistical method is used to set the regional weight according to the importance of the geological heritage; Determine the objective function for updating the UAV flight path, expressed as: in The objective function for updating the UAV flight path, is the path cost weight, is the path cost, is the number of waypoints, For the The UTM coordinates of the waypoints, is the Euclidean distance between adjacent waypoints, is the steering angle penalty weight, For the The heading angle of the waypoint, Penalize weights for high variance, For the The flight altitude of each waypoint, is the coverage weight, is the coverage, The area of ​​tourist hotspots verified by drones, is the total area of ​​tourist hotspots, is the blind area coverage weight coefficient, To add new coverage to the historical blind area, is the total area of ​​historical blind spots, is the risk factor weight, is the risk factor, For the The area weight of the block area, For drones The dwell time of the block area, For the path to The minimum distance to the center of the block area, is the smoothing factor, is the time weight, For time consistency, For the The flight speed of the segment path, For the Flight time of the path segment; The original flight path of the drone is used as the initial path of the ant colony, and the adaptive ant colony algorithm is used to update the pheromone concentration: in For the Update time and The pheromone concentration between For the Update time With the times The pheromone concentration between is the pheromone volatility intensity, is the number of ants in the ant colony, For the The ant in the Update time and The pheromone concentration increment between , also represents the adaptive weight, For elite ants The path weight of For elite ants In the Update time With the times The pheromone concentration increment between 、 The minimum and maximum values ​​of the ant paths that reach the end point in this iteration, is the pheromone intensity coefficient, For the Updated at the time The path length of the ants, is the coverage adaptive weight, is the number of unverified trajectory points, is the number of all trajectory points, is the cost adaptive weight, ; Adopting an elite strategy to retain the top three optimal paths in each generation and exponentially increase the pheromone concentration of the optimal path , using Bayesian optimization parameter combination; the parameter combination includes 、 、 、 、 and ; The pheromone concentration is updated multiple times, and the path with the highest pheromone concentration is selected as the optimal path. The objective function value is calculated based on the optimal path until the objective function is minimized and the output UAV flight path is stopped.

5. The method for collecting geological heritage information in a geological park based on multi-source data according to claim 1, characterized in that: The method for determining the data fusion area includes: The multi-source correction data includes macroscopic geological heritage correction images, geological heritage environment correction parameters and geological heritage update data; Based on the geographic coordinate matching of GIS surface images and low-altitude aerial images in the transfer database, the matching results are imported into ArcGIS software for image synthesis to output macro-geological relic images. The tourist trajectory and the macro-geological relic image are located according to the geographic coordinates. The elevation information and time information of the tourist trajectory are used to correct the corresponding information of the macro-geological relic image to obtain the macro-geological relic corrected image. The macro-geological relic corrected image collected this time is compared with the last time to determine the image change area, and the geographic coordinate range corresponding to the image change area is set as the first fusion area. Match the representative monitoring point in the transit database that is closest to the geographical coordinates of the tourist trajectory point, use the relic characteristics of the tourist trajectory point to correct the corresponding data of the geological relic environmental parameters of the representative monitoring point to obtain the geological relic environmental correction parameters, and set the geographical coordinates corresponding to the geological relic update data as the second fusion area.

6. The method for collecting geological heritage information in a geological park based on multi-source data according to claim 1, characterized in that: The method for determining the acquisition plan includes: Determine the change degree of geological relic information based on the deviation between the fusion area and the geological relic collection information. When the change degree of geological relic information is greater than the change threshold or the collection interval is greater than the set collection period, store the geological relic collection information in the local database and back it up to the cloud database. Otherwise, back up the geological relic collection information to the cloud database. The change degree of the geological relic information is calculated as follows: in is the degree of change of geological heritage information, 、 is the fusion region weight, is the number of the first fusion regions, is the number of the second fusion regions, For the The area of ​​the first fusion region, For the The geological heritage collection information deviation of the first fusion area, No. The area of ​​the second fusion region, For the The geological heritage collection information deviation of the second fusion area.

7. A geological park geological heritage information acquisition system based on multi-source data, used to execute the method according to any one of claims 1 to 6, characterized in that: include: Perception network module: used to build a multi-source perception network to obtain multi-source perception data of the geological park; the multi-source perception network includes satellite remote sensing technology to obtain GIS surface images, ground sensors to collect geological heritage environmental parameters, geological heritage information update platform to obtain geological heritage update data, and drones to obtain low-altitude aerial images; Platform module: used to verify user identity and spatiotemporal information, generate tourist trajectories based on the user's spatiotemporal information, and review user uploaded data to obtain updated data of the geological relics; Drone module: used to update the drone flight path based on the tourist trajectory and the original flight path of the drone, and execute the drone flight strategy based on the updated flight path to obtain low-altitude aerial images; A data processing module is configured to perform multi-source data matching on the multi-source perception data, correct the multi-source perception data to obtain multi-source corrected data, and determine a data fusion region based on the multi-source corrected data; the data fusion region includes a first fusion region and a second fusion region; Model module: used to construct a geological heritage multi-source data fusion model based on the multi-source correction data, and input the multi-source perception data of the geological park to be collected into the geological heritage multi-source data fusion model to obtain geological heritage collection information; Management module: used to view and manage the multi-source perception data and the geological heritage collection information, and determine the collection plan according to the data fusion area and collection time.

Images