Unmanned aerial vehicle airborne geophysical prospecting method and system based on big data

Abstract

The invention discloses an unmanned aerial vehicle airborne geophysical prospecting method and system based on big data, and the method comprises the steps: obtaining the positioning data of a global navigation satellite system and the measurement data of an inertial measurement unit, carrying out the correction and compensation through the combination of a differential positioning technology and a terrain height model, and obtaining an accurate three-dimensional coordinate data set; acquiring a magnetic field signal and a spectral signal according to the three-dimensional coordinate data set, and performing de-noising processing on the magnetic field signal and the spectral signal by adopting an adaptive filtering algorithm to obtain preprocessed signal data; judging whether the signal data volume of the preprocessed signal data exceeds a preset threshold value or not, if the signal data volume exceeds the preset threshold value, performing fragmentation processing on the preprocessed signal data through a distributed computing framework, extracting feature vectors in each piece of fragmented data by adopting a convolutional neural network, and generating a feature vector set; according to the invention, full-link optimization from data acquisition to ore body prediction is realized, and the precision and efficiency of geological exploration are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of unmanned aerial vehicle airborne geophysical prospecting, and particularly discloses an unmanned aerial vehicle airborne geophysical prospecting method and system based on big data. BACKGROUND

[0002] As an important means of modern mineral resource exploration, unmanned aerial vehicle airborne geophysical prospecting technology can efficiently obtain surface and underground resource information by integrating high-precision detection equipment and intelligent flight platforms, which has key significance for promoting the digital transformation and green exploration of the mining industry. This technology uses unmanned aerial vehicles to carry multiple sensors and combines intelligent data processing to significantly improve the efficiency and accuracy of exploration, and has been widely applied in iron ore, polymetallic ore and other resource exploration.

[0003] However, the existing exploration methods still have obvious limitations. Traditional ground exploration is limited by terrain, low efficiency and high cost, especially in complex environments such as deserts and high mountains; although conventional airborne geophysical prospecting has a wide coverage, the flight height is relatively high, and the resolution is insufficient, making it difficult to capture weak signals of deep ore bodies. These limitations reveal the core challenges in the field of unmanned aerial vehicle airborne geophysical prospecting.

[0004] Firstly, the high-precision positioning requirement under low-altitude flight becomes a technical bottleneck. The unmanned aerial vehicle needs to maintain ultra-low altitude flight of 50-150 meters in complex terrain while ensuring centimeter-level positioning accuracy to capture weak magnetic field or spectral signals, and any positioning deviation may lead to data distortion.

[0005] Secondly, the realization of high-precision positioning further deduces the problem of massive data processing. The data generated by ultra-low altitude flight is huge, and traditional processing methods are difficult to analyze and extract weak abnormal signals in real time, which limits the exploration efficiency.

[0006] In addition, the lack of data processing capacity also derives the urgent need for intelligent analysis, which needs to quickly filter out noise and build accurate geological models through advanced algorithms.

[0007] Therefore, how to realize centimeter-level positioning accuracy in ultra-low altitude flight, and develop efficient real-time data processing and intelligent analysis technology to ensure accurate capture of weak signals and rapid construction of geological models, has become a key problem to be solved in unmanned aerial vehicle airborne geophysical prospecting technology. SUMMARY

[0008] The present application provides an unmanned aerial vehicle airborne geophysical prospecting method and system based on big data, which aims to solve at least one of the defects in the prior art.

[0009] One aspect of the present application relates to an unmanned aerial vehicle airborne geophysical prospecting method based on big data, comprising the following steps: Obtain positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correct the positioning data through a differential positioning technology, combine a terrain height model to compensate the corrected data in height, and obtain a three-dimensional coordinate data set; According to the three-dimensional coordinate data set, collect magnetic field signals and spectrum signals, and use an adaptive filtering algorithm to denoise the magnetic field signals and spectrum signals to obtain preprocessed signal data; Determine whether the signal data quantity of the preprocessed signal data exceeds a preset threshold value, if the signal data quantity exceeds the preset threshold value, perform sharding processing on the preprocessed signal data through a distributed computing framework, use a convolutional neural network to extract feature vectors in each shard data, and generate a feature vector set; Input the feature vector set into a random forest classification model to obtain a geological anomaly classification result, combine pattern data in a ore body distribution knowledge base to determine a geological anomaly distribution atlas; According to the geological anomaly distribution atlas, perform three-dimensional inversion calculation on the magnetic field signals and spectrum signals to establish a spatial distribution model of underground ore bodies and generate three-dimensional geological structure data; Spatially register the three-dimensional geological structure data and the three-dimensional coordinate data set to output a three-dimensional geological model containing ore body distribution characteristics.

[0010] Further, the step of obtaining positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through a differential positioning technology, combining a terrain height model to compensate the corrected data in height, and obtaining a three-dimensional coordinate data set includes: Obtain positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, align the positioning data and the measurement data in time axis through a time synchronization algorithm, use Kalman filtering to preprocess the measurement data to obtain preprocessed data; Use a differential positioning technology to correct errors of the preprocessed data, generate correction data through differential calculation of a reference station and a mobile station, combine a coordinate conversion algorithm to unify the correction data to a preset reference coordinate system, and obtain corrected two-dimensional coordinate data; According to the corrected two-dimensional coordinate data, obtain corresponding digital elevation information from a pre-established terrain height model, adjust the height component of the corrected two-dimensional coordinate data through a height compensation algorithm, and obtain a preliminary three-dimensional coordinate data set; If the accuracy of the preliminary three-dimensional coordinate data set is lower than a preset threshold value, perform secondary fusion on the corrected positioning data and the measurement data through a data fusion algorithm, use a weighted average method to optimize the preliminary three-dimensional coordinate data set, and obtain a final three-dimensional coordinate data set.

[0011] Further, according to the three-dimensional coordinate data set, the magnetic field signal and the spectrum signal are collected, the adaptive filtering algorithm is used for denoising processing of the magnetic field signal and the spectrum signal, and the step of obtaining the preprocessed signal data includes: The magnetic field sensor and the spectrum sensor are calibrated through a sensor calibration algorithm, preset calibration parameters are used for deviation correction of sensor output, and calibrated magnetic field signals and spectrum signals are obtained; According to the calibrated magnetic field signal and the spectrum signal, the signal acquisition module synchronously acquires data within a preset time interval, and a time stamp alignment algorithm is used to ensure consistency with the time axis of the three-dimensional coordinate data set, to obtain an original signal data set; For the original signal data set, an adaptive filtering algorithm is used for denoising processing of the magnetic field signal and the spectrum signal, and environmental interference is removed by dynamically adjusting the filtering parameters, to obtain a preprocessed signal data set; If the signal-to-noise ratio of the preprocessed signal data set is lower than a preset threshold, then through a data fusion algorithm, the preprocessed signal data set is associated with the three-dimensional coordinate data set, a weighted average method is used to optimize the correspondence between the signal and the coordinate, and a preprocessed signal data is obtained.

[0012] Further, it is judged whether the signal data amount of the preprocessed signal data exceeds a preset threshold, and if the signal data amount exceeds the preset threshold, then through a distributed computing framework, the preprocessed signal data is processed in slices, and a convolutional neural network is used to extract feature vectors in each slice data, and the step of generating a feature vector set includes: If the signal data amount exceeds the preset threshold, then through a distributed computing framework, the preprocessed signal data is processed in slices, and according to the data dimension, a hash mapping algorithm is used to distribute to a distributed node, to obtain a slice data set; For the slice data set, a convolutional neural network is used to extract features of each slice data through a convolutional layer and a pooling layer, to generate a feature vector of each slice data, and to obtain a feature vector set.

[0013] Further, the feature vector set is input into a random forest classification model to obtain a geological anomaly classification result, and the step of determining a geological anomaly distribution atlas in combination with pattern data in a ore body distribution knowledge base includes: Classification input data is obtained from the feature vector set, a random forest classification model is used for classification processing of the feature vector, each feature vector is voted through traversal of a decision tree set, and a classification result is obtained; According to the classification result, pattern data is obtained from a pre-established ore body distribution knowledge base, a cosine similarity algorithm is used to calculate the similarity value between the classification result and the pattern data, and matched pattern data is determined; If the similarity value of the matched pattern data and the classification result exceeds a preset threshold value, the classification result is mapped to a two-dimensional spatial coordinate through a grid division algorithm to generate an initial geological anomaly distribution map; According to the initial geological anomaly distribution map, an interpolation algorithm is used to smooth the two-dimensional spatial coordinate, and combined with the spatial distribution characteristics of the pattern data, a final geological anomaly distribution atlas is generated.

