A multi-channel high-density electrical method data acquisition and processing integrated system

Abstract

The application discloses a kind of multi-channel high-density electrical method data acquisition processing integrated system, it is related to geophysical exploration technical field, the system includes: data acquisition module, data processing module, data fusion and inversion module, intelligent interpretation and anomaly identification module, data management and visualization module, quality control and output module, system interaction module.In the application, the synchronous excitation and multi-channel parallel acquisition technology based on pseudo-random coding realizes data acquisition;Adaptive filtering and wavelet transform are used for signal processing;High-resolution apparent resistivity profile is generated using Kriging interpolation fusion and terrain-constrained inversion algorithm;Combined with convolutional neural network and isolated forest algorithm, geological layering and anomaly automatic identification are realized;And through space-time database and WebGL engine, integrated management and three-dimensional visualization of data are supported, the operation efficiency, data quality and interpretation accuracy of electrical exploration are improved.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration technology, and in particular to an integrated system for multi-channel high-density electrical resistivity tomography (EDT) data acquisition and processing. Background Technology

[0002] Multi-channel high-density electrical resistivity data acquisition and processing integration is a geophysical exploration technology system that deeply integrates multi-channel parallel data acquisition with real-time / automated data processing. Through the collaborative work of multi-channel acquisition instruments, intelligent electrode arrays, and integrated software, it achieves seamless connection of the entire process from field data acquisition to indoor / field processing. Its core relies on multi-channel parallel acquisition to break through the traditional inefficient single-channel mode, and combined with an integrated platform to achieve noise filtering, terrain correction, real-time inversion, and two-dimensional / three-dimensional resistivity imaging, thereby improving operational efficiency.

[0003] Currently, the data acquisition process mostly adopts a sequential mode, which is inefficient and has poor anti-interference capabilities; the data processing and inversion processes are isolated from each other, and there is a lack of effective means to integrate data from multiple devices, resulting in limited resolution of the inversion model; geological interpretation relies heavily on human experience, with low levels of automation and intelligence, resulting in low efficiency and poor consistency.

[0004] Therefore, a multi-channel, high-density electrical resistivity tomography (EDT) data acquisition and processing integrated system is proposed to solve the above problems. Summary of the Invention

[0005] The main objective of this invention is to provide an integrated system for multi-channel high-density electrical resistivity data acquisition and processing to solve the problems mentioned in the background above.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is: a multi-channel high-density electrical resistivity tomography (OTT) data acquisition and processing integrated system, the system comprising: Data acquisition module: Through synchronous excitation technology based on pseudo-random coding and multi-channel parallel acquisition technology, it drives multiple pairs of electrodes and synchronously acquires full-waveform potential difference signals to generate raw electrical resistivity data; Data processing module: Through adaptive filtering and wavelet transform techniques, the raw electrical resistivity data is denoised, baseline corrected, and spike anomaly suppressed in real time to generate preprocessed electrical resistivity data; Data fusion and inversion module: Through dynamic data fusion technology based on Kriging interpolation and least squares inversion algorithm with terrain constraints, preprocessed electrical resistivity data from different device modes are fused to generate high-resolution two-dimensional and three-dimensional apparent resistivity profiles. Intelligent interpretation and anomaly identification module: Based on convolutional neural network and isolated forest algorithm, it automatically performs geological stratification, anomaly delineation and attribute identification on apparent resistivity profile, and generates geological interpretation results and anomaly reports; Data Management and Visualization Module: Employing a spatiotemporal database and a WebGL 3D rendering engine, it provides integrated storage, management, and dynamic 3D visualization of data throughout the entire process of acquisition, processing, and interpretation. Quality control and output module: Through a confidence interval-based data quality assessment algorithm and a standardized template engine, it monitors data quality in real time and automatically generates output charts and reports that conform to industry standards; System interaction module: Provides a web-based graphical user interface and API interface for task configuration, process control, parameter adjustment and result interaction.

[0007] Preferably, the data acquisition module includes a channel management unit, a synchronization activation unit, and a parallel acquisition unit; The channel management unit is used to configure and manage the channel connection status and working mode of up to hundreds of electrodes, and to achieve rapid switching of electrode paths through a high-speed electronic switch matrix. The synchronous excitation unit uses pseudo-random sequence coding technology and a high-precision constant current source to generate and synchronously apply excitation current to the designated electrode pair in order to reduce environmental noise interference and improve the signal-to-noise ratio. The parallel acquisition unit uses a high-precision analog-to-digital converter and field-programmable gate array technology to synchronously and in parallel acquire the full waveform potential difference signals of all measurement electrode pairs, generating raw electrical resistivity data.

[0008] Preferably, the data processing module includes a real-time noise reduction unit, a signal correction unit, and a data packaging unit; The real-time noise reduction unit uses an adaptive filtering algorithm based on the reference electrode to identify and suppress power frequency interference and random noise in real time. The signal correction unit analyzes the signal baseline drift through wavelet transform and performs automatic correction, while using the threshold method to suppress transient spike anomalies generated during the acquisition process. The data packaging unit adds timestamps, electrode positions, and device parameter metadata to the noise-reduced and corrected signals, and packages them to generate preprocessed electrical resistivity data in a standard format.

[0009] Preferably, the data fusion and inversion module includes a dynamic weighted fusion unit, a fast inversion calculation unit, and a result optimization unit. The dynamic weighted fusion unit performs dynamic weighted fusion based on the signal-to-noise ratio and spatial sampling density of data from different device modes, using the Kriging interpolation algorithm to generate a fused data field. The fast inversion calculation unit uses a least squares optimization inversion algorithm with terrain undulation constraints to perform fast iterative inversion calculations on the fused data field and generate two-dimensional and three-dimensional apparent resistivity distribution models. The result optimization unit optimizes the inversion results using regularization smoothing technology to reduce false anomalies and generate a high-resolution apparent resistivity profile.

