Cloud-based surveying and mapping data processing method and system

Abstract

The invention discloses a cloud-based surveying and mapping data processing method and system, and the method comprises the steps: collecting a multi-source surveying and mapping data set, and carrying out the data fusion of the multi-source surveying and mapping data set based on an improved hierarchical Transform model in a cloud platform; and establishing an MTL-DNN surveying and mapping data analysis model based on an MTL-DNN multi-task deep learning network, performing pruning operation on the model by using a fine-grained pruning device, and inputting the fused multi-source surveying and mapping data set into the target MTL-DNN surveying and mapping data analysis model to perform terrain modeling, ground feature classification and risk prediction, so that model parameters and calculation amount can be reduced, and the modeling efficiency is improved. The deployment efficiency is improved; accurate and efficient data support and decision basis are provided for the fields of surveying and mapping engineering, disaster early warning and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of data analysis, and particularly to a cloud-based mapping data processing method and system thereof. Background Art

[0002] In the field of surveying and mapping, with the continuous development of technologies such as unmanned aerial vehicles, satellite remote sensing, and ground measurement equipment, the acquisition methods of mapping data have become increasingly diverse, and the data volume has also increased explosively. Traditional mapping data processing methods are usually carried out locally and rely on high-performance local computers or servers. This not only has high requirements for hardware devices, increasing costs, but also has great limitations in data storage, computing resource sharing, and collaborative processing. Although existing cloud-based mapping data processing methods have solved the problems of data storage and resource sharing to a certain extent, there is still room for improvement in terms of data processing efficiency, accuracy, and intelligence. In the data preprocessing stage, the identification and processing of noise data are not precise enough; in the data feature extraction and analysis stage, there is a lack of efficient algorithm support, resulting in low accuracy and reliability of the processing results. Summary of the Invention

[0003] The purpose of the present invention is to solve the above problems and design a cloud-based mapping data processing method and system thereof.

[0004] Furthermore, in the above-mentioned cloud-based mapping data processing method, the cloud-based mapping data processing method includes the following steps: Collect satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocess the collected data to obtain a multi-source mapping data set, and transmit the multi-source mapping data set to the cloud platform; Perform data fusion on the multi-source mapping data set based on an improved hierarchical Transformer model in the cloud platform, introduce a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain a fused multi-source mapping data set; Establish an MTL-DNN mapping data analysis model based on the MTL-DNN multi-task deep learning network, and perform pruning operations on the model using a fine-grained pruner to obtain a target MTL-DNN mapping data analysis model; Input the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, ground object classification, and risk prediction, and obtain mapping data analysis results.

[0005] Further, in the above cloud-based mapping data processing method, collecting satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocessing the collected data to obtain a multi-source mapping data set, and transmitting the multi-source mapping data set to the cloud platform includes: Collect satellite remote sensing data, perform radiometric correction and geometric correction on the satellite remote sensing data based on a model of atmospheric transmission theory to obtain first mapping data; Obtain UAV LiDAR data, perform point cloud denoising on the UAV LiDAR data using a statistical filtering method, and perform coordinate transformation and point cloud resampling on the denoised data using linear interpolation to obtain second mapping data; Collect Internet of Things sensor data, perform outlier processing on the Internet of Things sensor data using polynomial interpolation, and synchronize the time of the data after outlier processing to obtain third mapping data; Perform normalization processing on the first mapping data, second mapping data, and third mapping data, and integrate the normalized data to obtain a multi-source mapping data set; Encrypt the multi-source mapping data set, and transmit the encrypted multi-source mapping data set to the cloud platform based on 5G communication.

[0006] Further, in the above cloud-based mapping data processing method, performing data fusion on the multi-source mapping data set using an improved hierarchical Transformer model in the cloud platform, introducing a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps to obtain a fused multi-source mapping data set, includes: Establish a hierarchical Transformer model in the cloud platform using a hierarchical structure of multi-layer Transformer encoders and decoders; Each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, an encoder-decoder attention mechanism is introduced to obtain an improved hierarchical Transformer model; Add a spatio-temporal attention module to the encoder and decoder of the improved hierarchical Transformer model, and use the spatio-temporal attention module to process the time dimension and space dimension information of the data respectively.

[0007] Further, in the above cloud-based mapping data processing method, performing data fusion on the multi-source mapping data set using an improved hierarchical Transformer model in the cloud platform, introducing a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps to obtain a fused multi-source mapping data set, further includes: Obtain multi-source mapping data sets in the cloud platform and input the multi-source mapping data sets into an improved hierarchical Transformer model; Use the underlying encoder of the model to perform preliminary feature extraction on the input multi-source data, and dynamically associate data with different resolutions and timestamps through a spatio-temporal attention module; Use the spatio-temporal attention mechanism to calculate attention weights according to spatial positions and distances, fuse high-resolution detailed features and low-resolution global features to obtain a fused multi-source mapping data set.

