High-density direct current method array type detection device and data fusion method

Abstract

The invention discloses a high-density direct current method array type detection device and a data fusion method, and relates to the technical field of geophysical exploration engineering. A data acquisition module; a signal preprocessing module; a space inversion module; a feature extraction module; an imaging and alarm module; potential data, contact impedance information and electrode positioning confidence which are obtained by multiple injection-measurement sequences are fused, and through multi-scale purification, local inversion and uncertainty perception fusion, the method is oriented to application scenes such as urban pipeline positioning, surface layer mineral exploration, underground water and archaeological detection and the like. A closed-loop optimization measurement process from data acquisition to imaging-scheduling is realized, a local inversion result and uncertainty thereof are taken as fusion weights, and spatial connectivity is maintained in a fusion process, so that artifacts caused by isolated noise can be effectively inhibited, and a real geometric shape of an anomalous body is maintained; therefore, the robustness and the engineering interpretability of an imaging result are remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of geophysical exploration engineering technology, specifically to a high-density DC electrical resistivity array detection device and data fusion method. Background Technology

[0002] High-density direct current resistivity tomography (DCS) has been widely used in engineering scenarios such as urban pipeline location, near-surface mineral exploration, and groundwater and archaeological detection due to its advantages of high resolution and wide coverage. The core of this technology is to inject DC excitation into an electrode array and collect potential signals, then process the data and perform inversion imaging to obtain the conductivity distribution of the underground medium, thereby identifying target anomalies.

[0003] Existing high-density DC resistivity array detection devices and data processing methods still face significant technical bottlenecks in practical engineering applications. A key shortcoming lies in the fact that current technologies often employ fixed injection paths and measurement plans, neglecting the uncertainty distribution of inversion results and the differences in confidence levels regarding contact impedance and electrode positioning during data fusion. They rely solely on a single data source or fixed weights for inversion imaging and anomaly identification. This results in the detection process being unable to dynamically adapt to the complexity of the subsurface medium, isolated noise easily forming artifacts, and difficulty in accurately preserving the true geometric shape and amplitude characteristics of anomalies. This not only reduces the robustness and engineering interpretability of the imaging results but also makes them susceptible to misinterpretation due to errors from a single data source, increasing the risk of engineering decisions.

[0004] In response to this problem, this application proposes a high-density DC current array detection device and data fusion method to solve the above-mentioned problems. Summary of the Invention

[0005] The purpose of this invention is to provide a high-density DC current-based array detection device and data fusion method to solve the problems in the existing technology where the detection process cannot dynamically adapt to the complexity of the underground medium, isolated noise is prone to forming artifact interference, and the true geometric shape and amplitude characteristics of anomalies are difficult to accurately preserve.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] In a first aspect, this application provides a high-density DC electrical array detection device, comprising:

[0008] The electrode array module is used to arrange multiple rows of injection / measurement electrodes in a high-density pattern and inject controllable DC excitation to obtain the original potential field.

[0009] The data acquisition module is used to align the original potential field in time and space to obtain aligned potential data;

[0010] The signal preprocessing module is used to perform contact impedance compensation and multi-scale noise suppression on the alignment potential data to obtain a purified potential field.

[0011] The spatial inversion module is used to perform spatial inversion on the purification potential field based on physical constraints to obtain a candidate map of conductivity distribution.

[0012] The feature extraction module is used to extract the geometric and amplitude parameters of the anomaly from the candidate conductivity distribution map to obtain the anomaly parameters;

[0013] The imaging and alarm module is used to generate localization images based on the abnormal parameters and output alarm and visualization results.

[0014] Furthermore, the electrode array module employs staggered injection paths and programmable segment spacing to generate depth-to-shallow differential sensitivity in adjacent measurements, thereby improving the deep response signal-to-noise ratio and preserving previous injection path information in subsequent inversions to assist in interlayer differentiation.

[0015] Furthermore, the data acquisition module synchronizes with the global time reference by fusing electrode position tracking (optical or magnetic positioning) to eliminate electrode positioning errors and clock drift, thereby obtaining high-precision time-space aligned potential data, which is then used as input for subsequent purification and inversion.

[0016] Furthermore, the signal preprocessing module employs synchronous demodulation and local baseline adaptive estimation based on a reference electrode grid to suppress contact impedance and low-frequency environmental drift while preserving spatial details, thereby obtaining a stable purification potential field. The signal preprocessing module further decomposes the signal at multiple scales to decouple noise and coupling terms at different spatial scales to avoid information loss.