[0010] Preferably, the multi-device data fusion in the dynamic weighted fusion unit includes the following steps: S1: Data spatial registration: Mapping preprocessed electrical resistivity data collected from different device modes onto regular grid nodes under the same measurement coordinate system; S2: Dynamic weight calculation: For each grid node, the Kriging interpolation weight coefficient is dynamically calculated based on the spatial density, signal-to-noise ratio, and distance from the surrounding data from different sources. S3: Fusion Interpolation Execution: Using the calculated dynamic weights, weighted interpolation is performed on data from multiple devices at grid nodes to generate a spatially continuous and information-complete fused data field; S4: Fusion quality verification: Calculate the spatial continuity index and variance of the fused data field. If the threshold is not reached, return to S2 to adjust the weight calculation strategy.

[0011] Preferably, the intelligent interpretation and anomaly recognition module includes an automatic hierarchical unit, an anomaly detection unit, and an attribute interpretation unit; The automatic stratification unit is based on a pre-trained convolutional neural network model and automatically identifies the electrical stratigraphic boundaries in the apparent resistivity profile to complete the automatic geological stratification. The anomaly detection unit uses the isolated forest unsupervised learning algorithm to automatically detect and delineate electrical anomalies that are significantly different from the background field on a hierarchical basis. The attribute interpretation unit combines the anomaly's shape, size, and resistivity range with knowledge base matching to provide a preliminary geological attribute interpretation, generating geological interpretation results and anomaly reports.

[0012] Preferably, the training and online optimization of the pre-trained convolutional neural network model in the automatic hierarchical unit includes the following steps: S1: Offline model pre-training: Collect a large amount of existing manually interpreted and labeled typical apparent resistivity profile data as a training set, train a U-Net structured convolutional neural network model, and learn the mapping relationship from resistivity images to hierarchical labels. S2: Online Incremental Learning: In practical applications, the system uses new interpretations that have been confirmed by experts and have high credibility as new samples, and employs incremental learning algorithms to fine-tune and optimize the network model to adapt it to the new geophysical characteristics of the work area. S3: Inference and Post-processing: The apparent resistivity profile to be interpreted is input into the optimized network model for forward propagation to obtain a preliminary hierarchical probability map. Then, the boundary continuity is optimized through morphological operations, and finally, the automatic hierarchical results are output.

[0013] Preferably, the data management and visualization module includes a spatiotemporal database unit, a 3D visualization engine unit, and a results integration and display unit; The spatiotemporal database unit adopts a hybrid database architecture to store and manage spatiotemporal data throughout the entire process from raw data to interpretation results, supporting efficient querying and backtracking. The three-dimensional visualization engine unit uses WebGL technology to build a browser-side three-dimensional rendering engine to realize dynamic interactive visualization of survey line layout, data profile, and three-dimensional inversion model; The results integration and display unit integrates and links the results of data collection, processing, inversion, and interpretation within the same spatiotemporal framework for display.

[0014] Preferably, the quality control and output module includes a quality assessment unit, a standardized drawing unit, and a report generation unit; The quality assessment unit calculates the confidence interval for each data point based on statistical principles, evaluates and marks suspicious and out-of-tolerance data in real time, and generates a data quality log. The standardized mapping unit has built-in map templates that conform to industry standards, and automatically draws apparent resistivity cross-sections, contour maps and three-dimensional renderings based on the inversion and interpretation results. The report generation unit integrates text, graphics, tables, and quality logs, and automatically generates a comprehensive results report according to a preset template.

[0015] Preferably, the system interaction module includes a Web graphical interface unit, a process control API unit, and a user management and logging unit; The Web graphical interface unit provides a browser-based interface for task creation, parameter setting, process monitoring, result viewing, and interactive analysis. The process control API unit provides an application programming interface, enabling users to submit batch tasks, customize process orchestration, and integrate with the system via scripts; The User Management and Logs unit is responsible for user authentication, access control, and the recording and auditing of all system operation logs.

[0016] The present invention has the following beneficial effects: 1. In this invention, during high-density electrical resistivity tomography (EPT) data acquisition, by integrating pseudo-random coding synchronous excitation technology and multi-channel parallel acquisition technology based on field-programmable gate arrays, multiple pairs of electrodes can be driven to work synchronously, and the potential difference signals of all measurement channels can be acquired synchronously with full waveform. This eliminates the traditional sequential acquisition mode and improves the efficiency of field data acquisition. At the same time, the pseudo-random coding excitation has white noise-like characteristics, which can effectively suppress periodic interference such as power frequency at specific frequencies. Combined with the high time consistency of parallel acquisition, the signal-to-noise ratio and integrity of the raw data are significantly improved, providing a foundation for subsequent high-precision processing and inversion, and solving the problems of slow acquisition speed and weak anti-interference ability of traditional equipment.

[0017] 2. In this invention, during data processing and geological interpretation, dynamic data fusion technology based on Kriging interpolation can adaptively and weightedly fuse observation data from different device modes to form a fused data field with more complete information and better spatial coverage. Furthermore, a least-squares inversion algorithm with topographic constraints is employed to obtain a high-resolution two-dimensional / three-dimensional apparent resistivity model. An automatic hierarchical model based on U-Net convolutional neural networks and an isolated forest anomaly detection algorithm are introduced to achieve intelligent interpretation of the apparent resistivity profile. This enables automatic division of electrical stratigraphic levels and delineation of anomalies, organically combining data fusion, constrained inversion, and artificial intelligence interpretation. This significantly reduces the ambiguity of inversion results, improves the objectivity, accuracy, and efficiency of geological interpretation, and reduces excessive reliance on human experience.