[0008] Further, in the above cloud-based mapping data processing method, an MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network, and a fine-grained pruning tool is used to perform pruning operations on the model to obtain a target MTL-DNN mapping data analysis model, including: An MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network. The shared layer is located at the bottom of the MTL-DNN model and is used to extract common features of the data; According to the different requirements of the three tasks of terrain modeling, feature classification, and risk prediction, task-specific layers are designed; The output of the shared layer serves as the input to the three task-specific layers at the same time. The task-specific layers are independent of each other and only share features at the shared layer.

[0009] Further, in the above cloud-based mapping data processing method, an MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network, and a fine-grained pruning tool is used to perform pruning operations on the model to obtain a target MTL-DNN mapping data analysis model, further including: Use the Adam optimizer to set the learning rate, momentum, and weight decay parameters of the MTL-DNN mapping data analysis model; Based on the fine-grained pruning tool, calculate the absolute value magnitudes of the weights in the network, set a pruning threshold, set the weights with absolute values less than the threshold to zero, and remove unimportant connections to obtain the target MTL-DNN mapping data analysis model.

[0010] Further, in the above cloud-based mapping data processing method, inputting the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, feature classification, and risk prediction to obtain mapping data analysis results, including: Through the spatial and temporal features extracted by the shared layer, combined with the processing of the task-specific layers, output the three-dimensional coordinates and height information of the terrain to obtain terrain modeling; The features such as spectra, textures, and shapes extracted through the shared layer, combined with the classifier in the task-specific layer, output the class probability distribution of the ground objects to obtain the ground object classification. Fuse the time series features, spatial features, and ground object features in the data, combine the historical risk data and relevant risk indicators, and through the classification algorithm, output the risk level or the probability of risk occurrence to obtain the risk prediction.

[0011] Furthermore, in a cloud-based surveying and mapping data processing system, the cloud-based surveying and mapping data processing system includes the following modules: The data acquisition and transmission module is used to collect satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocess the collected data to obtain a multi-source surveying and mapping data set, and transmit the multi-source surveying and mapping data set to the cloud platform. The surveying and mapping data fusion module is used to perform data fusion on the multi-source surveying and mapping data set based on the improved hierarchical Transformer model in the cloud platform, introduce the spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain the fused multi-source surveying and mapping data set. The analysis model establishment module is used to establish an MTL-DNN surveying and mapping data analysis model based on the MTL-DNN multi-task deep learning network, and use a pruning tool to perform pruning operations on the model to obtain the target MTL-DNN surveying and mapping data analysis model. The surveying and mapping data analysis module is used to input the fused multi-source surveying and mapping data set into the target MTL-DNN surveying and mapping data analysis model to perform terrain modeling, ground object classification, and risk prediction, and obtain the surveying and mapping data analysis results.

[0012] Furthermore, in a cloud-based surveying and mapping data processing system, the surveying and mapping data fusion module includes the following sub-modules: The establishment sub-module is used to establish a hierarchical Transformer model in the cloud platform using the hierarchical structure of multi-layer Transformer encoders and decoders. The introduction sub-module is such that each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, the encoder-decoder attention mechanism is introduced to obtain the improved hierarchical Transformer model. The addition sub-module is used to add a spatio-temporal attention module to the encoder and decoder of the improved hierarchical Transformer model, and use the spatio-temporal attention module to process the time dimension and space dimension information of the data respectively.

[0013] Furthermore, in a cloud-based surveying and mapping data processing system, the surveying and mapping data fusion module also includes the following sub-modules: An input sub-module for obtaining a multi-source mapping data set in a cloud platform and inputting the multi-source mapping data set into an improved hierarchical Transformer model; An extraction sub-module for preliminarily extracting features of the input multi-source data by using the underlying encoder of the model and dynamically associating data with different resolutions and timestamps through a spatio-temporal attention module; A fusion sub-module for calculating attention weights according to spatial positions and distances by using a spatio-temporal attention mechanism, and fusing high-resolution detailed features and low-resolution global features to obtain a fused multi-source mapping data set.

[0014] The beneficial effects are as follows: by collecting satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocessing the collected data to obtain a multi-source mapping data set, and transmitting the multi-source mapping data set to a cloud platform; performing data fusion on the multi-source mapping data set based on an improved hierarchical Transformer model in the cloud platform, introducing a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps to obtain a fused multi-source mapping data set; establishing an MTL-DNN mapping data analysis model based on an MTL-DNN multi-task deep learning network, and pruning the model by using a fine-grained pruning device to obtain a target MTL-DNN mapping data analysis model; inputting the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, ground object classification, and risk prediction to obtain mapping data analysis results. It can reduce model parameters and calculation amounts, and improve deployment efficiency; finally, realize terrain modeling, ground object classification, and risk prediction, output visualization results and verify optimization, and provide accurate and efficient data support and decision-making basis for fields such as mapping engineering and disaster warning. Description of the Drawings

[0015] By reading the detailed description of the preferred embodiments below, various other advantages and benefits will become clear to those of ordinary skill in the art. The drawings are only for the purpose of showing the preferred embodiments and are not considered to be a limitation of the present invention.