[0017] Furthermore, the spatial inversion module is based on a hierarchical-local coupled inversion process with prior physical boundary constraints. It is used to first lock the global conductivity trend with low-resolution tomography, and then refine it layer by layer in the high uncertainty region using local correction kernels to obtain the candidate conductivity distribution map that has both global stability and local resolution; and to record the uncertainty distribution for adaptive measurement scheduling.

[0018] Furthermore, the feature extraction module uses morphological consistency filtering and spatial coherence clustering to separate geometric connectivity information from candidate images and calculate the volume, peak contrast and connectivity index of anomalies to obtain structured anomaly parameters, thereby significantly reducing the false alarm rate based on isolated noise points.

[0019] Furthermore, the device also includes an adaptive measurement scheduling module, which is used to dynamically adjust the subsequent injection / measurement sequence based on the uncertainty distribution of the inversion module and the abnormal parameters obtained by the feature extraction (e.g., prioritize increasing injection samples in high uncertainty or abnormal regions), in order to maximize the detection sensitivity and positioning accuracy of the target region under a limited number of measurements, and obtain detection results with significantly improved measurement efficiency.

[0020] Secondly, this application provides a high-density DC electrical resistivity array-based detection data fusion method, applied to the high-density DC electrical resistivity array-based detection device described in the first aspect, the method comprising the following steps:

[0021] S1. Acquire raw potential signals from multiple injection / measurement sequences and different injection modes of the electrode array to obtain multiple sets of raw potential field data;

[0022] S2. Based on the electrode position and global time reference, perform position and time registration on the multiple sets of original potential field data to obtain aligned potential data;

[0023] S3. Perform contact impedance estimation and local baseline correction on the alignment potential data, and decompose the decoupled noise and coupling terms according to the spatial scale to obtain the purification potential field;

[0024] S4. Perform inversion on the purification potential field at the partition or layer scale and record the uncertainty distribution of each inversion result to obtain several local conductivity maps with uncertainty.

[0025] S5. Based on the local conductivity map with uncertainty, perform consistency correction and fusion at the spatial level to output the joint conductivity distribution and corresponding uncertainty map;

[0026] S6. Based on the joint conductivity distribution and uncertainty map, generate positioning imaging and structured anomaly parameters, and output a priority list for subsequent adaptive measurement scheduling.

[0027] Furthermore, in step S5, the uncertainty-aware data fusion adopts a multi-scale weighted fusion strategy from coarse to fine. The weights are jointly determined by the inversion uncertainty of each local conductivity map, the confidence level of the contact impedance estimation of the corresponding measurement, and the confidence level of the electrode position / attitude.

[0028] Compared with existing technologies, this invention provides a high-density DC electrical resistivity array detection device and data fusion method. This invention integrates potential data obtained from multiple injection-measurement sequences, contact impedance information, and electrode positioning confidence. Through multi-scale purification, local inversion, and uncertainty-aware fusion, it addresses application scenarios such as urban pipeline location, near-surface mineral exploration, groundwater and archaeological detection. It achieves a closed-loop optimized measurement process from data acquisition to imaging and scheduling, using local inversion results and their uncertainties as fusion weights and maintaining spatial connectivity during the fusion process. This effectively suppresses artifacts caused by isolated noise and preserves the true geometric shape of anomalies, thereby significantly improving the robustness and engineering interpretability of the imaging results. This invention enables the orderly integration of information obtained from different acquisition modes and regional inversions at various scales, reducing misinterpretations caused by errors from a single data source. It facilitates engineering decision-makers in scheduling remeasurements or drilling based on a confidence-driven priority list, forming an auditable chain of imaging evidence. Attached Figure Description

[0029] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0030] Figure 1 A block diagram of a high-density DC current array detection device provided in an embodiment of the present invention;

[0031] Figure 2 A flowchart of a high-density DC-DC array-based detection data fusion method provided in an embodiment of the present invention. Detailed Implementation

[0032] To enable those skilled in the art to better understand the technical solution of the present invention, the present invention will be further described in detail below with reference to the accompanying drawings.

[0033] As attached Figure 1 As shown:

[0034] Example 1:

[0035] A high-density DC current-based array detection device includes:

[0036] The electrode array module is used to arrange multiple rows of injection / measurement electrodes in a high-density pattern and inject controllable DC excitation to obtain the original potential field.