[0014] Further, according to the geological anomaly distribution atlas, a three-dimensional inversion calculation is performed on the magnetic field signal and the spectrum signal to establish a spatial distribution model of the underground ore body, and the steps of generating three-dimensional geological structure data include: The magnetic field signal and the spectrum signal data are obtained from the geological anomaly distribution atlas, and a data fusion algorithm is used to process the magnetic field signal and the spectrum signal through a weighted average method to obtain a fusion signal data set; According to the fusion signal data set, a three-dimensional inversion algorithm is used to calculate the fusion signal data set through an iterative least squares method to determine preliminary spatial distribution data; If the inversion accuracy of the preliminary spatial distribution data is lower than a preset threshold value, the preliminary spatial distribution data is subjected to spatial discretization processing through a grid division algorithm, and the discretized data is subjected to feature enhancement through a stereoscopic microscopy technique to obtain optimized spatial distribution data; According to the optimized spatial distribution data, a coordinate mapping algorithm is used to smooth the optimized spatial distribution data through an interpolation algorithm to generate three-dimensional geological structure data.

[0015] Further, the three-dimensional geological structure data and the three-dimensional coordinate data set are spatially registered, and the steps of outputting a three-dimensional geological model containing ore body distribution characteristics include: The spatial position information is obtained from the three-dimensional geological structure data and the three-dimensional coordinate data set, and a coordinate mapping algorithm is used to preliminarily align the three-dimensional geological structure data and the three-dimensional coordinate data set through a linear transformation method to obtain a preliminary registration data set; According to the preliminary registration data set, a grid division algorithm is used to discretize the preliminary registration data set through a uniform grid segmentation method to obtain a discretized registration data set; If the registration error of the discretized registration data set is higher than a preset threshold value, a support vector machine algorithm is used to enhance the features of the discretized registration data set through a classification method to obtain an optimized registration data set; According to the optimized registration data set, an interpolation smoothing algorithm is used to smooth the optimized registration data set through a spline interpolation method to generate a three-dimensional geological model containing ore body distribution characteristics.

[0016] Another aspect of the present application relates to a big data-based unmanned aerial vehicle airborne geophysical exploration system for implementing the above-mentioned big data-based unmanned aerial vehicle airborne geophysical exploration method, the big data-based unmanned aerial vehicle airborne geophysical exploration system comprising: A first acquisition module is configured to acquire positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correct the positioning data by using a differential positioning technology, combine a terrain height model to compensate the corrected data in height, and obtain a three-dimensional coordinate data set; A second acquisition module is configured to acquire magnetic field signals and spectrum signals according to the three-dimensional coordinate data set, and perform denoising processing on the magnetic field signals and the spectrum signals by using an adaptive filtering algorithm to obtain preprocessed signal data; A judgment module is configured to judge whether the signal data amount of the preprocessed signal data exceeds a preset threshold value, and if the signal data amount exceeds the preset threshold value, perform sharding processing on the preprocessed signal data by using a distributed computing framework, extract feature vectors in each sharded data by using a convolutional neural network, and generate a feature vector set; A determination module is configured to input the feature vector set into a random forest classification model, acquire a geological anomaly classification result, combine pattern data in a ore body distribution knowledge base, and determine a geological anomaly distribution atlas; A generation module is configured to perform three-dimensional inversion calculation on the magnetic field signals and the spectrum signals according to the geological anomaly distribution atlas, establish a spatial distribution model of underground ore bodies, and generate three-dimensional geological structure data; An output module is configured to perform spatial registration on the three-dimensional geological structure data and the three-dimensional coordinate data set, and output a three-dimensional geological model containing ore body distribution characteristics.

[0017] Further, the first acquisition module comprises: A first acquisition unit is configured to acquire positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, perform time axis alignment on the positioning data and the measurement data by using a time synchronization algorithm, and perform preprocessing on the measurement data by using Kalman filtering to obtain preprocessed data; A second acquisition unit is configured to perform error correction on the preprocessed data by using a differential positioning technology, generate correction data by differential calculation of a reference station and a mobile station, and unify the correction data to a preset reference coordinate system by using a coordinate conversion algorithm to obtain corrected two-dimensional coordinate data; A third acquisition unit is configured to acquire corresponding digital elevation information from a pre-established terrain height model according to the corrected two-dimensional coordinate data, adjust the height component of the corrected two-dimensional coordinate data by using a height compensation algorithm to obtain a preliminary three-dimensional coordinate data set; The fourth acquisition unit is configured to, if the accuracy of the preliminary three-dimensional coordinate data set is lower than a preset threshold, perform secondary fusion on the corrected positioning data and the measurement data by using a data fusion algorithm, optimize the preliminary three-dimensional coordinate data set by using a weighted average method, and obtain a final three-dimensional coordinate data set.

[0018] Further, the second acquisition module comprises: The fifth acquisition unit is configured to calibrate the magnetic field sensor and the spectrum sensor by using a sensor calibration algorithm, correct the deviation of the sensor output by using preset calibration parameters, obtain the calibrated magnetic field signal and spectrum signal, and output the calibrated magnetic field signal and spectrum signal. The sixth acquisition unit is configured to, according to the calibrated magnetic field signal and spectrum signal, synchronize the data collected by the signal acquisition module within a preset time interval, ensure the time axis of the three-dimensional coordinate data set by using a time stamp alignment algorithm, and obtain an original signal data set. The seventh acquisition unit is configured to, for the original signal data set, perform denoising processing on the magnetic field signal and the spectrum signal by using an adaptive filtering algorithm, remove the environmental interference by dynamically adjusting the filtering parameters, and obtain a preprocessed signal data set. The eighth acquisition unit is configured to, if the signal-to-noise ratio of the preprocessed signal data set is lower than a preset threshold, perform coordinate correlation on the preprocessed signal data set and the three-dimensional coordinate data set by using a data fusion algorithm, optimize the corresponding relationship between the signal and the coordinate by using a weighted average method, and obtain the preprocessed signal data.

[0019] The present application has the following beneficial effects: The present application provides a large data-based unmanned aerial vehicle airborne geophysical prospecting method and system, which obtains data of a global navigation satellite system and an inertial measurement unit, corrects and compensates by combining a differential positioning technology and a terrain height model, and obtains an accurate three-dimensional coordinate data set. On this basis, magnetic field and spectrum signals are collected and adaptive filtering denoising is performed. For large-scale data, distributed computing and convolutional neural network are used to extract features, and a random forest classification model is used in combination with a mineral body distribution knowledge base to generate a geological anomaly distribution atlas. Finally, an underground mineral body spatial distribution model is established by three-dimensional inversion calculation, and is registered with the three-dimensional coordinate data set to output a three-dimensional geological model containing the distribution characteristics of the mineral body. BRIEF DESCRIPTION OF DRAWINGS

[0020] Figure 1 The figure is a flowchart of an embodiment of the large data-based unmanned aerial vehicle airborne geophysical prospecting method of the present application. DETAILED DESCRIPTION

[0021] In order to better understand the above technical solutions, the above technical solutions will be described in detail below in combination with the drawings in the specification and specific embodiments.

[0022] AsFigure 1 As shown, the first embodiment of the present invention proposes a UAV aerial geophysical exploration method based on big data, comprising the following steps: Step S100: Acquire positioning data from a global navigation satellite system and measurement data from an inertial measurement unit, correct the positioning data using differential positioning technology, and perform height compensation on the corrected data using a terrain height model to obtain a three-dimensional coordinate data set.

[0023] Differential positioning is a satellite navigation technology that uses real-time or post-processing error correction algorithms to improve positioning accuracy through the collaborative work of a base station and a rover. Its core principle is to calculate satellite signal errors at the base station (with known precise coordinates) and transmit the error correction values ​​to the rover (user end). This eliminates interference factors such as atmospheric delay and satellite clock errors, achieving centimeter-level positioning accuracy.

[0024] Altitude compensation of corrected data is a technical process that, after basic data calibration has been completed, uses mathematical models or empirical parameters to perform secondary corrections to measurements to address systematic errors caused by spatial elevation differences. Its core goal is to eliminate observational data bias caused by terrain fluctuations or differences in equipment installation height, ensuring the accuracy and comparability of environmental parameters across regions and gradients.

[0025] A 3D coordinate dataset is a structured data set that records the location information of objects, point clouds, or environmental features based on a 3D spatial coordinate system. It typically contains numerical coordinates in the X, Y, and Z dimensions and can be labeled with attributes (such as color, temperature, and pressure) to enable multidimensional spatial analysis. Essentially, it builds digital twin models of physical or virtual scenes through the mapping of discrete or continuous spatial data.

[0026] Step S200: According to the three-dimensional coordinate data set, magnetic field signals and spectral signals are collected, and adaptive filtering algorithms are used to perform denoising on the magnetic field signals and spectral signals to obtain pre-processed signal data.

[0027] Adaptive filtering is an intelligent computing method that adjusts filter parameters in real time based on the statistical characteristics of the input signal to dynamically optimize signal processing effects (such as noise suppression, signal separation, or system modeling). Its core is to use feedback mechanisms and iterative optimization to enable the filter to maintain optimal estimation or feature extraction capabilities of the target signal even in non-stationary environments.

[0028] De-noising is a process of identifying and eliminating noise interference from signals, data or images while preserving the original effective information as much as possible through algorithms or technical means. Its core goal is to suppress or remove non-target components such as random noise, environmental interference and device errors, improve the signal-to-noise ratio (SNR) and usability of data, and is widely used in communication, medical imaging, speech processing, data science and other fields.