[0016] Figure 1 Schematic diagram of the first embodiment of a cloud-based mapping data processing method in an embodiment of the present invention; Figure 2 Schematic diagram of the second embodiment of a cloud-based mapping data processing method in an embodiment of the present invention; Figure 3 Schematic diagram of the first embodiment of a cloud-based mapping data processing system in an embodiment of the present invention. Detailed Embodiments

[0017] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present 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 only used to explain the present invention and are not used to limit the present invention.

[0018] Those skilled in the art of the present technology can understand that unless specifically stated, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the term "comprising" used in the specification of the present invention means the presence of the described features, integers, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components and / or their groups.

[0019] The present invention will be specifically described below with reference to the accompanying drawings. As Figure 1 shown, a cloud-based mapping data processing method, the cloud-based mapping data processing method includes the following steps: Step 101, collect satellite remote sensing data, UAV LiDAR data and Internet of Things sensor data, preprocess the collected data to obtain a multi-source mapping data set, and transmit the multi-source mapping data set to the cloud platform; Specifically, in this embodiment, satellite remote sensing data is collected, and radiation correction and geometric correction are performed on the satellite remote sensing data based on a model of atmospheric transmission theory to obtain the first mapping data; Obtain UAV LiDAR data, perform point cloud denoising on the UAV LiDAR data using a statistical filtering method, and perform coordinate transformation and point cloud resampling on the denoised data using linear interpolation to obtain the second mapping data; Collect Internet of Things sensor data, perform outlier processing on the Internet of Things sensor data using polynomial interpolation, and synchronize the time of the data after outlier processing to obtain the third mapping data; Perform normalization processing on the first mapping data, the second mapping data and the third mapping data, and integrate the normalized data to obtain a multi-source mapping data set; Encrypt the multi-source mapping data set, and transmit the encrypted multi-source mapping data set to the cloud platform based on 5G communication.

[0020] Step 102, perform data fusion on the multi-source mapping data set based on an improved hierarchical Transformer model in the cloud platform, introduce a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain a fused multi-source mapping data set; Specifically, in this embodiment, a hierarchical Transformer model is established in the cloud platform using a hierarchical structure of multiple Transformer encoders and decoders; Each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, an encoder-decoder attention mechanism is introduced to obtain an improved hierarchical Transformer model; A spatio-temporal attention module is added to the encoder and decoder of the improved hierarchical Transformer model, and the spatio-temporal attention module is used to process the time dimension and space dimension information of the data respectively.

[0021] Obtain a multi-source mapping data set in the cloud platform, and input the multi-source mapping data set into the improved hierarchical Transformer model; Use the bottom layer encoder of the model to perform preliminary feature extraction on the input multi-source data, and dynamically associate data with different resolutions and timestamps through the spatio-temporal attention module; Use the spatio-temporal attention mechanism to calculate the attention weights according to the spatial position and distance, and fuse the high-resolution detailed features and low-resolution global features to obtain a fused multi-source mapping data set.

[0022] Step 103: Establish an MTL-DNN mapping data analysis model based on the MTL-DNN multi-task deep learning network, and use a fine-grained pruning tool to perform pruning operations on the model to obtain a target MTL-DNN mapping data analysis model; Specifically, in this embodiment, an MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network. The shared layer is located at the bottom layer of the MTL-DNN model and is used to extract the common features of the data; Design task-specific layers according to the different requirements of the three tasks of terrain modeling, feature classification, and risk prediction; The output of the shared layer serves as the input to the three task-specific layers at the same time. The task-specific layers are independent of each other and only share features at the shared layer.

[0023] Use the Adam optimizer to set the learning rate, momentum, and weight decay parameters of the MTL-DNN mapping data analysis model; Calculate the absolute value of each weight in the network based on the fine-grained pruning tool, set a pruning threshold, set the weights with absolute values less than the threshold to zero, and remove unimportant connections to obtain the target MTL-DNN mapping data analysis model.

[0024] Step 104: Input the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, feature classification, and risk prediction to obtain mapping data analysis results.