[0037] Specifically, the electrode array module employs staggered injection paths and programmable segment spacing to generate depth-to-shallow differential sensitivity in adjacent measurements, thereby improving the deep response signal-to-noise ratio and preserving previous injection path information in subsequent inversions to assist in interlayer differentiation.

[0038] Furthermore, linear, area, or hybrid arrays are employed, with injection modes including programmable combinations of short, medium, and long spans; each mode is repeatedly measured to obtain sample stability. Data is acquired by a high-resolution ADC, along with the injected current, electrode pair identification, electrode number, environmental label, and initial contact impedance for each injection. Implementation details include setting the injection current amplitude and duration, selecting the steady-state window, calibrating the sampling rate and quantization depth, and on-site quality control (dry testing, reference electrode measurement, and rejection of abnormal measurements).

[0039] The data acquisition module is used to align the original potential field in time and space to obtain aligned potential data;

[0040] Specifically, the data acquisition module synchronizes with the global time reference by fusing electrode position tracking (optical or magnetic positioning) to eliminate electrode positioning errors and clock drift, thereby obtaining high-precision time-space aligned potential data, which is then used as input for subsequent purification and inversion.

[0041] Furthermore, multiple sets of observations are unified to the same spatial coordinate system and time reference. Spatial alignment determines the three-dimensional coordinates of the electrodes and records their attitude using total station, RTK-GNSS, ground laser, or optical calibration; temporal alignment achieves sub-millisecond synchronization via GPS / PPS, NTP, or hardware triggering. Alignment processing includes interpolation / resampling to correct sampling rate differences, correcting the signal reference direction based on electrode coordinates and attitude, and packet loss or anomalous sampling repair strategies. The output is an ordered potential dataset with precise coordinates and timestamps, providing a unified input for subsequent compensation and inversion.

[0042] The signal preprocessing module is used to perform contact impedance compensation and multi-scale noise suppression on the alignment potential data to obtain a purified potential field.

[0043] Specifically, the signal preprocessing module employs synchronous demodulation and local baseline adaptive estimation based on a reference electrode grid to suppress contact impedance and low-frequency environmental drift while preserving spatial details, thereby obtaining a stable purification potential field. The signal preprocessing module further decomposes the signal at multiple scales to decouple noise and coupling terms at different spatial scales to avoid information loss.

[0044] Furthermore, the prior contact impedance of each electrode is first obtained (4-line measurement or short-time steady-state estimation), and the confidence level is updated according to the number of measurements. Compensation is performed to eliminate the amplitude shift introduced by the contact through local gain correction or synchronous demodulation based on a reference network. Cleaning employs multi-scale processing (e.g., wavelet-based or block filtering) to distinguish between short-scale spatial noise, local coupling interference, and far-field low-frequency drift, using adaptive filtering, baseline subtraction, and neighborhood decoupling respectively. The goal is to preserve spatial edges and anomalous details while denoising, and output a noise / confidence estimate for each pixel for subsequent weight calculation.

[0045] The spatial inversion module is used to perform spatial inversion on the purification potential field based on physical constraints to obtain a candidate map of conductivity distribution.

[0046] Specifically, the spatial inversion module is based on a hierarchical-local coupled inversion process with prior physical boundary constraints. It is used to first lock the global conductivity trend with low-resolution tomography, and then refine it layer by layer in the high uncertainty region using local correction kernels to obtain the candidate conductivity distribution map that has both global stability and local resolution; and to record the uncertainty distribution for adaptive measurement scheduling.

[0047] Furthermore, the observation domain is divided into overlapping or hierarchical local blocks. For each block, a fast and scalable inversion strategy (e.g., low-resolution tomography first, followed by local refinement) is employed, and the inversion residuals and stability indices are recorded. To estimate uncertainty, perturbation resampling, parameter perturbation, or parallel small-scale Monte Carlo experiments can be used to obtain pixel-level or voxel-level standard deviations / confidence intervals. Key implementation points include selecting an appropriate regularization intensity, an adaptive block partitioning strategy to control computational load, and storing the uncertainty in a unified format in the results file for easy fusion and use.