[0029] Step S300, determine whether the signal data quantity of the pre-processed signal data exceeds the preset threshold. If the signal data quantity exceeds the preset threshold, perform sharding processing on the pre-processed signal data through a distributed computing framework, extract feature vectors in each sharded data using a convolutional neural network, and generate a feature vector set.

[0030] Distributed computing framework is a software architecture and technical system that supports collaborative execution of computing tasks on multiple independent computers (nodes). Through task splitting, parallel processing and data distribution mechanisms, it realizes a systematic solution for large-scale data processing, complex model training or real-time resource scheduling. Its core is a node cluster interconnected by network, providing high throughput, fault tolerance and horizontal expansion capability to meet the computing needs of massive data or high concurrency scenarios.

[0031] Sharding is a technique that divides large-scale data sets or system resources into multiple independent fragments (shards) for distributed storage, computation or management. Its core goal is to alleviate single-point performance bottlenecks, improve system throughput, reduce latency and enhance fault tolerance through horizontal expansion.

[0032] Convolutional Neural Network (CNN) is a deep learning model designed for processing grid topology data such as images, audio and video. Its core is through convolution operation, pooling operation and hierarchical feature extraction mechanism to realize local correlation modeling and efficient representation learning of input data. Convolutional Neural Network is a cornerstone model in computer vision and medical image analysis.

[0033] Feature vector set (feature space) is a linear subspace composed of all eigenvectors corresponding to a certain eigenvalue of a linear transformation or matrix and zero vectors.

[0034] Step S400, input the feature vector set into the random forest classification model to obtain the geological anomaly classification result, combine the pattern data in the ore body distribution knowledge base to determine the geological anomaly distribution atlas.

[0035] Random Forest is a supervised learning algorithm based on Ensemble Learning, which improves the classification accuracy and generalization ability by constructing multiple decision trees and voting (or averaging) their prediction results. The core idea is to combine "Bootstrap resampling" and "feature random selection" strategies to generate a diverse group of decision trees, reducing the risk of overfitting of a single decision tree, and outputting the final classification result through majority voting mechanism.

[0036] Geological anomaly classification results are based on specific classification criteria (such as genesis, size, and hosting medium) to systematically divide and classify geological bodies or geological processes with significant differences, aiming to reveal the genetic mechanism, spatial distribution pattern, and resource and environmental effects of geological anomalies. The core is to divide geological anomalies into categories with similar attributes through multi-dimensional feature analysis and threshold definition, providing scientific basis for mineral prediction, disaster prevention and control, and environmental evaluation.

[0037] The ore body distribution knowledge base is a structured data system integrating spatial distribution characteristics, attribute parameters, and geological correlation information of ore bodies, aiming to provide a standardized and interactive digital support platform for mineral resource exploration, development decision-making, and geological research through multi-source data fusion and intelligent analysis technology. The core is to convert key information such as ore body geometry, hosting environment, and genetic mechanism into quantifiable and searchable knowledge units through a unified data framework, realizing the visualization and dynamic updating of ore body distribution patterns.

[0038] Pattern data is an abstract representation of regularity and repeatability extracted from complex data sets, usually represented as mathematical descriptions of specific rules, structures, or models, used to represent the internal correlation or behavior trend of data. Its essence is to compress and generalize the original data through algorithms or statistical methods to form interpretable and reusable knowledge units to support classification, prediction, clustering, and other analysis tasks.

[0039] The geological anomaly distribution map is a systematic map that expresses the spatial distribution, intensity level, and genetic correlation of geological anomalies based on spatial geographic coordinates, integrating geological, geophysical, and geochemical multi-source anomaly data through visualization technology. The core is to reveal the macroscopic distribution pattern of geological anomalies and their internal relationship with mineralization, disaster risk, or environmental problems through layer superposition and comprehensive analysis.

[0040] Step S500, according to the geological anomaly distribution map, three-dimensional inversion calculation is performed on the magnetic field signal and the spectrum signal, and a spatial distribution model of the underground ore body is established, and three-dimensional geological structure data is generated.

[0041] Three-dimensional inversion calculation is a numerical simulation technique that inversely calculates the distribution of physical property parameters (such as resistivity, magnetic susceptibility, and density) of target geological bodies in three-dimensional space based on physical field data (such as electromagnetic, gravity, and magnetic methods) observed on the ground or underground, combined with geological models and mathematical optimization algorithms. The core is to minimize the difference between measured data and theoretical forward response through nonlinear iterative optimization, and finally build a three-dimensional physical property model consistent with the true geological structure.

[0042] The spatial distribution model of underground ore bodies is a digital three-dimensional model constructed by integrating geological exploration data (such as drilling, geophysical prospecting, and geochemical prospecting), geophysical parameters, and ore-forming geological conditions, used to quantitatively represent the geometric shape, occurrence characteristics, physical and chemical property distribution of ore bodies in three-dimensional space, and their relationship with surrounding rock structures. The spatial distribution model of underground ore bodies uses computers as carriers, adopts mathematical algorithms and visualization techniques, and dynamically displays the spatial distribution law of ore bodies, providing core basis for mineral exploration, reserve estimation, and mining design.

[0043] Three-dimensional geological structure data is a digital information set that represents the geometric shape (such as rock layer interface and fault zone distribution), physical property parameters (such as density, resistivity, and magnetic susceptibility), and topological relationships of underground geological bodies based on three-dimensional spatial coordinates (X, Y, Z). Its core is to integrate multiple data sources (such as drilling, geophysical prospecting, and remote sensing) to build a computer model that can describe the spatial distribution law and internal attribute variation of geological bodies, providing quantitative basis for geological analysis, resource exploration, and engineering decision-making.

[0044] Step S600, spatially register the three-dimensional geological structure data with the three-dimensional coordinate data set, and output a three-dimensional geological model containing ore body distribution characteristics.

[0045] Spatial registration is a technical process that unifies multi-source spatial data (such as remote sensing images, three-dimensional point clouds, and geological models) from different sensors, different times, or different coordinate systems through geometric transformation and algorithm optimization. The core is to eliminate the inconsistency and superimposability of multi-source data in position, direction, and scale through coordinate alignment, scale correction, and geometric deformation, supporting data fusion and comprehensive analysis.

[0046] The three-dimensional geological model containing the distribution characteristics of the ore body is a computer model for quantitatively representing the spatial distribution law of the ore body and the relationship between the ore body and the surrounding rock structure in a three-dimensional digital form by fusing geological exploration data (such as drilling, geophysical prospecting, and geochemical prospecting), ore body geometric parameters (shape and occurrence), and mineralization attributes (grade and lithology). The three-dimensional geological model is a framework based on spatial coordinates (X, Y, Z), combined with geological laws and mathematical algorithms, to dynamically reveal the ore body size, extension direction, thickness variation, and resource potential, and to provide visual decision support for mineral exploration, reserve evaluation, and mining design.

[0047] Further, the unmanned aerial vehicle airborne geophysical prospecting method based on big data provided in the embodiment comprises the following steps: In step S110, positioning data of a global navigation satellite system and measurement data of an inertial measurement unit are obtained, the positioning data and the measurement data are time-axis aligned through a time synchronization algorithm, and the measurement data is preprocessed through Kalman filtering to obtain preprocessed data.

[0048] Data fusion of a global navigation satellite system (GNSS) and an inertial measurement unit (IMU) is a common technology for high-precision positioning, widely used in unmanned driving, unmanned aerial vehicle navigation and other fields. Taking the unmanned aerial vehicle navigation scene as an example, the technical theme of the above content is analyzed in detail, and the implementation method and beneficial effects are described through specific embodiments. For example, the GNSS receiver obtains the positioning data of the unmanned aerial vehicle, including latitude, longitude and time stamp, with an accuracy of about 2 meters. The IMU provides acceleration and angular velocity data with a sampling frequency of 100 Hz.

[0049] The time synchronization algorithm ensures that the two sets of data are aligned on the time axis. In one possible implementation, the network time protocol (NTP) is used to calibrate the time stamps of GNSS and IMU, with an accuracy of milliseconds.

[0050] Suppose the GNSS data time stamp is T1=10:00:00.000 and the IMU data is T2=10:00:00.005. Through linear interpolation, the IMU data is aligned to the T1 moment, eliminating the time deviation and improving the subsequent fusion accuracy. Kalman filtering is used to preprocess the IMU data to eliminate noise. For example, the acceleration data of the IMU may have random drift due to vibration, and Kalman filtering estimates the true state through prediction and update steps, outputs smooth acceleration values, and reduces the error from 0.5 m / s² to 0.1 m / s². This preprocessing enhances the reliability of the data and lays the foundation for subsequent correction.

[0051] In step S120, differential positioning technology is used to correct the error of the preprocessed data, correction data is generated through differential calculation between the reference station and the mobile station, the correction data is unified to the preset reference coordinate system through the coordinate conversion algorithm, and the corrected two-dimensional coordinate data is obtained.

[0052] Differential positioning technology corrects GNSS errors through the cooperation of reference stations and mobile stations. In one embodiment, the reference station is located at a known coordinate (X0, Y0), receives the same satellite signals, and calculates the pseudo-range error.