[0025] Specifically, in this embodiment, the spatial and temporal features extracted by the shared layer are combined with the processing of the task-specific layer to output the three-dimensional coordinates and height information of the terrain, and terrain modeling is obtained; The features such as spectrum, texture, and shape extracted by the shared layer are combined with the classifier of the task-specific layer to output the class probability distribution of the ground objects, and ground object classification is obtained; The time series features, spatial features, and ground object features in the fusion data are combined with the historical risk data and relevant risk indicators, and through a classification algorithm, the risk level or the probability of risk occurrence is output, and risk prediction is obtained.

[0026] Its beneficial effect is that it can reduce the model parameters and computational amount, and improve the deployment efficiency; finally, terrain modeling, ground object classification, and risk prediction are realized, and visual results are output and verified and optimized, providing accurate and efficient data support and decision-making basis for fields such as surveying and mapping engineering and disaster warning.

[0027] Please refer to Figure 2 , in a cloud-based surveying and mapping data processing method, based on an improved hierarchical Transformer model in a cloud platform, data fusion is performed on a multi-source surveying and mapping data set, and a spatio-temporal attention mechanism is introduced to dynamically associate data with different resolutions and timestamps, and the steps for obtaining a fused multi-source surveying and mapping data set include the following: Step 201, establish a hierarchical Transformer model using a hierarchical structure of multiple layers of Transformer encoders and decoders in the cloud platform; Each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, an encoder-decoder attention mechanism is introduced to obtain an improved hierarchical Transformer model; Step 203, add spatio-temporal attention modules to the encoder and decoder of the improved hierarchical Transformer model, and use the spatio-temporal attention modules to process the time dimension and space dimension information of the data respectively.

[0028] The above introduces the embodiments of a cloud-based surveying and mapping data processing method of the present invention. Please refer to Figure 3 , in a cloud-based surveying and mapping data processing system, the cloud-based surveying and mapping data processing system includes the following modules: A data acquisition and transmission module, which is used to collect satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocess the collected data to obtain a multi-source surveying and mapping data set, and transmit the multi-source surveying and mapping data set to the cloud platform; A mapping data fusion module, which is used to perform data fusion on multi-source mapping data sets based on an improved hierarchical Transformer model in a cloud platform, introduce a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain a fused multi-source mapping data set; An analysis model establishment module, which is used to establish an MTL-DNN mapping data analysis model based on the MTL-DNN multi-task deep learning network, and use a pruning tool to perform pruning operations on the model to obtain a target MTL-DNN mapping data analysis model; A mapping data analysis module, which is used to input the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, feature classification, and risk prediction, and obtain mapping data analysis results.