[0048] The feature extraction module is used to extract the geometric and amplitude parameters of the anomaly from the candidate conductivity distribution map to obtain the anomaly parameters;

[0049] Specifically, the feature extraction module uses morphological consistency filtering and spatial coherence clustering to separate geometric connectivity information from candidate images and calculate the volume, peak contrast and connectivity index of anomalies to obtain structured anomaly parameters, thereby significantly reducing the false alarm rate based on isolated noise points.

[0050] Furthermore, based on the local inversion outputs and their uncertainties, a weighted fusion is performed in both space and scale. The weights are jointly determined by the inversion uncertainty, contact impedance confidence, and electrode position / attitude confidence. The fusion employs a coarse-to-fine multi-scale strategy to first lock in macroscopic trends and then refine local details. Spatial connectivity preservation (such as neighborhood consistency constraints or graph cut smoothing) is introduced during the fusion process to avoid isolated artifacts, and a confidence-first or parallel re-inversion strategy is adopted for conflicting regions. The final output is a joint conductivity distribution and corresponding uncertainty field for engineering interpretation and scheduling decisions.

[0051] The imaging and alarm module is used to generate localization images based on the abnormal parameters and output alarm and visualization results.

[0052] Specifically, anomaly detection and structured description are performed on the joint distribution: depth, volume, peak comparison, and connectivity indicators of anomalies are extracted through thresholding, morphological segmentation, and connectivity analysis; confidence and priority are calculated for each anomaly. Visualization results (isosurfaces, profiles, planar plots, KML / GeoJSON export) are generated, and a coordinate-based list of re-measurement / drilling priorities is output for closed-loop adaptive measurement. The interface supports writing the results back to the measurement and control system to automatically adjust subsequent injection / measurement sequences, achieving a measurement-inversion-scheduling closed loop.

[0053] Specifically, the device also includes an adaptive measurement scheduling module, which is used to dynamically adjust the subsequent injection / measurement sequence based on the uncertainty distribution of the inversion module and the abnormal parameters obtained by the feature extraction (e.g., prioritize increasing injection samples in high uncertainty or abnormal regions), in order to maximize the detection sensitivity and positioning accuracy of the target region under a limited number of measurements, and obtain detection results with significantly improved measurement efficiency.

[0054] As can be seen from the above, this invention integrates potential data obtained from multiple injection-measurement sequences, contact impedance information, and electrode positioning reliability. Through multi-scale purification, local inversion, and uncertainty perception fusion, it aims to achieve a closed-loop optimized measurement process from data acquisition to imaging and scheduling for application scenarios such as urban pipeline positioning, near-surface mineral exploration, and groundwater and archaeological exploration.

[0055] By using local inversion results and their uncertainties as fusion weights and maintaining spatial connectivity during the fusion process, artifacts caused by isolated noise can be effectively suppressed, and the true geometric shape of anomalies can be preserved, thereby significantly improving the robustness and engineering interpretability of imaging results. This invention enables the orderly integration of information obtained from different acquisition modes and regional inversions at a scale, reducing misinterpretations caused by errors from a single data source. It facilitates engineering decision-makers in arranging re-measurements or drilling based on a confidence-driven priority list, forming an auditable chain of imaging evidence.

[0056] Example 2:

[0057] Location of shallow-buried public utility pipelines in urban areas:

[0058] Objectives and scenarios:

[0059] When inspecting metal pipelines and buried components at a depth of ≤5m in urban blocks, it is necessary to consider both spatial resolution and stability under the influence of noise.

[0060] On-site deployment and data collection:

[0061] Electrode array: a linear array of 2 rows with staggered layout, totaling 120 electrodes with equal spacing of 0.5m (covering a length of approximately 59.5m).

[0062] Injection / measurement sequence: staggered overlapping injection is used: each time, four injection path combinations (short-range / medium-range / overlapping span) are selected between two adjacent columns, and each injection combination is measured three times to take the average. A total of 240 injection-measurement pairs are used in a single measurement cycle.

[0063] Injection current: constant current 100mA ADC (±1%), each injection lasts 2.0s (steady-state observation window 1.5s).

[0064] Sampling and acquisition equipment: 24-bit ADC, sampling rate 200Hz, measured potential is output as an average value in a 100ms window.

[0065] Initial contact impedance measurement: Perform 4-wire contact impedance measurement on each electrode and record the average contact impedance (typically 5–200Ω, commonly 20–120Ω on urban roads).

[0066] in:

[0067] Electrode spacing (0.5m): affects lateral spatial resolution. It is determined on-site by the designer and confirmed by measuring tape and laser rangefinder.