[0053] The mobile station (drone) receives the correction value broadcast by the reference station and adjusts the GNSS data in real time. For example, the original positioning data is (X1, Y1), the error is 2 meters, and after applying differential correction, the coordinate error is reduced to 0.2 meters.

[0054] The correction data is unified to the WGS84 reference coordinate system through a coordinate conversion algorithm to ensure the consistency of coordinates from different data sources.

[0055] Step S130, according to the corrected two-dimensional coordinate data, the corresponding digital elevation information is obtained from the pre-established terrain height model, and the height component of the corrected two-dimensional coordinate data is adjusted through a height compensation algorithm to obtain a preliminary three-dimensional coordinate data set.

[0056] According to the corrected two-dimensional coordinate (X, Y), the elevation Z of the corresponding point is obtained from the digital elevation model (DEM). For example, the coordinate (X1, Y1) corresponds to a DEM elevation of 50 meters.

[0057] The height compensation algorithm adjusts the Z value, taking into account the terrain undulations and GNSS height errors, to obtain a preliminary three-dimensional coordinate (X, Y, Z).

[0058] Step S140, if the accuracy of the preliminary three-dimensional coordinate data set is lower than the preset threshold, then the corrected positioning data and the measured data are fused again through a data fusion algorithm, and a weighted average method is used to optimize the preliminary three-dimensional coordinate data set to obtain a final three-dimensional coordinate data set.

[0059] The accuracy of the preliminary three-dimensional coordinate data set is: (1) In formula (1), represents the accuracy evaluation value of the preliminary three-dimensional coordinate data set, represents the total number of coordinate points, 、 、 represents the three-dimensional coordinate component of the th measurement point, 、 、 represents the corresponding reference coordinate value, and formula (1) is used to calculate the average deviation between the coordinate data set and the reference value to evaluate whether the accuracy is lower than the preset threshold.

[0060] The final three-dimensional coordinate vector obtained after data fusion is: (2) In formula (2), Represents the final three-dimensional coordinate vector obtained after data fusion, represents the corrected positioning data coordinate vector, represents the measurement data coordinate vector, represents the weight coefficient for correcting the positioning data, represents the weight coefficient of the measurement data. Formula (2) uses the weighted average method to achieve the fusion optimization of the two data sources of corrected positioning data and measurement data.

[0061] The normalized weight value of the data source is: (3) In formula (3), represents the normalized weight value of the k-th data source, Indicates the The measurement uncertainty or error standard deviation of each data source, Indicates the total number of data sources involved in the fusion, Indicates the The error standard deviation of each data source is used. Formula (3) assigns corresponding weights according to the accuracy level of each data source. The higher the accuracy of the data source, the greater the weight it obtains.

[0062] Assuming the initial coordinate accuracy is 0.3 meters, which does not reach the preset threshold of 0.1 meters, the data fusion algorithm is activated. Preferably, a weighted average method is used, with a weight of 0.7 for GNSS data and a weight of 0.3 for IMU data. After fusion, the final three-dimensional coordinates (X', Y', Z') are generated with an accuracy of 0.08 meters.

[0063] This method significantly improves positioning accuracy, particularly in environments with weak GNSS signals, such as urban canyons and tunnels, where IMU data complements GNSS. The resulting 3D coordinate dataset provides a reliable navigation basis for drone aviation, ensuring accurate path planning and obstacle avoidance, and enhancing system safety and stability.

[0064] Furthermore, in the big data-based UAV geophysical prospecting method provided in this embodiment, step S200 includes: Step S210: calibrate the magnetic field sensor and the spectral sensor using a sensor calibration algorithm, use preset calibration parameters to perform deviation correction on the sensor output, and obtain calibrated magnetic field signals and spectral signals.

[0065] In the UAV aerial navigation scenario, the sensor calibration algorithm is used to ensure the accuracy of the magnetic field sensor and the spectral sensor data. The magnetic field sensor detects the geomagnetic field direction to assist the UAV attitude estimation; the spectral sensor collects the environmental spectral characteristics to assist the environmental perception. The calibration algorithm corrects the sensor bias by preset parameters. For example, the magnetic field sensor may be disturbed by the metal structure of the UAV, and the output deviates from the true geomagnetic field direction by 5 degrees. During calibration, the known geomagnetic field data is used as a reference to adjust the sensor output, and the bias is reduced to 0.5 degrees.

[0066] The spectral sensor may have wavelength shift due to changes in light. By calibrating the spectral curve in the laboratory, the wavelength error is reduced from 10 nanometers to 2 nanometers after correction. The calibrated signal is more reliable and provides high-quality data for subsequent processing.

[0067] Step S220, according to the calibrated magnetic field signal and the spectral signal, the signal acquisition module synchronously acquires data in a preset time interval, and the time stamp alignment algorithm ensures the time axis consistent with the three-dimensional coordinate data set, to obtain the original signal data set.

[0068] The signal acquisition module synchronously acquires the magnetic field signal and the spectral signal at intervals of 10 milliseconds. The time stamp alignment algorithm ensures the data synchronization with the three-dimensional coordinate data set. For example, the coordinate data time stamp is 10:00:00.000, the magnetic field signal time stamp is 10:00:00.002, and the linear interpolation is used to align to 10:00:00.000 to eliminate the time bias. The synchronized original signal data set contains the magnetic field intensity, direction and spectral characteristics, and forms a unified time axis with the coordinate data, which is convenient for subsequent fusion.

[0069] Step S230, for the original signal data set, an adaptive filtering algorithm is used to denoise the magnetic field signal and the spectral signal, and the environmental interference is removed by dynamically adjusting the filtering parameters, to obtain the preprocessed signal data set.

[0070] The adaptive filtering algorithm denoises the original signal data set. The magnetic field signal may be disturbed by power lines, producing high-frequency noise; the spectral signal may be affected by changes in cloud cover, introducing low-frequency drift. The adaptive filtering dynamically adjusts parameters such as bandwidth or cutoff frequency according to the signal characteristics. For example, the magnetic field signal noise amplitude is reduced from 0.1 microtesla to 0.02 microtesla after filtering; the spectral signal noise is reduced from 5% to 1%. The signal-to-noise ratio of the denoised signal data set is significantly improved, providing clear data for subsequent analysis.

[0071] Step S240, if the signal-to-noise ratio of the preprocessed signal data set is lower than the preset threshold, the data fusion algorithm is used to associate the preprocessed signal data set with the three-dimensional coordinate data set, and the weighted average method is used to optimize the correspondence between the signal and the coordinate, to obtain the preprocessed signal data.

[0072] The signal-to-noise ratio of the preprocessed signal dataset is: (4) In formula (4), represents the signal-to-noise ratio, represents the signal power, represents the noise power, represents a preset signal-to-noise ratio threshold, and the data fusion algorithm is triggered when the calculated signal-to-noise ratio is lower than the preset threshold.

[0073] The fused coordinate association result is: (5) In formula (5), represents the fused coordinate association result, represents the total number of data points, represents the i-th signal data, represents the i-th coordinate data, represents the i-th three-dimensional coordinate data, represents a regularization parameter, and represents a coordinate association function. The optimized signal data is: (6)

[0074] In formula (6), represents the optimized signal data, represents the number of signals participating in the weighted average, represents the weight of the i-th signal, represents the i-th signal data, represents the standard deviation of the i-th signal, represents the standard deviation of the i-th signal, and the weight is proportional to the signal quality. If the signal-to-noise ratio of the preprocessed signal dataset is lower than the threshold, such as lower than 20 decibels, the data fusion algorithm is started. The fusion algorithm associates the signal dataset with the three-dimensional coordinate dataset and optimizes the corresponding relationship through weighted averaging. For example, the magnetic field signal indicates the direction of the unmanned aerial vehicle, and the weight is set to 0.6; the coordinate data provides the position, and the weight is 0.4.

[0075]

[0076] ​​​​​​​After fusion, the signal and coordinates are precisely matched, generating optimized signal data. For example, magnetic field signals help correct coordinate deviations, while spectral signals provide information about road surface material. The optimized data provides an accurate basis for path planning. The fusion process considers the spatial relationship between signal characteristics and coordinates to ensure data consistency. It can be understood that the above method generates high-quality signal data through calibration, synchronization, denoising, and fusion. Calibration improves sensor accuracy, synchronization ensures time consistency, denoising enhances signal clarity, and fusion optimizes data association. These steps support each other, forming a complete data processing chain that provides reliable support for autonomous driving navigation.

[0077] Furthermore, in the big data-based UAV aerial geophysical prospecting method provided in this embodiment, step S300 includes: Step S310: If the amount of signal data exceeds a preset threshold, the pre-processed signal data is sharded through a distributed computing framework, and a hash mapping algorithm is used to distribute the data to distributed nodes according to the data dimension to obtain a sharded data set.

[0078] The trigger of the signal data volume is judged by the following formula: (7) In formula (7), Indicates the distributed processing trigger flag, represents a signal data set, Indicates the amount of data. Indicates the preset threshold. When the data volume exceeds the threshold, distributed processing is triggered; otherwise, single-machine processing is used.