[0029] Specifically, the present invention can also be implemented through the following steps: I. Data collection and preprocessing; (1) Data collection; Satellite remote sensing data collection; Define the mapping area scope, accuracy requirements, and required spectral information of the project, etc., and select appropriate satellite data sources based on this, such as high-resolution series satellites, Sentinel satellites, etc. Within a predetermined acquisition time window, start the satellite data receiving device to receive and store satellite remote sensing data in real time. UAV LiDAR data collection; Control the UAV to fly according to the planned route, and at the same time start the LiDAR sensor to collect data, and record the flight attitude data (longitude, latitude, altitude, heading angle, pitch angle, roll angle, etc.) and LiDAR point cloud data in real time. Internet of Things sensor data collection; Set the data collection frequency and transmission time interval of the sensor, start the sensor to collect data, and the data collection terminal receives and stores the data sent by the sensor in real time. (2) Data preprocessing; Satellite remote sensing data preprocessing; Radiometric correction: Due to the influence of factors such as atmospheric scattering and sensor response characteristics, there are radiometric errors in satellite remote sensing data, and radiometric correction is required. Use a radiometric correction model, such as a model based on atmospheric transmission theory, to correct the data, eliminate the influence of the atmosphere and the sensor, and restore the true radiance of the ground object. Geometric correction: During the acquisition of satellite remote sensing data, due to factors such as satellite attitude changes, the curvature of the earth, and terrain undulations, geometric distortion will occur and geometric correction is required. Select appropriate ground control points and establish a geometric correction model, such as a polynomial model, collinear equation model, etc., to correct the data and convert the image to the correct position in the geographic coordinate system. Data cropping and mosaicking: According to the scope of the surveying and mapping area, crop the corrected satellite remote sensing data to remove data from irrelevant areas. If the surveying and mapping area is large and covered by multiple satellite images, image mosaicking is required to ensure that the mosaicked image is seamless and geometrically consistent. Noise removal: Use filtering algorithms, such as median filtering, Gaussian filtering, etc., to remove noise from satellite remote sensing data and improve the quality of the image. Preprocessing of UAV LiDAR data; Point cloud denoising: UAV LiDAR data may contain noise points, such as points generated by vegetation shaking, ground debris reflection, etc., and denoising processing is required. Use methods such as statistical filtering, radius filtering, voxel grid filtering, etc., to remove outliers and noise points and retain effective ground and feature point cloud data. Coordinate transformation: Convert the coordinate system of UAV LiDAR data to the unified coordinate system required by the project, such as the WGS84 coordinate system, the national geodetic coordinate system, etc. First, obtain the attitude data and GNSS positioning data during the UAV flight, and use the coordinate transformation model to convert the LiDAR point cloud data from the sensor coordinate system to the world coordinate system, and then convert it to the unified target coordinate system. Point cloud resampling: To improve the uniformity and resolution of point cloud data, resampling processing is performed on LiDAR point cloud data. Use interpolation algorithms, such as nearest neighbor interpolation, linear interpolation, etc., to increase the point cloud density in sparse areas and appropriately reduce the number of point clouds in dense areas to make the point cloud data more uniform. Feature separation: Use the height information, density characteristics, etc. of the point cloud to separate the ground points and feature points (buildings, trees, vehicles, etc.). Use a ground point extraction algorithm based on slope and curvature, such as the Progressive Encryption Aspect Ratio (PEARL) algorithm, to extract the ground point cloud, generate a Digital Elevation Model (DEM), and at the same time separate the feature point cloud data. Preprocessing of Internet of Things sensor data; Outlier Detection and Repair: Outliers may appear in IoT sensor data due to sensor failures, communication interference, etc., and need to be detected and repaired. Statistical methods (Z-score method, IQR method) or machine learning algorithms (Isolation Forest, One-Class SVM) are used to detect outliers. For a small number of outliers, interpolation methods (linear interpolation, polynomial interpolation) can be used for repair; for a large number of outliers or data with sensor failures, they should be marked as invalid data and processed subsequently. Data Format Unification: The data formats output by different types of IoT sensors may be different, and they need to be converted into a unified data format, such as CSV, JSON, etc., to facilitate subsequent data fusion and processing. Define unified data fields and data types to ensure that each sensor data contains necessary information such as timestamps, location information, monitoring parameters, etc. Time Synchronization: Since there may be differences in the acquisition times of IoT sensors, it is necessary to perform time synchronization processing on the data. Taking the acquisition times of satellite remote sensing data and UAV LiDAR data as the reference, calibrate the timestamps of IoT sensor data to ensure the consistency of multi-source data in the time dimension. (3) Obtain a multi-source mapping dataset; Integrate the preprocessed satellite remote sensing data, UAV LiDAR data, and IoT sensor data according to a unified data format and coordinate system to establish a multi-source mapping dataset. During the data integration process, conduct quality inspections on the data to ensure the integrity, accuracy, and consistency of the data. Record metadata such as the data source, acquisition time, and preprocessing process for subsequent data management and traceability. II. Data Fusion; (1) Construction of an improved hierarchical Transformer model; Model Architecture Design; Building the basic hierarchical Transformer structure: Adopt a hierarchical structure of multiple layers of Transformer encoders and decoders. Each layer of the encoder consists of a multi-head self-attention mechanism and a feed-forward neural network, which are used for feature extraction and encoding of the input data; the decoder also consists of a multi-head self-attention mechanism and a feed-forward neural network, and at the same time introduce an encoder-decoder attention mechanism to handle the context information during the decoding process. Introduction of spatio-temporal attention mechanism: Spatio-temporal attention modules are added to the encoder and decoder of the hierarchical Transformer model. The spatio-temporal attention modules process the time dimension and space dimension information of the data respectively. For the time dimension, using the correlation of the time series, the attention weights between data at different timestamps are calculated to dynamically associate data at different times; for the space dimension, considering the spatial position and geometric relationship of the data, the attention weights between data at different spatial positions are calculated to achieve effective fusion of spatial data at different resolutions. Design of multi-source data input interface: In order to adapt to the different formats and characteristics of satellite remote sensing data, UAV LiDAR data and Internet of Things sensor data, a multi-source data input interface is designed. The satellite remote sensing data is converted into image feature vectors, the UAV LiDAR data is converted into point cloud feature vectors, and the Internet of Things sensor data is converted into numerical feature vectors. Different types of feature vectors are mapped to a feature space of the same dimension through the embedding layer for input into the hierarchical Transformer model for processing.