[0018] 3. In this invention, by constructing an integrated system architecture covering the entire process of acquisition, processing, inversion, interpretation, and output, and using a spatiotemporal database for unified data management, and utilizing a WebGL engine to achieve 3D dynamic visualization, this invention seamlessly integrates the originally scattered operational processes, realizing closed-loop management and efficient flow of exploration data. The web-based graphical interface and API provide flexible task configuration and control methods, supporting both convenient field operations and background batch processing and system integration. Overall, this invention improves the automation and intelligence level of high-density electrical resistivity tomography (EDT) exploration and the efficiency of results output, ensures data quality and the traceability and standardization of interpretation results, and effectively reduces the overall exploration cost. Attached Figure Description

[0019] Figure 1 This is a framework diagram of a multi-channel high-density electrical resistivity data acquisition and processing integrated system according to the present invention; Figure 2 This is a flowchart of a multi-channel high-density electrical resistivity data acquisition and processing integrated system according to the present invention. Detailed Implementation

[0020] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] Please see Figures 1-2 This invention provides a technical solution: a multi-channel high-density electrical resistivity tomography (OTT) data acquisition and processing integrated system, comprising: Data acquisition module: Through synchronous excitation technology based on pseudo-random coding and multi-channel parallel acquisition technology, it drives multiple pairs of electrodes and synchronously acquires full-waveform potential difference signals to generate raw electrical resistivity data; Data processing module: Through adaptive filtering and wavelet transform techniques, the raw electrical resistivity data is denoised, baseline corrected, and spike anomaly suppressed in real time to generate preprocessed electrical resistivity data; Data fusion and inversion module: Through dynamic data fusion technology based on Kriging interpolation and least squares inversion algorithm with terrain constraints, preprocessed electrical resistivity data from different device modes are fused to generate high-resolution two-dimensional and three-dimensional apparent resistivity profiles. Intelligent interpretation and anomaly identification module: Based on convolutional neural network and isolated forest algorithm, it automatically performs geological stratification, anomaly delineation and attribute identification on apparent resistivity profile, and generates geological interpretation results and anomaly reports; Data Management and Visualization Module: Employing a spatiotemporal database and a WebGL 3D rendering engine, it provides integrated storage, management, and dynamic 3D visualization of data throughout the entire process of acquisition, processing, and interpretation. Quality control and output module: Through a confidence interval-based data quality assessment algorithm and a standardized template engine, it monitors data quality in real time and automatically generates output charts and reports that conform to industry standards; System interaction module: Provides a web-based graphical user interface and API interface for task configuration, process control, parameter adjustment and result interaction.

[0022] The data acquisition module includes a channel management unit, a synchronization activation unit, and a parallel acquisition unit; The channel management unit is used to configure and manage the channel connection status and operating mode of up to hundreds of electrodes, and achieves rapid switching of electrode paths through a high-speed electronic switch matrix. The synchronous excitation unit employs pseudo-random sequence coding technology and a high-precision constant current source to generate and synchronously apply excitation current to a designated electrode pair, thereby reducing environmental noise interference and improving the signal-to-noise ratio. This includes the following steps: Generation and loading of pseudo-random encoded sequences: An m-sequence (maximum-length linear feedback shift register sequence) is used as the core encoder for the excitation signal. Its generation is achieved through a... Implemented using a linear feedback shift register, the period of this sequence is... It exhibits autocorrelation characteristics similar to white noise, which is beneficial for separating signal and noise in subsequent correlation detection. The recursive formula for sequence generation is: ; in, The sequence value at the current moment (0 or 1). For feedback coefficients, This represents modulo-2 addition (XOR operation). The generated binary sequence is then converted into a digital instruction sequence that controls the polarity and amplitude of the output current of the high-precision constant current source after waveform mapping, and is pre-stored in the cache of the excitation controller. Controlled output of a high-precision constant current source: The excitation controller is activated based on the applied coded sequence and the set current amplitude. A coded, modulated excitation current is generated through a digital-to-analog converter (DAC) and a precision V / I conversion circuit. Its output satisfies:

[0023] Meanwhile, the constant current source has a built-in feedback loop that samples the output current in real time and adjusts it through a PID controller to ensure that it can maintain high accuracy and stability under complex load changes. Strict synchronization of multi-channel excitation signals: A global clock source based on a high-stability crystal oscillator is used to provide a unified time base for all excitation channels. At the moment the excitation command is triggered, all selected excitation electrode pairs apply the generated signal synchronously according to the same clock edge. The synchronization mechanism eliminates time jitter between channels, ensuring that all measurement points observe a fully synchronized field source excitation, providing a time alignment basis for subsequent data fusion processing; The parallel acquisition unit synchronously and in parallel acquires the full-waveform potential difference signals of all measurement electrode pairs through a high-precision analog-to-digital converter and field-programmable gate array technology, generating raw electrical resistivity data, including the following steps: Full waveform signal conditioning and anti-aliasing filtering. All measurement electrode pairs. Potential difference signal induced above First, the signal passes through a front-end analog conditioning circuit, which includes an instrumentation amplifier (for high common-mode rejection ratio amplification), a programmable gain amplifier (adapted to different signal ranges), and an anti-aliasing low-pass filter with an adjustable cutoff frequency. Based on the main frequency of the pseudo-random coding sequence and the system's highest sampling rate Adaptive settings, strictly follow The criteria are to prevent spectral aliasing; Synchronous Sampling and Quantization Based on High-Precision ADC: Conditioned multi-channel analog signals are fed into a multi-channel synchronous sampling analog-to-digital converter (ADC). All ADC channels share the same high-precision sampling clock, which is obtained by frequency division from the global clock source by the digital clock management module inside the FPGA, ensuring high synchronization at the sampling time. At each rising edge of the sampling clock, all channels simultaneously sample the input voltage. Sample and quantize into digital code values The conversion relationship is as follows:

[0024] in, For the current programmable gain, This is the ADC reference voltage. To achieve ADC resolution, the sampled raw digital bitstream is fed into the FPGA in real time for subsequent processing. FPGA-based real-time data stream processing and packaging: The FPGA receives parallel data streams from the ADC and executes a real-time processing pipeline. First, the data is digitally filtered and preliminarily verified. Then, each sample point data is associated and packaged with its corresponding timestamp, channel ID, and phase information of the current excitation sequence. Finally, the FPGA continuously outputs the organized raw electrical resistivity data packets containing complete waveform information and metadata to the subsequent data processing modules through a high-speed communication interface.