[0079] The hash value of a data point is calculated using the following formula: (8) In formula (8), Represents the hash value of the data point, Indicates the data points, Indicates the number of data dimensions, Indicates the The hash coefficient of the dimension, Indicates that the data point is The eigenvalue of dimension, Represents the hash modulus parameter.

[0080] The sharded data set is obtained by the following formula: (9) In formula (9), Indicates the assignment to The shard data set of each node, Indicates the a data sample, a hash map value representing the data, a total number of distributed nodes, a total number of data samples, a node number.

[0081] In the scenario of unmanned aerial navigation, the amount of signal data may rapidly accumulate due to high-frequency acquisition, and needs to be efficiently processed through a distributed computing framework. The distributed computing framework divides and distributes data to multiple computing nodes to reduce the pressure on a single node.

[0082] The hash mapping algorithm generates a unique identifier according to the data dimension, such as the strength of the magnetic field signal or the wavelength of the spectral signal, and uniformly distributes the data. For example, the magnetic field signal data is divided into 10 subsets according to the strength interval, and the spectral signal is divided into 8 subsets according to the wavelength range, and each subset is allocated to a different node to ensure load balancing. After fragmentation, each node processes the data in parallel to generate a fragmented data set. This approach is suitable for large-scale data scenarios, such as the case where sensors on urban roads collect millions of records per second.

[0083] In step S320, for the fragmented data set, a convolutional neural network is used to extract features from each fragmented data through convolutional layers and pooling layers, generate a feature vector for each fragmented data, and obtain a feature vector set.

[0084] Specifically, for the fragmented data set, the convolutional neural network is used for feature extraction. The convolutional layer scans the data through a sliding window to capture the local strength variation of the magnetic field signal or the wavelength distribution characteristics of the spectral signal. For example, the convolution kernel of the magnetic field signal can be set to 3x3 to extract the directional variation pattern, and the convolution kernel of the spectral signal can be set to 5x5 to extract the wavelength continuity feature. The pooling layer further compresses the data and retains key features, such as the maximum strength of the magnetic field signal or the peak wavelength of the spectral signal. For example, the pooling operation reduces the spatial resolution of the magnetic field signal by 50% to retain the main directional information, and the pooling of the spectral signal retains the main spectral peak to reduce redundant data. The generated feature vector set contains the core information of each fragment, such as the directional feature vector of the magnetic field signal and the wavelength feature vector of the spectral signal.

[0085] The feature extraction process needs to consider the characteristics of the data. The magnetic field signal may exhibit nonlinear changes due to environmental interference, and the convolutional neural network captures these complex patterns through multiple convolutional layers. For example, in a tunnel scenario, the magnetic field signal is distorted due to metal structure interference, and the convolutional layer can extract distortion features to generate a high-dimensional feature vector.

[0086] The spectral signal can present intermittent characteristics due to light obstruction, and the pooling layer can retain key wavelength information to ensure the robustness of the feature vector. The feature vector set provides high-quality input for subsequent data fusion or path planning. For example, the feature vector can be used to identify the spectral features of road signs or correct the magnetic field direction of the UAV attitude. This method efficiently processes large-scale signal data through distributed computing and feature extraction, providing reliable support for autonomous navigation.

[0087] Further, the embodiment provides a UAV airborne geophysical prospecting method based on big data, and step S400 includes: Step S410, obtain classification input data from the feature vector set, classify the feature vector using a random forest classification model, and obtain a classification result by voting each feature vector through a set of decision trees.

[0088] In the geological exploration scene, the feature vector set is composed of feature vectors of magnetic field signals or spectral signals, which need to be further classified to identify geological anomalies. The random forest classification model generates a classification result by voting feature vectors through multiple decision trees.

[0089] Specifically, each feature vector contains the direction feature of the magnetic field signal or the wavelength feature of the spectral signal, and the random forest synthesizes the classification results of each tree by traversing the set of decision trees. For example, a certain feature vector represents a magnetic field direction anomaly, and 80 out of 100 decision trees determine that it is "abnormal", so the classification result is "geological anomaly". This voting mechanism ensures the robustness of classification.

[0090] Step S420, according to the classification result, obtain pattern data from a pre-established ore body distribution knowledge base, calculate the similarity value between the classification result and the pattern data using a cosine similarity algorithm, and determine the matching pattern data.

[0091] The cosine similarity value between the classification result and the pattern data is calculated by the following formula: (10) In formula (10), denotes the cosine similarity value between the classification result and the pattern data, denotes the th feature component of the classification result vector, denotes the th feature component of the pattern data vector, denotes the dimension number of the feature vector. Formula (10) measures the similarity of two vectors by calculating the cosine value of the angle between them, and the value closer to 1 indicates higher similarity.

[0092] The final matching pattern data number is: (11) In formula (11), Indicates the pattern data number that is finally matched, Represents the total number of pattern data in the knowledge base, Indicates the current classification result. attribute values, Indicates the The first attribute values, Represents the total number of attributes used for matching. Formula (11) determines the best matching pattern data by finding the maximum similarity.

[0093] The complete structure of the pattern data is as follows: (12) In formula (12), Indicates the first A data set of ore body distribution patterns, Indicates the The first mode geometric characteristic parameters, represents the number of geometric feature parameters, Indicates the The first mode attribute characteristic parameters, Indicates the number of attribute feature parameters.

[0094] The classification results are matched with pattern data in the ore body distribution knowledge base. The pattern data stores known ore body distribution characteristics, such as the direction pattern of magnetic field anomalies or the peak pattern of spectral wavelengths.

[0095] The cosine similarity algorithm calculates the similarity between the classification result and the pattern data. For example, if the similarity between the feature vector of the classification result and the iron ore pattern data in the knowledge base is 0.95, exceeding the preset threshold of 0.9, it is considered a match. Preferably, the knowledge base should be regularly updated to cover a variety of ore body types to ensure matching accuracy.

[0096] Step S430: If the similarity value between the matched pattern data and the classification result exceeds a preset threshold, the classification result is mapped to two-dimensional space coordinates through a grid division algorithm to generate an initial geological anomaly distribution map.

[0097] Determine a reasonable threshold range based on statistical distribution: (13) In formula (13), Represents the preset threshold for similarity determination, represents the mean of historical similarity data, represents the standard deviation of historical similarity data, Indicates the threshold adjustment coefficient.

[0098] The coordinate mapping of the classification results to the two-dimensional space is achieved through the following formula: (14) In formula (14), represents the normalized coordinates of the grid points in two-dimensional space, and Represent the row and column indexes of the grid respectively, and Indicates that the grid is and The step size in the direction, and Indicates the offset of the coordinate system origin, and express and The coordinate range of the direction.

[0099] After confirming the matching pattern data, the classification results are mapped to two-dimensional spatial coordinates to generate an initial geological anomaly distribution map. A gridding algorithm converts the spatial information of the feature vectors into grid coordinates. For example, the magnetic field anomaly area is divided into a 100×100 grid, with each grid corresponding to a classification result, labeled "abnormal" or "normal." This mapping method facilitates visualization of anomaly distribution.

[0100] The grid size needs to be adjusted according to the exploration area. For example, a 50×50 grid can be used in large mining areas to improve the resolution.

[0101] Step S440: Based on the initial geological anomaly distribution map, an interpolation algorithm is used to smooth the two-dimensional spatial coordinates, and the final geological anomaly distribution map is generated by combining the spatial distribution characteristics of the pattern data.

[0102] The initial geological anomaly distribution map is smoothed using an interpolation algorithm to generate the final geological anomaly distribution map. Interpolation algorithms fill in the gaps between grids. For example, kriging interpolation can be used to estimate the anomaly intensity in the intermediate region based on the anomaly values ​​of adjacent grids. For example, if the anomaly value of a grid is 0.8 and the value of the adjacent grid is 0.6, the interpolated value in the intermediate region will be 0.7.

[0103] Combined with the spatial distribution characteristics of model data, such as the spatial gradient of magnetic field anomalies in iron ore deposits, the map clearly shows the boundaries and intensity of the anomaly areas. This smoothing process improves the readability of the map and provides precise guidance for subsequent mineral development.

[0104] The above process needs to consider the complexity of the geological environment. For example, the magnetic field signal may present a non-uniform distribution due to rock layer interference, and the random forest classification can effectively identify the interference mode through multi-dimensional feature analysis. Similarly, the cosine similarity matching and interpolation processing combined with the pattern data of the knowledge base ensure that the map reflects the true distribution rule of the ore body. This method works through multiple steps in cooperation, providing efficient and reliable anomaly identification and distribution analysis support for geological exploration.

[0105] Further, the unmanned aerial vehicle airborne geophysical prospecting method based on big data provided by the embodiment comprises the following steps: In step S510, the magnetic field signal and the spectrum signal data are obtained from the geological anomaly distribution map, and a data fusion algorithm is used to process the magnetic field signal and the spectrum signal by a weighted average method to obtain a fusion signal data set.