[0030] Initialization of model parameters; Use the pre-trained Transformer model parameters as initial values, such as the parameters of models like BERT and ViT, and fine-tune the parameters of the embedding layer, attention mechanism, and feed-forward neural network of the model in combination with the characteristics of surveying and mapping data. For the parameters of the spatio-temporal attention module, they are initialized according to the characteristics of the time and space dimensions. For example, the weight matrix of time attention is initialized as a uniform distribution, and the weight matrix of space attention is initialized as a Gaussian distribution related to the spatial distance. (2) Data fusion process; Input of multi-source data: The preprocessed satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data are converted into feature vectors of a unified dimension through the multi-source data input interface and input into the improved hierarchical Transformer model. Hierarchical feature extraction and spatio-temporal correlation: In the bottom encoder of the model, preliminary feature extraction is performed on the input multi-source data, and data at different resolutions and timestamps are dynamically associated through the spatio-temporal attention module. For spatial data at different resolutions, the spatio-temporal attention mechanism calculates the attention weights according to the spatial position and distance to fuse the detailed features of high resolution and the global features of low resolution; for data at different timestamps, the spatio-temporal attention mechanism calculates the attention weights according to the time sequence and time interval to capture the change trend and correlation of the data over time. In the hierarchical encoder and decoder, the features are gradually abstracted and fused, and the spatio-temporal attention module of each layer further performs spatio-temporal correlation processing on the features output by the upper layer, and finally a fused feature vector is obtained at the output layer of the model. Fusion data generation: Convert the fused feature vectors output by the model into a format that matches the input data to generate a fused multi-source surveying and mapping dataset. The fused dataset contains the spatio-temporal information and features of multi-source data, which can more comprehensively reflect the actual situation of the surveyed area. III. Establishment and pruning of the MTL-DNN surveying and mapping data analysis model; (I) Construction of the MTL-DNN model; Network structure design; Shared layer design: The shared layer is located at the bottom of the MTL-DNN model and is used to extract the common features of multi-source surveying and mapping data, such as spatial features, temporal features, spectral features, etc. The shared layer can adopt structures such as fully connected layers, convolutional neural networks (CNNs), or recurrent neural networks (RNNs), and select a suitable network structure according to the characteristics of multi-source data. For satellite remote sensing images and UAV LiDAR point cloud data, CNN can be used for spatial feature extraction; for time series data of Internet of Things sensors, RNN or LSTM can be used for temporal feature extraction. Task-specific layer design: Design task-specific layers according to the different requirements of the three tasks of terrain modeling, object classification, and risk prediction. The task-specific layer is located above the shared layer, and each task corresponds to an independent sub-network, which is used to process the specific features of the task and output the task results. The sub-network for the terrain modeling task can adopt a fully connected layer or a 3D convolutional layer to generate a three-dimensional model of the terrain; the sub-network for the object classification task can adopt a Softmax layer to output the class probabilities of objects; the sub-network for the risk prediction task can adopt a regression layer to output the continuous values of the risk levels. Network connection method: The output of the shared layer serves as the input of the three task-specific layers at the same time. The task-specific layers are independent of each other, and only feature sharing is carried out at the shared layer to achieve parameter sharing and feature reuse of multi-task learning. Selection of activation function and loss function; Activation function: Use appropriate activation functions in the shared layer and task-specific layers, such as ReLU, Sigmoid, Tanh, etc. The ReLU function can alleviate the problem of gradient disappearance and improve the training speed of the model, and is commonly used in hidden layers; the Sigmoid function and Softmax function are commonly used in output layers for probability output of classification tasks. Loss function: Select the corresponding loss function according to different task types. The terrain modeling task belongs to a regression task and adopts the mean squared error (MSE) loss function; the object classification task belongs to a classification task and adopts the cross-entropy loss function; for the risk prediction task, according to the different prediction targets, the mean squared error loss function or the cross-entropy loss function can be adopted (when the risk level is divided into discrete categories). The total loss function of multi-task learning is the weighted sum of the loss functions of each task, and the weights are adjusted according to the importance and data scale of each task. (2) Model Pruning; Pruning Strategy Selection: Pruning Based on Weights: Calculate the absolute value of each weight in the network, set a pruning threshold, and set the weights with absolute values less than the threshold to zero to remove unimportant connections. This method is simple and effective and can reduce the number of model parameters without significantly affecting the model performance. Pruning Threshold Setting: Determine the appropriate pruning threshold through experiments and verification. During the pruning process, gradually increase the pruning ratio. After each pruning, use the validation set to evaluate the model performance. When the model performance drops to an acceptable range, stop pruning. For example, first perform 5% weight pruning and evaluate the model performance. If the performance drops by no more than 5%, continue to increase the pruning ratio until the performance drop exceeds the set threshold. Fine-tuning the Model after Pruning: Fine-tune the pruned model to recover the performance lost due to pruning. During fine-tuning, retrain the model using the training data, set the learning rate to 1 / 10 or lower than before pruning, and the number of training epochs to 10 - 20 epochs, so that the model adapts to the pruned network structure while retaining important connections and neurons, improving the model accuracy and stability. Obtaining the Target MTL-DNN Mapping Data Analysis Model: The model after pruning and fine-tuning is the target MTL-DNN mapping data analysis model. This model has fewer parameters and computational requirements while maintaining high performance, and is more suitable for deployment and operation on cloud or edge devices. IV. Mapping Data Analysis; (1) Model Performing Tasks; Terrain Modeling: The terrain modeling task sub-network of the model receives the fused feature vectors, extracts spatial and temporal features through the shared layer, and combines the processing of the task-specific layer to output the three-dimensional coordinates and height information of the terrain. Using this information, adopt three-dimensional modeling algorithms such as triangulation network construction and surface fitting to generate the three-dimensional terrain model of the mapping area. During the modeling process, the digital elevation model (DEM) and digital surface model (DSM) can be combined to improve the accuracy and detail performance of terrain modeling. Feature Classification: The feature classification task sub-network analyzes the fused features, extracts features such as spectrum, texture, and shape through the shared layer, and combines the classifier of the task-specific layer to output the class probability distribution of the features. According to the preset feature class labels, select the class with the highest probability as the classification result of the features, realizing the recognition and classification of different feature types such as buildings, vegetation, water bodies, and roads. Post-processing techniques such as morphological filtering and region growing can be used to optimize the classification results and improve the accuracy and consistency of classification. Risk prediction: The risk prediction task sub-network utilizes the time series features, spatial features, and ground object features in the fused data, combines historical risk data and relevant risk indicators (such as terrain slope, ground object type, rainfall, etc.), and outputs the risk level or the probability of risk occurrence through regression or classification algorithms. For geological hazard risk prediction (landslides, debris flows), a risk prediction model can be established, comprehensively considering various factors such as terrain, vegetation coverage, and hydrological conditions, to provide a basis for risk assessment and management of the surveyed area.