[0068] Injection current (100mA): Ensures sufficient signal amplitude while meeting on-site safety regulations; set and recorded by a constant current source.

[0069] Injection duration (2.0s): Ensures steady-state potential is reached, controlled and recorded by the device firmware at regular intervals.

[0070] ADC sampling rate and resolution: determined by the data acquisition card specifications (device factory specifications), used to calculate the observation average and noise statistics.

[0071] Contact impedance: Measured electrode by electrode using a 4-wire measuring instrument on site, serving as the basis for subsequent contact impedance compensation and confidence assessment.

[0072] Spatiotemporal alignment:

[0073] Electrode position measurement: A total station (optical measurement) and on-site reference stakes were used. The electrode coordinate accuracy was ±5mm (plane) / ±10mm (elevation), and the attitude label of each electrode was recorded.

[0074] Global time base: Uses GPS / local NTP to synchronize device clocks, with time synchronization accuracy ≤1ms.

[0075] Output: Aligned potential data file (including electrode coordinates and timestamps for each measurement).

[0076] Contact Impedance Compensation and Multi-Scale Cleaning

[0077] Contact impedance estimation: Based on the injected current and the measured potential difference each time, local corrections and confidence scores are made in conjunction with the initial contact impedance correction table.

[0078] Multi-scale purification: First, low-frequency drift baseline correction is performed (removing long-term drift magnitudes of 0.1–5mV), and then the local white noise and coupling interference are suppressed according to spatial scales (0.5m / 2m / 5m).

[0079] Output: Purification potential field (including noise estimation per pixel).

[0080] Local inversion and uncertainty record:

[0081] The observation domain is divided into several overlapping partitions (each partition contains approximately 30–50 electrode data points). A fast linearization inversion is performed on each partition to obtain a local conductivity map, and the uncertainty is estimated using Monte Carlo-type random perturbations (the standard deviation estimate for each pixel is recorded).

[0082] Record the confidence index for each inversion (e.g., inversion residual, iterative stability marker).

[0083] Uncertainty-aware data fusion:

[0084] A multi-scale coarse-fine weighted fusion method is adopted (weights are derived from local inversion uncertainty, contact impedance confidence, and electrode position confidence). Spatial connectivity (continuity of anomaly boundary) is forcibly maintained during the fusion process to avoid artifacts introduced by isolated high-confidence regions.

[0085] Output: Joint conductivity distribution map and corresponding uncertainty field.

[0086] Imaging and scheduling output:

[0087] Extract anomaly parameters (volume, maximum contrast, connectivity index) from the joint map; generate a positioning report (latitude and longitude / station coordinates), an imaging map, and a list of priority survey areas (for further refined on-site measurements).

[0088] Actual on-site measurement data:

[0089] Test target: Metal pipe sections buried at a depth of 0.8–3.2m.

[0090] Detection cycle (acquisition + processing): Acquisition takes 6 hours (including deployment, injection, and measurement cycles), and offline processing (all steps) takes approximately 35 minutes per dataset (desktop 8-core CPU).

[0091] On-site noise observation (before purification): DC drift ±2.1mV; short-time noise standard deviation within the measurement period 0.6mV.

[0092] Mean contact impedance: 68Ω; coefficient of variation: 0.45.

[0093] Mean local inversion uncertainty (pixel level): 0.12 S / m (standard deviation).

[0094] Results: Average positioning error 1.1m; detection probability 92%; false alarm rate 6%.

[0095] Example 3:

[0096] Near-surface deep target exploration:

[0097] Objectives and scenarios:

[0098] Exploration of mineral / aquifer anomalies, with a target depth of 5–30m, requires maintaining good depth sensitivity and uncertainty control over a large coverage area.

[0099] On-site deployment and data collection:

[0100] Electrode array: 20×20 area grid, 400 electrodes in total, grid spacing 1.0m (covering an area of ​​about 19×19m).

[0101] Injection / measurement sequence: A programmable segmented spacing injection strategy (“interlaced injection and programmable segmented spacing” described in claim 2) was adopted, including three types of injection: short span (1m), medium span (3m) and long span (8–12m), for a total of 480 injection-measurement pairs. The long span was used to improve the sensitivity to deep parts.

[0102] Injection current: constant current 200mA ADC (±1%), injection duration 3.0s (steady-state window 2.5s).