[0025] The data processing module includes a real-time noise reduction unit, a signal correction unit, and a data packaging unit; The real-time noise reduction unit employs an adaptive filtering algorithm based on a reference electrode to identify and suppress power frequency interference and random noise in real time, including the following steps: Reference electrode signal acquisition and interference feature extraction: At least one reference electrode R is deployed in a quiet location, far from the main excitation source, outside the measuring electrode array. The system synchronously acquires the potential signal between the reference electrode and the system common ground. This signal is considered to mainly contain power frequency (e.g., 50 / 60Hz and its harmonics) and broadband environmental noise, but essentially contains no useful response signal from the target geological body. A fast Fourier transform is performed on the signal to analyze its dominant interference frequency in real time. and its amplitude ; Adaptive filtering based on the minimum mean square error criterion: The reference electrode signal... As input to the adaptive filter, the original signal of any measurement channel is used. As the desired response, the least mean square adaptive algorithm is employed, the core of which is to iteratively update the weight coefficient vector of the finite impulse response (FIR) filter. At each discrete time point The output of the filter As a response Zhongyu Estimation of relevant interference components, error signal The formula for updating the weight coefficients is as follows: ; in, To achieve convergence step size, the algorithm is adaptively adjusted based on signal power to ensure algorithm stability. Statistical suppression and output of random noise: the signal after adaptive filtering The signal still contains random noise. To address this, the unit applies a sliding window-based statistical filter to calculate the mean and standard deviation of the signal within the window. For deviations from the mean exceeding ( For singularities that are set to a preset threshold (e.g., 2 or 3), the median value within the window is used for replacement, thereby smoothing and suppressing random impulse noise, and finally outputting a signal that has undergone two-step noise reduction processing. ; The signal correction unit analyzes signal baseline drift using wavelet transform and performs automatic correction. Simultaneously, it uses a threshold method to suppress transient spikes generated during acquisition, including the following steps: Baseline drift estimation based on wavelet multi-scale decomposition: for the denoised signal Discrete wavelet transform (DWT) is performed, and a wavelet basis with compact support and regularity is selected to decompose the signal into multiple scales; among them, the lowest frequency approximation coefficients are... ( The maximum decomposition level (representing the baseline drift and slow change trend of the signal) is used to reconstruct the signal by using only the lowest frequency approximation coefficients and setting all detail coefficients to zero, thus obtaining the baseline drift component. : ; in, Indicates the inverse discrete wavelet transform; Automatic baseline correction and removal: The estimated baseline drift component is directly subtracted from the original denoised signal to obtain the baseline-corrected signal. ; Transient spike suppression based on improved threshold function: for baseline-corrected signals A wavelet transform is performed again to analyze its high-frequency detail coefficients. Since transient spikes (such as external electromagnetic pulses) appear as points with prominent amplitudes in the high-frequency detail coefficients of the wavelet domain, this unit employs an improved soft-threshold function for processing each detail coefficient. ( As a scale, (For position), the processed coefficients are: ; in, It is the noise standard deviation estimate of the detail coefficients at the j-th scale (usually robustly estimated by dividing the absolute value of the median of the coefficients at that scale by 0.6745). The threshold adjustment factor is used; wavelet reconstruction is performed using the processed coefficients and the retained low-frequency approximation coefficients to obtain the final corrected signal. ; The data packaging unit adds metadata such as timestamps, electrode positions, and device parameters to the noise-reduced and corrected signals, and packages them to generate preprocessed electrical resistivity data in a standard format.