[0031] The above shows and describes 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 by the above embodiments. The above embodiments and the descriptions in the specification are only preferred examples of the present invention and are not used to limit the present invention. Without departing from the spirit and scope of the present invention, the present invention will have various changes and improvements, and these changes and improvements all fall within the scope of the present invention claimed. The scope of protection claimed by the present invention is defined by the appended claims and their equivalents.

Claims

1. A cloud-based mapping data processing method, characterized in that The cloud-based mapping data processing method includes the following steps: Collect satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocess the collected data to obtain a multi-source mapping data set, and transmit the multi-source mapping data set to the cloud platform; Based on the improved hierarchical Transformer model in the cloud platform, perform data fusion on the multi-source mapping data set, introduce a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain a fused multi-source mapping data set; Establish an MTL-DNN mapping data analysis model based on the MTL-DNN multi-task deep learning network, and use a fine-grained pruning tool to perform pruning operations on the model to obtain a target MTL-DNN mapping data analysis model; Input the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, feature classification, and risk prediction, and obtain mapping data analysis results.

2. The method for processing surveying and mapping data based on cloud according to claim 1, wherein The step of collecting satellite remote sensing data, UAV LiDAR data, and Internet of Things sensor data, preprocessing the collected data to obtain a multi-source mapping data set, and transmitting the multi-source mapping data set to the cloud platform includes: Collect satellite remote sensing data, and perform radiometric correction and geometric correction on the satellite remote sensing data based on a model of atmospheric transmission theory to obtain first mapping data; Obtain UAV LiDAR data, use a statistical filtering method to denoise the point cloud of the UAV LiDAR data, and perform coordinate transformation and point cloud resampling on the denoised data using linear interpolation to obtain second mapping data; Collect Internet of Things sensor data, use polynomial interpolation to process outliers in the Internet of Things sensor data, and synchronize the time of the data after outlier processing to obtain third mapping data; Perform normalization processing on the first mapping data, second mapping data, and third mapping data, and integrate the normalized data to obtain a multi-source mapping data set; Encrypt the multi-source mapping data set, and transmit the encrypted multi-source mapping data set to the cloud platform based on 5G communication.

3. A cloud-based mapping data processing method as claimed in claim 1, wherein, The step of performing data fusion on the multi-source mapping data set based on the improved hierarchical Transformer model in the cloud platform, introducing a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtaining a fused multi-source mapping data set includes: Establish a hierarchical Transformer model in the cloud platform using a hierarchical structure of multi-layer Transformer encoders and decoders; Each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, an encoder-decoder attention mechanism is introduced to obtain an improved hierarchical Transformer model; Add a spatio-temporal attention module to the encoder and decoder of the improved hierarchical Transformer model, and use the spatio-temporal attention module to process the time dimension and space dimension information of the data respectively.

4. The method for processing mapping data based on cloud as claimed in claim 3, wherein, The improved hierarchical Transformer model based on the cloud platform is used to perform data fusion on the multi-source mapping data set, introducing a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, obtaining a fused multi-source mapping data set, and further including: Obtain the multi-source mapping data set in the cloud platform, and input the multi-source mapping data set into the improved hierarchical Transformer model; Use the underlying encoder of the model to perform preliminary feature extraction on the input multi-source data, and dynamically associate data with different resolutions and timestamps through the spatio-temporal attention module; Use the spatio-temporal attention mechanism to calculate the attention weights according to the spatial position and distance, and fuse the detailed features of high resolution and the global features of low resolution to obtain the fused multi-source mapping data set.