[0103] Sampling and Equipment: 24-bit ADC, sampling at 100Hz, with a steady-state window of 250ms for averaging the output.

[0104] Initial contact impedance measurement: 4-wire measurement, the average contact impedance of relatively loose soil in the field is 30Ω (range 8–90Ω).

[0105] Grid spacing (1.0m): While ensuring depth coverage, a certain lateral resolution is guaranteed. This is set through topographic surveying and stake placement.

[0106] Long-span injection (8–12m): This is achieved by selecting the combination of distal electrodes on the array using a programmable injection device, thereby increasing the deep response component.

[0107] Injection current (200mA): To increase the amplitude of deep signals, a higher current is selected (within the range of equipment and safety permits).

[0108] Contact impedance measurement and position measurement: Ground RTK-GNSS (open field) combined with ground laser scanning is used, and the electrode positioning accuracy is ±2cm.

[0109] Spatiotemporal alignment:

[0110] The electrode positions are fused from RTK-GNSS and ground scanning, and the time reference is synchronized using a hardware PPS (pulses per second) signal with a time accuracy of ≤0.5ms.

[0111] Output: High-precision aligned potential data (including electrode coordinates and timestamps for each measurement).

[0112] Contact Impedance Compensation and Multi-Scale Cleaning

[0113] The contact impedance is based on the initial measurement and fine-tuned using short-time steady-state differential estimation after each injection; the multi-scale decomposition scale is 1m / 3m / 10m, which are used to suppress electrode level noise, local coupling and far-field environment drift respectively.

[0114] The ambient noise level was low, and the standard deviation of pixel noise after purification was approximately 0.18mV.

[0115] Local inversion and uncertainty record:

[0116] The array is divided into 4×4 overlapping blocks (approximately 100 electrodes per block). The overall conductivity trend is first determined by low-resolution tomography, and then local refinement inversion is added in the high uncertainty region (uncertainty threshold > 0.25 S / m); a pixel-level uncertainty map is recorded.

[0117] Uncertainty-aware data fusion:

[0118] Multi-scale coarse-to-fine weighted fusion: The weights are generated from three confidence terms (local inversion uncertainty, contact impedance confidence, and electrode position confidence) (the weight vector is obtained through normalization mapping), and spatial connectivity is forcibly preserved during fusion (neighborhood consistency constraint) to avoid isolated high-confidence noise points forming artifacts.

[0119] Output: Joint conductivity distribution map (covering 19×19m) and uncertainty field, with high-priority areas to be retested marked.

[0120] Imaging and scheduling output:

[0121] Anomaly parameter extraction: depth estimation (5–28m range), volume estimation, peak contrast, and connectivity indices. Generate priority-based recommendations for resurvey / sampling well locations.

[0122] Actual on-site data testing:

[0123] Target: Ore bodies with lower conductivity than the surrounding sedimentary body (depth 8–22 m).

[0124] Data collection cycle: Data collection including deployment takes about 14 hours; offline processing (all steps) takes about 55 minutes per dataset (desktop 16-core CPU + GPU acceleration for parallel inversion).

[0125] On-site noise observation (before purification): DC drift ±0.6mV; short-time noise standard deviation 0.25mV.

[0126] Mean contact impedance: 30Ω; coefficient of variation: 0.32.

[0127] Mean local inversion uncertainty (pixel level): 0.09 S / m (standard deviation).

[0128] Results (see table): Average positioning error 0.9m; Detection probability 95%; False alarm rate 4%.

[0129] Comparative test instructions and data acquisition methods:

[0130] Baseline: Conventional linear injection + single-shot global inversion method (excluding multi-scale cleanup, local inversion uncertainty recording, and the aforementioned uncertainty-aware weighted fusion and spatial connectivity constraints). Baseline parameters are matched to the field as closely as possible (same number of electrodes and injection current), but multiple injection mode differential and adaptive strategies are not employed.

[0131] All data were acquired using the same set of acquisition hardware or equipment of equivalent specifications; a dedicated 4-wire measuring instrument was used for contact impedance measurement; electrode coordinates were measured by a total station (Example 2) or RTK-GNSS (Example 3); environmental noise was obtained by statistical analysis of drift between measurements (without injection); and inversion uncertainty was estimated and recorded by the local perturbation method.

[0132] To ensure repeatability, each method was repeated three times under the same target and deployment conditions, and the average result was taken as the final indicator.