[0106] In geological exploration, extracting the magnetic field signal and the spectrum signal data from the geological anomaly distribution map is a key step. The magnetic field signal usually reflects the magnetic difference of the underground rock layer, while the spectrum signal reveals the spectral characteristics of the mineral composition. The magnetic field signal may contain the magnetic field strength value of a certain region, such as 100nT, while the spectrum signal may record the reflectivity at a specific wavelength, such as 0.6.

[0107] The data fusion algorithm integrates these two types of signals by a weighted average method to generate a fusion signal data set. In one embodiment, the weights of the magnetic field signal and the spectrum signal are 0.7 and 0.3 respectively, which are determined based on signal reliability and exploration target. For example, the fusion signal value of a certain point is 0.7x100nT+0.3x0.6=70.18. This fusion method balances the contribution of the two signals and improves the comprehensiveness of the data.

[0108] In step S520, according to the fusion signal data set, a three-dimensional inversion algorithm is used to calculate the fusion signal data set by an iterative least squares method to determine the preliminary spatial distribution data.

[0109] The observation data vector in the fusion signal data set is calculated by the following formula: (15) In formula (15), represents the observation data vector of the fusion signal data set, represents the Green's function matrix describing the relationship between the spatial position and the observation data, represents the three-dimensional spatial distribution parameter vector to be solved, represents the observation noise vector.

[0110] The fusion signal dataset is used for a three-dimensional inversion algorithm to calculate the preliminary spatial distribution data by the iterative least squares method. The iterative process adjusts the model parameters multiple times to make the calculation results approximate the observation data. For example, the initial model assumes that the magnetic field in a certain area is uniform, and after 5 iterations, the preliminary spatial distribution data reflecting the gradient change of the magnetic field is generated.

[0111] In step S530, if the inversion accuracy of the preliminary spatial distribution data is lower than the preset threshold, the preliminary spatial distribution data is spatially discretized by a grid division algorithm, and the discrete data is enhanced by a stereoscopic microscopy technique to obtain optimized spatial distribution data.

[0112] The inversion accuracy of the preliminary spatial distribution data is calculated and compared with the preset threshold by the following formula: (16) In formula (16), represents the inversion accuracy evaluation index, represents the total number of spatial distribution data points, represents the th observation data value, represents the th inversion calculation value.

[0113] The grid division discretization processing of the preliminary spatial distribution data is implemented by the following formula: (17) In formula (17), represents the discretization weight of the grid unit, , , represent the grid spacing in three directions respectively, represents the total volume, represents the spatial distribution function value at the coordinates .

[0114] The feature enhancement processing of the discrete data by the stereoscopic microscopy technique is implemented by the following formula: (18) In formula (18), represents the data after feature enhancement, represents the original discrete data, represents the feature data extracted by the stereoscopic microscopy technique, represents the original data weight coefficient, represents the stereoscopic feature weight coefficient, represents the Laplace operator.

[0115] If the inversion accuracy falls below a preset threshold, such as 0.85, further optimization is required. The gridding algorithm discretizes the space into 100×100×50 grids, with each grid storing the fused signal value. For example, a grid value of 70 nT is used. Stereoscopic microscopy then enhances the discretized data to highlight key features. For example, by analyzing the signal gradient between grids, the edge features of the magnetic field anomaly are enhanced, making the boundaries of the anomaly more distinct.

[0116] Step S540: Based on the optimized spatial distribution data, a coordinate mapping algorithm is used to smooth the optimized spatial distribution data through an interpolation algorithm to generate three-dimensional geological structure data.

[0117] The optimized spatial distribution data are smoothed by a coordinate mapping algorithm to generate three-dimensional geological structure data.

[0118] Interpolation algorithms, such as spline interpolation, fill in the gaps between grid cells. For example, if the signal values ​​of two adjacent grid cells in a certain area were 70 nT and 65 nT, the interpolated value of the midpoint would be 67.5 nT. This smoothing process makes 3D geological data more continuous, clearly showing the distribution of underground ore bodies, such as the spatial morphology of iron ore deposits.

[0119] The coordinate mapping algorithm needs to adjust the grid resolution according to the size of the exploration area. For example, a 50×50×25 grid can be used in large mining areas to improve accuracy.

[0120] The generation of the fused signal dataset takes into account the complexity of the geological environment. For example, the magnetic field signal may be non-uniform due to interference from rock formations, while the spectral signal is affected by surface vegetation.

[0121] The weighted averaging method dynamically adjusts weights to reduce interference. Stereoscopic microscopy enhances the ability to identify anomalies by analyzing grid data from multiple angles. This multi-step collaborative approach ensures the accuracy of 3D geological data, providing a reliable basis for mineral exploration.

[0122] Furthermore, in the big data-based UAV geophysical prospecting method provided in this embodiment, step S600 includes: Step S610: Acquire spatial position information from the 3D geological structure data and the 3D coordinate data set, and use a coordinate mapping algorithm to preliminarily align the 3D geological structure data and the 3D coordinate data set through a linear transformation method to obtain a preliminary registration data set.

[0123] Obtaining spatial position information from three-dimensional geological structure data and three-dimensional coordinate data sets is a basic step of geological modeling. Three-dimensional geological structure data usually contains physical properties of underground ore bodies, such as density or magnetism, while three-dimensional coordinate data sets record the longitude, latitude and depth information of spatial points. For example, the three-dimensional geological structure data of a certain mining area may record the density value of a point as 2.7 g / cm³, and the coordinate data set records the spatial position of the point as (x: 500 m, y: 300 m, z: -200 m).

[0124] Through the coordinate mapping algorithm, linear transformation method is used to preliminarily align the two types of data. Linear transformation aligns the reference system of the coordinate data set with the reference system of the geological structure data through translation and rotation operations. For example, assuming that the origin of the coordinate data set deviates from the origin of the geological structure data by 10 m, linear transformation will translate all coordinate points by 10 m, ensuring the spatial consistency of the two data sets, thereby generating a preliminary registration data set. This alignment operation ensures that the data for subsequent processing has a unified spatial reference.

[0125] Step S620, according to the preliminary registration data set, a grid division algorithm is used to perform spatial discretization processing on the preliminary registration data set through uniform grid division method, to obtain a discretized registration data set.

[0126] Based on the preliminary registration data set, the grid division algorithm performs spatial discretization processing through uniform grid division method. Uniform grid division divides the three-dimensional space into regular cubic grids, such as each grid with a side length of 5 m, generating a 100×100×50 grid system. Each point in the preliminary registration data set is assigned to the corresponding grid, for example, the density value of a certain point 2.7 g / cm³ is recorded in the grid (i: 50, j: 30, k: 20). The discretized data set facilitates subsequent algorithm processing.

[0127] Step S630, if the registration error of the discretized registration data set is higher than the preset threshold, a support vector machine algorithm is used to perform feature enhancement on the discretized registration data set through classification method, to obtain an optimized registration data set.

[0128] The registration error of the discretized registration data set is calculated by the following formula: (19) In formula (19), represents the registration error, represents the number of samples in the discretized registration data set, represents the registration result, represents the preset target value.

[0129] The optimized feature set after feature enhancement is: (20) In formula (20), represents the optimized feature set after feature enhancement, represents the support vector machine algorithm, represents the original discretized registration data set, represents the corresponding classification label set.

[0130] The optimized registration data set is: (21) In formula (21), represents the optimized registration data set, represents the feature fusion operation, represents the original discretized registration data set, represents the optimized feature set after feature enhancement.

[0131] If the registration error is higher than the preset threshold, such as 0.9, further optimization is needed. The support vector machine algorithm enhances the features of the discretized registration data set through a classification method. For example, the algorithm identifies grid points with abnormal density values and marks them as ore body boundary features, generating an optimized registration data set. This classification method highlights the features of the ore body edge and improves the discrimination of the data.

[0132] In step S640, according to the optimized registration data set, an interpolation smoothing algorithm is used to smooth the optimized registration data set through a spline interpolation method, generating a three-dimensional geological model containing ore body distribution characteristics.

[0133] The optimized registration data set is processed by the interpolation smoothing algorithm and filled with gaps between grids using the spline interpolation method. For example, the density values of two adjacent grids are 2.7 g / cm³ and 2.5 g / cm³, respectively, and the density value of the middle point calculated in the spline interpolation is 2.6 g / cm³. This smoothing processing makes the data transition more natural, and finally generates a three-dimensional geological model containing ore body distribution characteristics. For example, the model clearly shows the spatial form of a certain iron ore body, such as the ore body presenting an elliptical distribution at a depth of -200 m. This model provides intuitive spatial information for mineral exploration and helps to accurately locate the position of the ore body.

[0134] The application relates to a big data-based unmanned aerial vehicle (UAV) airborne geophysical exploration system for realizing the big data-based UAV airborne geophysical exploration method.