5. A cloud-based mapping data processing method as claimed in claim 1, wherein The MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network, and the model is pruned by a fine-grained pruning tool to obtain the target MTL-DNN mapping data analysis model, including: Establish the MTL-DNN mapping data analysis model based on the MTL-DNN multi-task deep learning network. The shared layer is located at the bottom of the MTL-DNN model and is used to extract the common features of the data; Design task-specific layers according to the different requirements of the three tasks of terrain modeling, feature classification and risk prediction; The output of the shared layer is used as the input of the three task-specific layers at the same time. The task-specific layers are independent of each other and only share features at the shared layer.

6. The method for processing surveying and mapping data based on cloud as claimed in claim 5, wherein, The MTL-DNN mapping data analysis model is established based on the MTL-DNN multi-task deep learning network, and the model is pruned by a fine-grained pruning tool to obtain the target MTL-DNN mapping data analysis model, and further including: Use the Adam optimizer to set the learning rate, momentum and weight decay parameters of the MTL-DNN mapping data analysis model; Calculate the absolute value of each weight in the network based on the fine-grained pruning tool, set a pruning threshold, set the weights with absolute values less than the threshold to zero, remove the unimportant connections, and obtain the target MTL-DNN mapping data analysis model.

7. The method for processing surveying and mapping data based on cloud as claimed in claim 1, wherein Input the fused multi-source mapping data set into the target MTL-DNN mapping data analysis model to perform terrain modeling, feature classification and risk prediction, and obtain the mapping data analysis result, including: Extract the spatial and temporal features through the shared layer, and combine the processing of the task-specific layer to output the three-dimensional coordinates and height information of the terrain to obtain terrain modeling; Extract features such as spectrum, texture, and shape through the shared layer, and combine the classifier of the task-specific layer to output the category probability distribution of the features to obtain feature classification; Fuse the time series features, spatial features and feature features in the data, combine the historical risk data and relevant risk indicators, and output the risk level or the probability of risk occurrence through the classification algorithm to obtain risk prediction.

8. A cloud-based mapping data processing system, characterized in that, The cloud-based mapping data processing system includes the following modules: The data acquisition and transmission module is used to collect satellite remote sensing data, UAV LiDAR data and Internet of Things sensor data, preprocess the collected data to obtain a multi-source surveying and mapping data set, and transmit the multi-source surveying and mapping data set to the cloud platform; The surveying and mapping data fusion module is used to perform data fusion on the multi-source surveying and mapping data set based on the improved hierarchical Transformer model in the cloud platform, introduce a spatio-temporal attention mechanism to dynamically associate data with different resolutions and timestamps, and obtain a fused multi-source surveying and mapping data set; The analysis model establishment module is used to establish an MTL-DNN surveying and mapping data analysis model based on the MTL-DNN multi-task deep learning network, and use a pruning tool to perform pruning operations on the model to obtain a target MTL-DNN surveying and mapping data analysis model; The surveying and mapping data analysis module is used to input the fused multi-source surveying and mapping data set into the target MTL-DNN surveying and mapping data analysis model to perform terrain modeling, feature classification and risk prediction, and obtain surveying and mapping data analysis results.

9. A cloud-based mapping data processing system according to claim 8, characterized in that, The surveying and mapping data fusion module includes the following sub-modules: The establishment sub-module is used to establish a hierarchical Transformer model in the cloud platform using the hierarchical structure of multi-layer Transformer encoders and decoders; The introduction sub-module is such that each layer of the encoder in the model consists of a multi-head self-attention mechanism and a feed-forward neural network, and the decoder consists of a multi-head self-attention mechanism and a feed-forward neural network. At the same time, an encoder-decoder attention mechanism is introduced to obtain an improved hierarchical Transformer model; The addition sub-module is used to add a spatio-temporal attention module to the encoder and decoder of the improved hierarchical Transformer model, and use the spatio-temporal attention module to process the time dimension and space dimension information of the data respectively.

10. A cloud-based mapping data processing system according to claim 8, characterized in that, The surveying and mapping data fusion module also includes the following sub-modules: The input sub-module is used to obtain the multi-source surveying and mapping data set in the cloud platform and input the multi-source surveying and mapping data set into the improved hierarchical Transformer model; The extraction sub-module is used to perform preliminary feature extraction on the input multi-source data using the underlying encoder of the model, and dynamically associate data with different resolutions and timestamps through the spatio-temporal attention module; The fusion sub-module is used to calculate attention weights according to spatial positions and distances using the spatio-temporal attention mechanism, and fuse the high-resolution detailed features and low-resolution global features to obtain a fused multi-source surveying and mapping data set.

Images