[0026] The data fusion and inversion module includes a dynamic weighted fusion unit, a fast inversion calculation unit, and a result optimization unit. The dynamic weighted fusion unit performs dynamic weighted fusion based on the signal-to-noise ratio and spatial sampling density of data from different device modes, using the Kriging interpolation algorithm to generate a fused data field, including the following steps: S1: Data Spatial Registration: Mapping preprocessed electrical resistivity data acquired from different device modes onto regular grid nodes in the same measurement coordinate system, including the following steps: The system receives preprocessed electrical resistivity data from the data processing module, which originates from different device models (such as Winner and Schlumberger). First, all data points (each containing its electrode coordinates and apparent resistivity value) are processed. Unified conversion to the same Cartesian coordinate system Below; subsequently, a regular computational grid is defined above this region, with the grid node positions denoted as... The data from all discrete and irregularly distributed devices are mapped onto this unified grid framework to establish a spatial reference for subsequent interpolation and fusion. S2: Dynamic Weight Calculation: For each grid node, the Kriging interpolation weight coefficients are dynamically calculated based on the spatial density, signal-to-noise ratio, and distance from the surrounding data from different sources. This includes the following steps: For each target grid node Search for all valid data points from different device modes within its neighborhood. ,in The core of Kriging interpolation is to assign optimal weights to these data points. This makes the estimated value Unbiased and with the minimum estimated variance, the weights are obtained by solving the Kriging equations: ; in, It is a variation function used to quantify the correlation of data over spatial distance h. The variation function is dynamically constructed as follows: It is no longer just a function of distance, but a combination of data points. signal-to-noise ratio and its local spatial sampling density dynamic variation function Its form can be simplified to: ; in, It is a basic variation model (such as a spherical model) and a signal-to-noise ratio weighting function (high signal-to-noise ratio data have increased weight). It is a density-weighted function (the influence range of data points in high-density areas is relatively reduced), which makes the weights... It can dynamically adapt to data quality and spatial distribution; S3: Fusion Interpolation Execution: Using the calculated dynamic weights, weighted interpolation is performed on the data from multiple devices at the grid nodes to generate a spatially continuous and information-complete fused data field, including the following steps: The dynamic weights calculated in step S2 Calculate each grid node fused apparent resistivity value This operation is performed on all grid nodes to generate a spatially continuous fused apparent resistivity data field that covers the entire detection area. This data field combines the advantages of different device modes, making reasonable inferences through kriging in data-sparse areas and highlighting high-quality data through dynamic weights in data-dense areas. S4: Fusion Quality Verification: Calculate the spatial continuity index and variance of the fused data field. If the threshold is not met, return to S2 to adjust the weight calculation strategy, including the following steps: The quality metrics of the fused data field are calculated, mainly including spatial continuity metrics (such as the statistical distribution of gradient values ​​between adjacent grid nodes) and variance metrics (the spatial average of the variance estimated by Kriging). These two metrics are compared with preset thresholds. If the requirements are not met (such as the presence of discontinuous abrupt changes or excessively high overall variance), the process returns to step S2, and the signal-to-noise ratio weighting function in the dynamic variogram is adjusted. or density weighting function The parameters are adjusted, and the weights and interpolations are recalculated until the fused data field meets the quality requirements. The fast inversion calculation unit employs a least-squares optimization inversion algorithm with terrain undulation constraints to perform rapid iterative inversion calculations on the fused data field, generating two-dimensional and three-dimensional apparent resistivity distribution models, including the following steps: Establish a forward and inverse model with terrain constraints: use the fused data field as the observation data vector. Define the resistivity model parameter vector of the subsurface medium. Its dimension is consistent with the number of cells in the inversion grid (usually denser than the fused grid), establishing a nonlinear forward modeling operator that connects the model and the data. The specific method is as follows: based on the elevation data of the measurement area, the terrain interface is used as the top boundary of the inversion grid, so that the top cells of the grid completely match the surface undulations, thereby enabling forward modeling calculations. Accurately incorporate terrain effects in real time; Construct the objective function and solve iteratively. The inversion problem is reduced to minimizing the following regularized objective function. : ; The first item is the data fit difference. The first term is the data weight matrix (usually a diagonal matrix of the inverse of the observation error), and the second term is the regularization term (model constraint term). As a regularization factor, As a reference model, The roughness matrix of the model is used; linearized Gauss-Newton method and other optimization algorithms are employed for iterative solution to achieve the desired result. Minimal Model Each iteration requires calculating the sensitivity matrix (Jacobi matrix) and solving the system of linear equations; Convergence Criteria and Inversion Result Output: Convergence criteria are set, typically including the rate of decrease of the objective function, the model update increment, or the maximum number of iterations. When any one of the convergence criteria is met, iteration stops, and the final output is... This is the two-dimensional or three-dimensional resistivity distribution model obtained by inversion, which has already taken into account the constraints of terrain undulation; The result optimization unit optimizes the inversion results using regularized smoothing techniques to reduce spurious anomalies and generate a high-resolution apparent resistivity profile, including the following steps: Regularized smoothing filter for inversion results: Model obtained from direct inversion This unit addresses potential high-frequency oscillations or local spurious anomalies caused by data noise or inversion ambiguity. It applies gradient-domain guided filtering-based regularization smoothing to the model and solves for a minimization: ; in, The smoothed model, For gradient operators, The weight matrix is ​​an anisotropic smoothing matrix. This is the smoothing intensity factor; Anomaly Focusing under Geological Rationality Constraints: On the basis of smoothing, in order to further suppress scattered false anomalies and focus on real anomalies, an edge-preserving focusing optimization technique is adopted; for example, a minimum support functional based on reweighting is introduced, and the weights of the model parameters are adjusted iteratively so that the final model tends to present blocky or concentrated features, reducing scattered and cloud-like anomalies, thereby making the boundaries of geological bodies in the resistivity profile clearer and sharper. Generating the final high-resolution apparent resistivity profile: the smoothed and focused model The data is re-meshed according to spatial location and depth, and output as images or 3D volume data to generate the final high-resolution apparent resistivity profile or 3D model for use by the subsequent intelligent interpretation module.