[0133] A comprehensive comparison is shown in Table 1 below;

[0134] Table 1

[0135] Indicators / Methods Example 2 (Shallow Burial in Urban Areas) Example 3 (Deep Exploration) Baseline (conventional single-shot global inversion) Number of covered electrodes 120 400 Same as the corresponding embodiment Injection - Logarithmic measurement (times) 240 480 Same quantity (for comparability) Injected current (mA) 100 200 same Average contact resistance (Ω) 68 30 Same as the on-site measurement Total time for data acquisition and processing (data acquisition does not include deployment; processing is offline). 6 h / 35 min 14 h / 55 min 6 h / 20 min (faster processing) Positioning error (m, average) 1.1 0.9 3.2 Detection probability (%) 92 95 78 False alarm rate / False positive (%) 6 4 18 Average uncertainty decreased (compared to baseline, %) 35% 48% — SNR improvement (approximately compared to baseline) +6 dB +9 dB 0 dB Handling computational burden (relative) Moderate (approximately 1.6 × baseline) High (approximately 2.7 times the baseline) 1× Prioritize test point reduction (at the same confidence level) Saves approximately 28% of retesting points Saves approximately 42% of retesting points —

[0136] In the table:

[0137] Positioning error: The average Euclidean distance between the ground-measured marked point (known burial depth / coordinates) and the nearest point shown in the imaging result; the true value is obtained through ground measurement / excavation or drilling verification.

[0138] Detection probability: The proportion of targets successfully detected among a number of known targets (drilled / recorded); a true positive is defined as an imaging anomaly with a true value overlap of ≥50%.

[0139] False alarm rate: The percentage of locations where no real target is detected as abnormal, based on on-site verification and subsequent excavation / drilling confirmation.

[0140] Uncertainty reduction: The percentage difference between the baseline and the average pixel uncertainty output by this method, calculated from the pixel-level uncertainty mapping.

[0141] SNR improvement: estimated by the change in the ratio of the measured target peak value to the standard deviation of the ambient noise (in dB), obtained from observations before and after purification.

[0142] Computational burden: expressed as processing time ratio (on the same processor), including fusion and uncertainty estimation overhead.

[0143] Comparative analysis:

[0144] Differential sensitivity between shallow and deep layers and composite injection design: By interleaved injection and programmable segment spacing, different sensitivity components for shallow and deep responses can be obtained in a single observation, improving the detectability of deep targets (SNR improved by about +9dB in Example 3).

[0145] Robust observations are achieved through contact impedance and multi-scale cleanup: prior contact impedance measurement and local adaptive correction, combined with multi-scale noise decoupling, significantly reduce artifacts introduced by poor contact and low-frequency drift (false alarm rate in Example 2 decreased from 18% to 6%).

[0146] Local inversion and recording of uncertainties make fusion more reliable: recording uncertainties and using them as fusion weight inputs avoids the influence of single strong noise points on global imaging, improving the interpretability and engineering usability of fusion results; the uncertainty is reduced by an average of 35–48%.

[0147] Multi-scale weighted fusion for uncertainty perception: By introducing electrode position confidence and contact confidence during fusion and forcing the preservation of spatial connectivity, it can reduce artifacts while maintaining the geometric shape of the anomalous body, thereby improving the accuracy of engineering interpretation.

[0148] Adaptive measurement scheduling (driven by uncertainty output): By outputting a priority list, subsequent measurement resources are concentrated in areas of high uncertainty / high anomaly, thereby improving efficiency with limited resources (Example 3 reduces unnecessary retest points by approximately 42%).

[0149] The above data are typical values ​​from the field. Actual values ​​may be affected by soil conductivity, weather, surface cover, etc. All values ​​in the table are averaged based on three repeated field measurements.

[0150] As shown above, the staggered injection strategy generates complementary responses of varying depths, and the uncertainty of the inversion output guides subsequent injection / measurement scheduling, enabling the concentration of measurement resources in high-value, low-confidence areas. This closed-loop mechanism, while ensuring imaging resolution and positioning accuracy, improves sensitivity to deep or weakly contrasted targets, reduces unnecessary repeated measurements and field work time, and enhances the system's practicality and economy under complex geological conditions and constrained operating conditions.