[0135] Further, the big data-based UAV airborne geophysical exploration system provided by the embodiment comprises a first acquisition module, a second acquisition module, a judgment module, a determination module, a generation module and an output module. The first acquisition module is used for acquiring positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through a differential positioning technology, combining a terrain height model to perform height compensation on the corrected data, and obtaining a three-dimensional coordinate data set. The second acquisition module is used for collecting magnetic field signals and spectrum signals according to the three-dimensional coordinate data set, performing denoising processing on the magnetic field signals and the spectrum signals through an adaptive filtering algorithm, and obtaining preprocessed signal data. The judgment module is used for judging whether the signal data amount of the preprocessed signal data exceeds a preset threshold value. If the signal data amount exceeds the preset threshold value, the preprocessed signal data is subjected to sharding processing through a distributed computing framework, a convolutional neural network is used to extract feature vectors in each shard data, and a feature vector set is generated. The determination module is used for inputting the feature vector set into a random forest classification model, acquiring a geological anomaly classification result, combining mode data in a ore body distribution knowledge base, and determining a geological anomaly distribution atlas. The generation module is used for performing three-dimensional inversion calculation on the magnetic field signals and the spectrum signals according to the geological anomaly distribution atlas, establishing a spatial distribution model of an underground ore body, and generating three-dimensional geological structure data. The output module is used for performing spatial registration on the three-dimensional geological structure data and the three-dimensional coordinate data set, and outputting a three-dimensional geological model containing ore body distribution characteristics.

[0135] Further, the big data-based UAV airborne geophysical exploration system provided by the embodiment comprises a first acquisition module, a second acquisition module, a judgment module, a determination module, a generation module and an output module. The first acquisition module is used for acquiring positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through a differential positioning technology, combining a terrain height model to perform height compensation on the corrected data, and obtaining a three-dimensional coordinate data set. The second acquisition module is used for collecting magnetic field signals and spectrum signals according to the three-dimensional coordinate data set, performing denoising processing on the magnetic field signals and the spectrum signals through an adaptive filtering algorithm, and obtaining preprocessed signal data. The judgment module is used for judging whether the signal data amount of the preprocessed signal data exceeds a preset threshold value. If the signal data amount exceeds the preset threshold value, the preprocessed signal data is subjected to sharding processing through a distributed computing framework, a convolutional neural network is used to extract feature vectors in each shard data, and a feature vector set is generated. The determination module is used for inputting the feature vector set into a random forest classification model, acquiring a geological anomaly classification result, combining mode data in a ore body distribution knowledge base, and determining a geological anomaly distribution atlas. The generation module is used for performing three-dimensional inversion calculation on the magnetic field signals and the spectrum signals according to the geological anomaly distribution atlas, establishing a spatial distribution model of an underground ore body, and generating three-dimensional geological structure data. The output module is used for performing spatial registration on the three-dimensional geological structure data and the three-dimensional coordinate data set, and outputting a three-dimensional geological model containing ore body distribution characteristics.

[0136] Preferably, the big data-based unmanned aerial vehicle airborne geophysical exploration system provided in this embodiment comprises a fifth acquisition unit, a sixth acquisition unit, a seventh acquisition unit and an eighth acquisition unit. The fifth acquisition unit is configured to calibrate the magnetic field sensor and the spectrum sensor by a sensor calibration algorithm, correct the sensor output by using preset calibration parameters, and obtain calibrated magnetic field signals and spectrum signals. The sixth acquisition unit is configured to collect data in a preset time interval by using a signal collection module according to the calibrated magnetic field signals and spectrum signals, ensure the time axis of the three-dimensional coordinate dataset by using a time stamp alignment algorithm, and obtain an original signal dataset. The seventh acquisition unit is configured to perform denoising processing on the magnetic field signals and the spectrum signals by using an adaptive filtering algorithm for the original signal dataset, remove environmental interference by dynamically adjusting filtering parameters, and obtain a preprocessed signal dataset. The eighth acquisition unit is configured to perform coordinate association on the preprocessed signal dataset and the three-dimensional coordinate dataset by using a data fusion algorithm if the signal-to-noise ratio of the preprocessed signal dataset is lower than a preset threshold, optimize the correspondence between the signals and the coordinates by using a weighted average method, and obtain the preprocessed signal data.

[0137] The big data-based unmanned aerial vehicle airborne geophysical exploration method and system provided in this embodiment can obtain data of a global navigation satellite system and an inertial measurement unit, correct and compensate by combining differential positioning technology and a terrain height model, and obtain an accurate three-dimensional coordinate dataset. On this basis, magnetic field and spectrum signals are collected and adaptive filtering denoising is performed. For large-scale data, distributed computing and a convolutional neural network are used to extract features, and a random forest classification model is used in combination with a mineral body distribution knowledge base to generate a geological anomaly distribution atlas. Finally, a spatial distribution model of underground mineral bodies is established by three-dimensional inversion calculation, and is registered with the three-dimensional coordinate dataset to output a three-dimensional geological model containing mineral body distribution characteristics. The beneficial effects achieved by this embodiment are as follows: ‌1. High-precision three-dimensional spatial positioning GNSS (such as Beidou and GPS) and inertial navigation system (INS) are fused, RTK differential positioning technology is combined, horizontal positioning accuracy is improved to centimeter level (±1 cm), elevation accuracy is controlled within ±2 cm, and reliable positioning ability is maintained in signal shielding areas such as tunnels and dense forests; the ionospheric error and terrain undulation influence are effectively compensated by a terrain height model (such as EHP technology) and double-frequency carrier phase difference technology, and vertical direction cumulative error is reduced.

[0138] ‌2. Multi-source data fusion and denoising capability Adaptive filtering technology suppresses environmental noise (such as electromagnetic interference, atmospheric disturbance) in magnetic field, spectral signal, improves data signal-to-noise ratio, and lays a foundation for feature extraction; Distributed computing framework (such as RTMC system) supports massive data parallel processing, significantly shortens the feature extraction period of large-scale geological data.

[0139] ‌3、Intelligent feature recognition and classification optimization‌ Convolutional neural network (CNN) automatically extracts deep features of spectral and magnetic field data (such as mineral reflectance features and magnetic anomaly shapes), reduces the subjectivity of manual interpretation; Random forest model combined with ore body distribution knowledge base realizes high-precision classification of geological anomaly areas (such as discrimination between mineralized and non-mineralized areas), and the classification accuracy is improved by 20%-30%.

[0140] ‌4、High-fidelity reconstruction of three-dimensional geological model‌ Three-dimensional inversion algorithm combined with multi-physical field constraints (such as gravity and electromagnetic data) quantitatively analyzes the shape and physical parameters of underground ore bodies, and the matching degree of the model with the actual ore body space is more than 90%; The model is dynamically registered with the three-dimensional coordinate data set, supporting millimeter-level precision display of ore body burial depth, dip angle and other geometric properties.

[0141] ‌5、Resource exploration efficiency and cost optimization‌ Full-process automation reduces human intervention, shortens single exploration period by 40%-50%, and reduces exploration cost by more than 30% in complex terrain areas; The generated three-dimensional geological model can directly guide the optimization of drilling target area, reduce the number of invalid drill holes, and improve the resource exploration rate by more than 25%.

[0142] In summary, the embodiment realizes the full-link optimization from data acquisition to ore body prediction through multi-technology cooperation and data-driven modeling, and provides a high-precision and high-efficiency digital solution for mineral exploration.

[0143] Although the preferred embodiments of the present application have been described, those skilled in the art can make further changes and modifications to the embodiments once they know the basic inventive concept. Therefore, the appended claims are intended to include the preferred embodiments and all changes and modifications falling within the scope of the present application. Obviously, those skilled in the art can make various modifications and changes to the present application without departing from the spirit and scope of the present application. Thus, if these modifications and changes of the present application fall within the scope of the claims of the present application and their equivalents, the present application also intends to include these modifications and changes.

Claims

1. A big data based UAV airborne geophysical method characterized in that, The method comprises the following steps: Obtaining positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through differential positioning technology, combining a terrain height model to compensate the height of the corrected data, and obtaining a three-dimensional coordinate data set; According to the three-dimensional coordinate data set, collecting magnetic field signals and spectrum signals, and using an adaptive filtering algorithm to denoise the magnetic field signals and spectrum signals to obtain preprocessed signal data; Judging whether the signal data amount of the preprocessed signal data exceeds a preset threshold, if the signal data amount exceeds the preset threshold, performing sharding processing on the preprocessed signal data through a distributed computing framework, using a convolutional neural network to extract feature vectors in each sharded data to generate a feature vector set; Inputting the feature vector set into a random forest classification model to obtain a geological anomaly classification result, combining pattern data in a ore body distribution knowledge base to determine a geological anomaly distribution atlas; According to the geological anomaly distribution atlas, performing three-dimensional inversion calculation on the magnetic field signals and spectrum signals to establish a spatial distribution model of underground ore bodies and generate three-dimensional geological structure data; Spatially registering the three-dimensional geological structure data and the three-dimensional coordinate data set to output a three-dimensional geological model containing ore body distribution characteristics.

2. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of obtaining positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through differential positioning technology, combining a terrain height model to compensate the height of the corrected data, and obtaining a three-dimensional coordinate data set comprises: Obtaining positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, aligning the positioning data and measurement data on the time axis through a time synchronization algorithm, preprocessing the measurement data using Kalman filtering to obtain preprocessed data; Using differential positioning technology to correct errors of the preprocessed data, generating correction data through differential calculation of a reference station and a mobile station, and combining a coordinate conversion algorithm to unify the correction data to a preset reference coordinate system to obtain corrected two-dimensional coordinate data; According to the corrected two-dimensional coordinate data, obtaining corresponding digital elevation information from a pre-established terrain height model, and adjusting the height component of the corrected two-dimensional coordinate data through a height compensation algorithm to obtain a preliminary three-dimensional coordinate data set; If the accuracy of the preliminary three-dimensional coordinate data set is lower than a preset threshold, performing secondary fusion of the corrected positioning data and measurement data through a data fusion algorithm, and using a weighted average method to optimize the preliminary three-dimensional coordinate data set to obtain a final three-dimensional coordinate data set.

3. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of obtaining positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correcting the positioning data through differential positioning technology, combining a terrain height model to compensate the height of the corrected data, and obtaining a three-dimensional coordinate data set comprises: Calibrating the magnetic field sensor and the spectrum sensor through a sensor calibration algorithm, and correcting the deviation of sensor output using preset calibration parameters to obtain calibrated magnetic field signals and spectrum signals; According to the calibrated magnetic field signal and the spectrum signal, data is synchronously collected in a preset time interval by using a signal collection module, time stamp alignment algorithm is used to ensure consistency with the time axis of the three-dimensional coordinate data set, and an original signal data set is obtained; For the original signal data set, an adaptive filtering algorithm is used to denoise the magnetic field signal and the spectrum signal, environmental interference is removed by dynamically adjusting the filtering parameters, and a preprocessed signal data set is obtained; If the signal-to-noise ratio of the preprocessed signal data set is lower than a preset threshold, the preprocessed signal data set is associated with the three-dimensional coordinate data set through a data fusion algorithm, the corresponding relationship between the signal and the coordinate is optimized by using a weighted average method, and a preprocessed signal data is obtained.

4. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of judging whether the signal data amount of the preprocessed signal data exceeds a preset threshold, if the signal data amount exceeds the preset threshold, performing sharding processing on the preprocessed signal data through a distributed computing framework, and extracting feature vectors in each shard data by using a convolutional neural network to generate a feature vector set includes: If the signal data amount exceeds the preset threshold, the preprocessed signal data is sharded through a distributed computing framework, and is distributed to distributed nodes according to the data dimension by using a hash mapping algorithm to obtain a shard data set; For the shard data set, a convolutional neural network is used to extract features from each shard data through convolutional layers and pooling layers to generate feature vectors of each shard data and obtain a feature vector set.

5. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of inputting the feature vector set into a random forest classification model, obtaining a geological anomaly classification result, and determining a geological anomaly distribution map in combination with pattern data in a ore body distribution knowledge base includes: Obtain classification input data from the feature vector set, classify the feature vectors by using a random forest classification model, vote each feature vector by traversing a decision tree set, and obtain a classification result; According to the classification result, obtain pattern data from a pre-established ore body distribution knowledge base, calculate the similarity value between the classification result and the pattern data by using a cosine similarity algorithm, and determine the matching pattern data; If the similarity value between the matching pattern data and the classification result exceeds a preset threshold, map the classification result to a two-dimensional spatial coordinate by using a grid division algorithm to generate an initial geological anomaly distribution map; According to the initial geological anomaly distribution map, smooth the two-dimensional spatial coordinate by using an interpolation algorithm, and generate a final geological anomaly distribution map in combination with the spatial distribution characteristics of the pattern data.

6. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of performing three-dimensional inversion calculation on the magnetic field signal and the spectrum signal according to the geological anomaly distribution map, establishing a spatial distribution model of the underground ore body, and generating three-dimensional geological structure data includes: Obtain magnetic field signal and spectrum signal data from the geological anomaly distribution map, process the magnetic field signal and the spectrum signal by using a data fusion algorithm through a weighted average method to obtain a fusion signal data set; According to the fusion signal data set, calculate the fusion signal data set by using a three-dimensional inversion algorithm through an iterative least squares method to determine preliminary spatial distribution data; If the inversion accuracy of the preliminary spatial distribution data is lower than a preset threshold, the preliminary spatial distribution data is subjected to spatial discretization processing through a grid division algorithm, feature enhancement of the discretized data is performed using a stereomicroscopy technique, and optimized spatial distribution data is obtained; According to the optimized spatial distribution data, the optimized spatial distribution data is subjected to smoothing processing using a coordinate mapping algorithm through an interpolation algorithm, and three-dimensional geological structure data is generated.

7. The big data based UAV airborne geophysical survey method of claim 1, wherein, The step of spatially registering the three-dimensional geological structure data and the three-dimensional coordinate data set and outputting a three-dimensional geological model containing ore body distribution characteristics includes: Spatial position information is obtained from the three-dimensional geological structure data and the three-dimensional coordinate data set, and the three-dimensional geological structure data and the three-dimensional coordinate data set are preliminarily aligned using a coordinate mapping algorithm through a linear transformation method, and a preliminary registration data set is obtained; According to the preliminary registration data set, the preliminary registration data set is subjected to spatial discretization processing using a grid division algorithm through a uniform grid segmentation method, and a discretized registration data set is obtained; If the registration error of the discretized registration data set is higher than a preset threshold, feature enhancement of the discretized registration data set is performed using a support vector machine algorithm through a classification method, and an optimized registration data set is obtained; According to the optimized registration data set, the optimized registration data set is subjected to smoothing processing using an interpolation smoothing algorithm through a spline interpolation method, and a three-dimensional geological model containing ore body distribution characteristics is generated.

8. A big data based UAV airborne geophysical system for implementing the big data based UAV airborne geophysical method according to any one of claims 1 to 7, characterized in that, The unmanned aerial vehicle airborne geophysical prospecting system based on big data includes: The first acquisition module is configured to acquire positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, correct the positioning data through differential positioning technology, and combine a terrain height model to perform height compensation on the corrected data, thereby obtaining a three-dimensional coordinate data set; The second acquisition module is configured to acquire magnetic field signals and spectrum signals based on the three-dimensional coordinate data set, and perform denoising processing on the magnetic field signals and spectrum signals using an adaptive filtering algorithm, thereby obtaining preprocessed signal data; The determination module is configured to input the feature vector set into a random forest classification model, acquire a geological anomaly classification result, and determine a geological anomaly distribution map in combination with pattern data in a ore body distribution knowledge base; The generation module is configured to perform three-dimensional inversion calculation on the magnetic field signals and spectrum signals based on the geological anomaly distribution map, establish a spatial distribution model of underground ore bodies, and generate three-dimensional geological structure data; The output module is configured to spatially register the three-dimensional geological structure data and the three-dimensional coordinate data set, and output a three-dimensional geological model containing ore body distribution characteristics. The first acquisition module includes:

9. The big data based UAV airborne geophysical system of claim 8, wherein, ​ The first acquisition unit is configured to acquire positioning data of a global navigation satellite system and measurement data of an inertial measurement unit, perform time axis alignment on the positioning data and the measurement data through a time synchronization algorithm, and perform preprocessing on the measurement data through Kalman filtering to obtain preprocessed data. The second acquisition unit is configured to perform error correction on the preprocessed data through a differential positioning technology, generate correction data through differential calculation of a reference station and a mobile station, unify the correction data to a preset reference coordinate system through a coordinate conversion algorithm, and obtain corrected two-dimensional coordinate data. The third acquisition unit is configured to acquire corresponding digital elevation information from a pre-established terrain height model according to the corrected two-dimensional coordinate data, adjust a height component of the corrected two-dimensional coordinate data through a height compensation algorithm, and obtain a preliminary three-dimensional coordinate data set. The fourth acquisition unit is configured to, if the accuracy of the preliminary three-dimensional coordinate data set is lower than a preset threshold, perform secondary fusion on the corrected positioning data and the measurement data through a data fusion algorithm, optimize the preliminary three-dimensional coordinate data set through a weighted average method, and obtain a final three-dimensional coordinate data set.

10. The big data based UAV airborne geophysical system of claim 8, wherein, The second acquisition module includes: The fifth acquisition unit is configured to calibrate a magnetic field sensor and a spectrum sensor through a sensor calibration algorithm, correct deviations of sensor outputs through preset calibration parameters, and obtain calibrated magnetic field signals and spectrum signals. The sixth acquisition unit is configured to acquire data within a preset time interval through a signal acquisition module according to the calibrated magnetic field signals and the spectrum signals, ensure time axis consistency with the three-dimensional coordinate data set through a time stamp alignment algorithm, and obtain an original signal data set. The seventh acquisition unit is configured to, for the original signal data set, perform denoising processing on the magnetic field signals and the spectrum signals through an adaptive filtering algorithm, remove environmental interference through dynamic adjustment of filtering parameters, and obtain a preprocessed signal data set. The eighth acquisition unit is configured to, if the signal-to-noise ratio of the preprocessed signal data set is lower than a preset threshold, perform coordinate correlation between the preprocessed signal data set and the three-dimensional coordinate data set through a data fusion algorithm, optimize the corresponding relationship between signals and coordinates through a weighted average method, and obtain preprocessed signals.

Images