[0027] The intelligent interpretation and anomaly recognition module includes an automatic hierarchical unit, an anomaly detection unit, and an attribute interpretation unit; The automatic stratification unit, based on a pre-trained convolutional neural network model, automatically identifies electrical stratigraphic boundaries in the apparent resistivity profile and completes automatic geological stratification, including the following steps: S1: Offline Model Pre-training: Collect a large amount of manually interpreted and labeled typical apparent resistivity profile data as the training set, train a U-Net-structured convolutional neural network model, and learn the mapping relationship from resistivity images to hierarchical labels, including the following steps: A large dataset of apparent resistivity profiles, meticulously labeled by geological experts across stratigraphic boundaries from numerous historical exploration projects, was collected to form the training dataset. Each sample includes input (a normalized two-dimensional apparent resistivity image) and labels (a layered mask image of the same size as the image, with different integers representing different geological stratigraphic layers). The dataset is trained using a U-Net network architecture. This network contains symmetrical encoder and decoder paths, fusing shallow details and deep semantic features through skip connections, making it suitable for image segmentation tasks. The training process is supervised learning, minimizing the cross-entropy loss function between the predicted and ground truth masks. To optimize network parameters : ; in, It is the total number of pixels. It is the number of layer categories (including background). The true label (one-hot encoded) of pixel i belonging to category c. This is the probability that pixel i belongs to class c, predicted by the network, and iteratively updated using backpropagation and the Adaptive Moment Estimation (Adam) optimizer. The process continues until the model's segmentation accuracy on the validation set converges, resulting in the initial pre-trained model. ; S2: Online Incremental Learning: In practical applications, the system uses new interpretations confirmed by experts or with high credibility as new samples, and employs incremental learning algorithms to fine-tune and optimize the network model to adapt it to the new geophysical characteristics of the work area. This includes the following steps: pre-trained model Deployed into the integrated system, the newly generated apparent resistivity profiles in practical applications, after being reviewed, corrected, or confirmed by experts through the interactive interface, are treated as high-quality new samples. An elastic weight consolidation strategy is then used for online incremental learning, optimizing the objective function to: ; in, It is the loss of the new sample. It is a parameter In the pre-trained model The value in These are the parameters corresponding to the diagonal of the Fisher information matrix. The estimated value is used to quantify the importance of this parameter to older tasks. It is a hyperparameter that controls the strength of constraints. Through this mechanism, the model can learn the geophysical characteristics of new work areas while avoiding catastrophic forgetting of the stratification rules already learned, and achieve adaptive optimization. Inference and Post-processing: The apparent resistivity profile to be interpreted is input into the optimized network model for forward propagation to obtain a preliminary hierarchical probability map. Then, morphological operations are used to optimize the boundary continuity, and finally, the automatic hierarchical results are output, including the following steps: When a new apparent resistivity profile to be interpreted is input, the system feeds it into the currently optimized model for forward propagation inference. The network outputs a hierarchical probability map of the same size as the input, with each pixel containing the probability value of belonging to each geological layer. By taking the category corresponding to the highest probability of each pixel, a preliminary discrete hierarchical label map is obtained. Since the direct output may contain small isolated regions or jagged boundaries, the cells then perform morphological post-processing: first, morphological opening operation (erosion followed by dilation) is used to remove small spurious noise points, and then morphological closing operation (dilation followed by erosion) is used to bridge the small gaps between adjacent regions of the same category, thereby optimizing the continuity and smoothness of the layer boundaries, and finally outputting an automatically hierarchical result map with a clear structure. The anomaly detection unit employs the isolated forest unsupervised learning algorithm to automatically detect and delineate electrical anomalies that are significantly different from the background field based on a hierarchical structure. This includes the following steps: Feature Space Construction and Data Preparation: Based on the automatic stratification results, anomaly detection is performed on the data within each geological stratum (or the entire profile). Each spatial location is treated as a sample point, and a multi-dimensional feature vector is constructed for each sample point. The features mainly include: the apparent resistivity value at that location, the average resistivity of its stratum, the deviation of the resistivity at that location from the average value of its stratum, and the spatial gradient at that location. After normalization, the feature vectors of all sample points constitute the dataset for anomaly detection. ; Isolation Forest Construction and Path Length Calculation: An unsupervised isolation forest algorithm is used for anomaly detection. This algorithm recursively splits the data by randomly selecting features and split values ​​to construct multiple isolation trees (iTrees). When constructing each iTree, the dataset is used to... random subsampling For each sample point, the tree branches grow until each sample point is isolated or the maximum depth is reached. Outliers, due to their significantly different feature values ​​from the majority of points, are expected to be isolated at a shallower depth (i.e., after fewer segmentations). Its path length in a single iTree The average path length is defined as the number of edges traversed from the root node to its isolated leaf node, calculated across all trees. ; S3: Anomaly score calculation and anomaly identification. Based on average path length. Calculate sample points abnormal scores : ; in, Given the subsample size The average path length is used for standardization and outlier scores. The closer the score is to 1, the more likely the point is to be an anomaly. Set an anomaly threshold (such as 0.7 or determined statistically). Mark all sample points with anomaly scores exceeding the threshold as anomalies. Perform spatial clustering analysis on the marked anomalies in a two-dimensional profile or three-dimensional space. Merge spatially adjacent anomalies into the same anomaly body and calculate the boundary, center position, and spatial distribution range of each anomaly body to complete the automatic detection and delineation of anomalies. The attribute interpretation unit combines the anomaly's morphology, size, and resistivity range with knowledge base matching to provide a preliminary geological attribute interpretation, generating geological interpretation results and anomaly reports, including the following steps: Anomaly feature quantification and knowledge base retrieval. For each anomaly identified by the anomaly detection unit, its quantized feature description vector is extracted. The main features include: morphological features (such as aspect ratio, area / volume, and boundary curvature), scale features (such as length along the survey line and maximum thickness), electrical features (such as the average resistivity inside the anomaly, the ratio to the resistivity of the surrounding rock, and the spatial gradient of resistivity), and spatial relationship features (such as the geological stratum and burial depth). At the same time, the system maintains an expandable geological and geophysical knowledge base, which stores empirical values ​​or theoretical ranges for various typical geological targets in the above feature dimensions. Rule-based and similarity-based attribute matching. This involves matching the feature description vector of the anomaly. The system matches geological target templates with various data points in the knowledge base. The matching process combines rule-based reasoning and similarity calculation. Rule-based reasoning is based on explicit expert knowledge, such as extremely low resistivity and a flattened shape suggesting an aquifer. Similarity calculation uses weighted Euclidean distance or cosine similarity to measure the similarity. With template feature vector The degree of closeness, the matching scoring function can be expressed as: ; in, It is the feature dimension. It is the weight of the j-th feature. It is a similarity function. It is the number of rules. It is an indicator function (1 if the exception body satisfies the r-th rule, 0 otherwise). It is the rule weight; Generate interpretation results and comprehensive report: Based on the matching and scoring results, assign one or more possible geological attribute interpretations to each anomaly, and attach a confidence level rating (e.g., high, medium, low). The unit integrates all stratification results, delineated anomalies and their attribute interpretations, confidence levels, and other information to generate structured geological interpretation results; at the same time, it automatically generates an anomaly report containing key maps (such as stratified profile maps and anomaly distribution maps) and textual descriptions, clearly indicating potential geological targets, their spatial locations, and electrical characteristics, providing preliminary and quantitative reference for geological decision-making.

[0028] The data management and visualization module includes a spatiotemporal database unit, a 3D visualization engine unit, and a results integration and display unit; The spatiotemporal database unit adopts a hybrid database architecture to store and manage spatiotemporal data throughout the entire process from raw data to interpretation results, supporting efficient querying and backtracking, including the following steps: To achieve complete backtracking of processing history, the unit implements strict data version management. Any data processing or interpretation operation (such as modifying inversion parameters and recalculating) is considered to generate a new version. The derivation relationship between versions is recorded through a directed acyclic graph. The core version tracking algorithm maintains a version linked list and a parent version pointer for each data entity. When a user needs to backtrack to a certain historical state, the system can quickly reconstruct the data view at that moment based on the version graph. All data access operations are recorded in the audit log to ensure the traceability and integrity of data operations. The 3D visualization engine unit utilizes WebGL technology to build a browser-side 3D rendering engine, enabling dynamic and interactive visualization of survey line layout, data profiles, and 3D inversion models. This includes the following steps: Construction of a WebGL-based multi-source data integrated rendering scene: On the user's browser, a 3D rendering context is created using the WebGL 2.0 API and front-end graphics framework. This constructs a unified scene capable of simultaneously hosting and rendering surface terrain models, survey line layout geometry, 2D / 3D resistivity data profiles, and 3D anomaly interpretation volumes. First, the digital elevation model is converted into a triangular mesh terrain surface. Then, the survey line coordinates and electrode point positions are rendered as wireframes and point clouds. For the 2D resistivity profile, it is processed into texture patches perpendicular to the survey lines. For the 3D resistivity model, it is converted into isosurface meshes or scalar field data required for volume rendering using 3D texture slicing or moving cube algorithms. This unifies exploration data of different dimensions and types into the same 3D world coordinate system. Dynamic Scheduling and LOD Rendering of Large-Scale Geophysical Data: Due to the massive data volume of 3D resistivity models and other data, it is impossible to load all of it into video memory at once. This unit implements a dynamic data scheduling mechanism, preprocessing large-scale 3D volume data into multi-resolution pyramid tiles on the server side; on the client side, based on the current viewpoint's position, orientation, and view frustum clipping, it calculates in real time which data tiles are within the field of view and their required level of detail, loading the required tiles through asynchronous requests; for already loaded geometric data (such as terrain and isosurfaces), level of detail technology is applied: dynamically switching the mesh refinement used for rendering based on the distance between the object and the viewpoint, which is achieved through a... The core logic of the selection function implementation can be simplified as follows: ; in, It is the distance from the object to the camera. It is the vertical field of view. It is an adjustment coefficient, and this mechanism ensures a balance between smooth interactive frame rates and massive data visualization; Interactive analysis functionality and shader programming implementation: To achieve dynamic interactive visualization, the engine encapsulates a series of interactive functions: dragging the mouse to rotate / translate / zoom the scene, clicking to query the spatial coordinates and attribute values ​​of any point, cutting profiles, and adjusting the color map table; The results integration and display unit integrates and links the results of data collection, processing, inversion, and interpretation within the same spatiotemporal framework, including the following steps: The system constructs a spatiotemporal relational graph, where nodes represent various data entities and edges represent the relationships between them. The types of relationships include: spatial inclusion (e.g., a measuring point belongs to a certain measuring line), temporal sequence / derivation (e.g., original data A is processed to generate data B), spatial reference (e.g., an interpretation anomaly C is located in a certain area of ​​the inversion profile D), and logical association (e.g., anomaly E corresponds to geological template F in the knowledge base).

[0029] The quality control and output module includes a quality assessment unit, a standardized charting unit, and a report generation unit; The quality assessment unit calculates the confidence interval for each data point based on statistical principles, evaluates and marks suspicious or out-of-tolerance data in real time, and generates a data quality log, including the following steps: Data quality feature extraction and error model establishment: This unit receives preprocessed electrical resistivity data from the data acquisition and processing module. For each data point, it extracts its quality-related features, mainly including: the signal amplitude of the observed voltage value, the background noise level during the acquisition process, the standard deviation of repeated observations of the signal, and the status code of the data acquisition system's self-check feedback. Based on historical data and instrument calibration information, it establishes preliminary error estimation models for different types of observations, assigning a priori measurement error variance to each data point. ; Real-time quality assessment based on statistical confidence intervals: for a new data point (usually apparent resistivity) Within its nearest spatiotemporal window, calculate the average of all reliable data points. and standard deviation Construct a confidence interval for the data point; for data that roughly conforms to a normal distribution, its Confidence interval is ,in The quantiles are those of the standard normal distribution. Simultaneously, by incorporating prior error information, a comprehensive quality score is calculated, which can be expressed as: ; in, It is an indicator function (if the data point is in the confidence interval). (If it is inside, it is 1; otherwise, it is 0). These are weighting coefficients. and It is an adjustment parameter. The formula combines statistical outliers, relative bias, and instrument prior error to obtain a score between 0 and 1. Quality grading and log generation: based on comprehensive quality score The data points are divided into different quality levels, for example: Excellent ( ),good( ),suspicious( ) and out of tolerance / invalid ( For suspicious and out-of-tolerance data points, corresponding quality tags are added to the data stream or database. The unit automatically generates a structured data quality log, which records the overall quality statistics of each batch of data by time or measurement line, providing key quality information for subsequent processing and report generation. The standardized mapping unit has built-in map templates that conform to industry standards, and automatically draws apparent resistivity cross-sections, contour maps and 3D renderings based on the inversion and interpretation results; The report generation unit integrates text, graphics, tables, and quality logs, and automatically generates a comprehensive results report according to a preset template.

[0030] The system interaction module includes a Web graphical interface unit, a process control API unit, and a user management and logging unit; The Web graphical interface unit provides a browser-based interface for task creation, parameter setting, process monitoring, result viewing, and interactive analysis. The process control API unit provides an application programming interface, enabling users to submit batch tasks, customize process orchestration, and integrate with the system via scripts; The User Management and Logs unit is responsible for user authentication, access control, and the recording and auditing of all system operation logs.

[0031] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus.

[0032] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A multi-channel high-density electrical resistivity tomography (OTT) data acquisition and processing integrated system, characterized in that, The system includes: Data acquisition module: Through synchronous excitation technology based on pseudo-random coding and multi-channel parallel acquisition technology, it drives multiple pairs of electrodes and synchronously acquires full-waveform potential difference signals to generate raw electrical resistivity data; Data processing module: Through adaptive filtering and wavelet transform techniques, the raw electrical resistivity data is denoised, baseline corrected, and spike anomaly suppressed in real time to generate preprocessed electrical resistivity data; Data fusion and inversion module: Through dynamic data fusion technology based on Kriging interpolation and least squares inversion algorithm with terrain constraints, preprocessed electrical resistivity data from different device modes are fused to generate high-resolution two-dimensional and three-dimensional apparent resistivity profiles. Intelligent interpretation and anomaly identification module: Based on convolutional neural network and isolated forest algorithm, it automatically performs geological stratification, anomaly delineation and attribute identification on apparent resistivity profile, and generates geological interpretation results and anomaly reports; Data Management and Visualization Module: Employing a spatiotemporal database and a WebGL 3D rendering engine, it provides integrated storage, management, and dynamic 3D visualization of data throughout the entire process of acquisition, processing, and interpretation. Quality control and output module: Through a confidence interval-based data quality assessment algorithm and a standardized template engine, it monitors data quality in real time and automatically generates output charts and reports that conform to industry standards; System interaction module: Provides a web-based graphical user interface and API interface for task configuration, process control, parameter adjustment and result interaction.

2. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The data acquisition module includes a channel management unit, a synchronization activation unit, and a parallel acquisition unit; The channel management unit is used to configure and manage the channel connection status and working mode of up to hundreds of electrodes, and to achieve rapid switching of electrode paths through a high-speed electronic switch matrix. The synchronous excitation unit uses pseudo-random sequence coding technology and a high-precision constant current source to generate and synchronously apply excitation current to the designated electrode pair in order to reduce environmental noise interference and improve the signal-to-noise ratio. The parallel acquisition unit uses a high-precision analog-to-digital converter and field-programmable gate array technology to synchronously and in parallel acquire the full waveform potential difference signals of all measurement electrode pairs, generating raw electrical resistivity data.

3. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The data processing module includes a real-time noise reduction unit, a signal correction unit, and a data packaging unit; The real-time noise reduction unit uses an adaptive filtering algorithm based on the reference electrode to identify and suppress power frequency interference and random noise in real time. The signal correction unit analyzes the signal baseline drift through wavelet transform and performs automatic correction, while using the threshold method to suppress transient spike anomalies generated during the acquisition process. The data packaging unit adds timestamps, electrode positions, and device parameter metadata to the noise-reduced and corrected signals, and packages them to generate preprocessed electrical resistivity data in a standard format.

4. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The data fusion and inversion module includes a dynamic weighted fusion unit, a fast inversion calculation unit, and a result optimization unit. The dynamic weighted fusion unit performs dynamic weighted fusion based on the signal-to-noise ratio and spatial sampling density of data from different device modes, using the Kriging interpolation algorithm to generate a fused data field. The fast inversion calculation unit uses a least squares optimization inversion algorithm with terrain undulation constraints to perform fast iterative inversion calculations on the fused data field and generate two-dimensional and three-dimensional apparent resistivity distribution models. The result optimization unit optimizes the inversion results using regularization smoothing technology to reduce false anomalies and generate a high-resolution apparent resistivity profile.

5. The integrated system for multi-channel high-density electrical resistivity tomography (OTT) data acquisition and processing according to claim 4, characterized in that: The multi-device data fusion in the dynamic weighted fusion unit includes the following steps: S1: Data spatial registration: Mapping preprocessed electrical resistivity data collected from different device modes onto regular grid nodes under the same measurement coordinate system; S2: Dynamic weight calculation: For each grid node, the Kriging interpolation weight coefficient is dynamically calculated based on the spatial density, signal-to-noise ratio, and distance from the surrounding data from different sources. S3: Fusion Interpolation Execution: Using the calculated dynamic weights, weighted interpolation is performed on data from multiple devices at grid nodes to generate a spatially continuous and information-complete fused data field; S4: Fusion quality verification: Calculate the spatial continuity index and variance of the fused data field. If the threshold is not reached, return to S2 to adjust the weight calculation strategy.

6. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The intelligent interpretation and anomaly recognition module includes an automatic hierarchical unit, an anomaly detection unit, and an attribute interpretation unit; The automatic stratification unit is based on a pre-trained convolutional neural network model and automatically identifies the electrical stratigraphic boundaries in the apparent resistivity profile to complete the automatic geological stratification. The anomaly detection unit uses the isolated forest unsupervised learning algorithm to automatically detect and delineate electrical anomalies that are significantly different from the background field on a hierarchical basis. The attribute interpretation unit combines the anomaly's shape, size, and resistivity range with knowledge base matching to provide a preliminary geological attribute interpretation, generating geological interpretation results and anomaly reports.

7. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 6, characterized in that: The training and online optimization of the pre-trained convolutional neural network model in the automatic hierarchical unit includes the following steps: S1: Offline model pre-training: Collect a large amount of existing manually interpreted and labeled typical apparent resistivity profile data as a training set, train a U-Net structured convolutional neural network model, and learn the mapping relationship from resistivity images to hierarchical labels. S2: Online Incremental Learning: In practical applications, the system uses new interpretations that have been confirmed by experts and have high credibility as new samples, and employs incremental learning algorithms to fine-tune and optimize the network model to adapt it to the new geophysical characteristics of the work area. S3: Inference and Post-processing: The apparent resistivity profile to be interpreted is input into the optimized network model for forward propagation to obtain a preliminary hierarchical probability map. Then, the boundary continuity is optimized through morphological operations, and finally, the automatic hierarchical results are output.

8. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The data management and visualization module includes a spatiotemporal database unit, a 3D visualization engine unit, and a results integration and display unit. The spatiotemporal database unit adopts a hybrid database architecture to store and manage spatiotemporal data throughout the entire process from raw data to interpretation results, supporting efficient querying and backtracking. The three-dimensional visualization engine unit uses WebGL technology to build a browser-side three-dimensional rendering engine to realize dynamic interactive visualization of survey line layout, data profile, and three-dimensional inversion model; The results integration and display unit integrates and links the results of data collection, processing, inversion, and interpretation within the same spatiotemporal framework for display.

9. The integrated system for multi-channel high-density electrical resistivity tomography (OTT) data acquisition and processing according to claim 1, characterized in that: The quality control and output module includes a quality assessment unit, a standardized charting unit, and a report generation unit. The quality assessment unit calculates the confidence interval for each data point based on statistical principles, evaluates and marks suspicious and out-of-tolerance data in real time, and generates a data quality log. The standardized mapping unit has built-in map templates that conform to industry standards, and automatically draws apparent resistivity cross-sections, contour maps and three-dimensional renderings based on the inversion and interpretation results. The report generation unit integrates text, graphics, tables, and quality logs, and automatically generates a comprehensive results report according to a preset template.

10. The integrated system for multi-channel high-density electrical resistivity tomography data acquisition and processing according to claim 1, characterized in that: The system interaction module includes a Web graphical interface unit, a process control API unit, and a user management and log unit. The Web graphical interface unit provides a browser-based interface for task creation, parameter setting, process monitoring, result viewing, and interactive analysis. The process control API unit provides an application programming interface, enabling users to submit batch tasks, customize process orchestration, and integrate with the system via scripts; The User Management and Logs unit is responsible for user authentication, access control, and the recording and auditing of all system operation logs.

Images