[0151] like Figure 2 As shown, in one embodiment, this application also provides a high-density DC resistivity array-type detection data fusion method, applied to the high-density DC resistivity array-type detection device described in Embodiment 1. The method includes the following steps:

[0152] S1, Multi-source potential acquisition, is used to acquire raw potential signals from multiple injection / measurement sequences and different injection modes of the electrode array to obtain multiple sets of raw potential field data;

[0153] S2, Spatiotemporal alignment, is used to perform position and time registration on the multiple sets of original potential field data based on the electrode position and global time reference to obtain aligned potential data;

[0154] S3, Contact Impedance Compensation and Multi-Scale Purification, is used to perform contact impedance estimation and local baseline correction on the alignment potential data, and decouple noise and coupling terms according to spatial scale to obtain the purification potential field.

[0155] S4. Local inversion uncertainty recording is used to invert the purification potential field at the partition or layer scale and record the uncertainty distribution of each inversion result to obtain several local conductivity maps with uncertainty.

[0156] S5. Uncertainty-aware data fusion is used to perform consistency correction and fusion at the spatial level based on the local conductivity map with uncertainty, and output the joint conductivity distribution and the corresponding uncertainty map.

[0157] S6. Imaging and scheduling output, used to generate positioning imaging and structured anomaly parameters based on the joint conductivity distribution and uncertainty map, and output a priority list for subsequent adaptive measurement scheduling.

[0158] Furthermore, in step S5, the uncertainty-aware data fusion adopts a multi-scale weighted fusion strategy from coarse to fine. The weights are jointly determined by the inversion uncertainty of each local conductivity map, the confidence level of the contact impedance estimation of the corresponding measurement, and the confidence level of the electrode position / attitude.

[0159] Its beneficial effects are the same as those of the embodiment of a high-density DC current array detection device, and will not be repeated here.

[0160] The foregoing has only described certain exemplary embodiments of the present invention by way of illustration. Undoubtedly, those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the foregoing drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A high-density DC current array detection device, characterized in that, include: The electrode array module is used to arrange multiple rows of injection / measurement electrodes in a high-density pattern and inject controllable DC excitation to obtain the original potential field. The data acquisition module is used to align the original potential field in time and space to obtain aligned potential data; The signal preprocessing module is used to perform contact impedance compensation and multi-scale noise suppression on the alignment potential data to obtain a purified potential field. The spatial inversion module is used to perform spatial inversion on the purification potential field based on physical constraints to obtain a candidate map of conductivity distribution. The feature extraction module is used to extract the geometric and amplitude parameters of the anomaly from the candidate conductivity distribution map to obtain the anomaly parameters; The imaging and alarm module is used to generate localization images based on the abnormal parameters and output alarm and visualization results.

2. The high-density DC current array detection device according to claim 1, characterized in that, The electrode array module employs staggered injection paths and programmable segment spacing to generate depth-to-shallow differential sensitivity in adjacent measurements, thereby improving the deep response signal-to-noise ratio and preserving previous injection path information in subsequent inversions to assist in interlayer differentiation.

3. The high-density DC current array detection device according to claim 1, characterized in that, The data acquisition module synchronizes electrode position tracking with a global time reference to eliminate electrode positioning errors and clock drift, thereby obtaining high-precision time-space aligned potential data. This aligned potential data is then used as input for subsequent purification and inversion.

4. The high-density DC current array detection device according to claim 1, characterized in that, The signal preprocessing module employs synchronous demodulation and local baseline adaptive estimation based on a reference electrode grid to suppress contact impedance and low-frequency environmental drift while preserving spatial details, thereby obtaining a stable purification potential field.

5. The high-density DC current array detection device according to claim 1, characterized in that, The spatial inversion module is based on a hierarchical-local coupled inversion process with prior physical boundary constraints. It is used to first lock the global conductivity trend with low-resolution tomography, and then refine it layer by layer in the high uncertainty region using local correction kernels to obtain the candidate conductivity distribution map that has both global stability and local resolution; and to record the uncertainty distribution for adaptive measurement scheduling.

6. The high-density DC current array detection device according to claim 1, characterized in that, The feature extraction module uses morphological consistency filtering and spatial coherence clustering to separate geometric connectivity information from candidate images and calculate the volume, peak contrast and connectivity index of anomalies to obtain structured anomaly parameters, thereby significantly reducing the false alarm rate based on isolated noise points.

7. The high-density DC current array detection device according to claim 1, characterized in that, The device also includes an adaptive measurement scheduling module, which is used to dynamically adjust the subsequent injection / measurement sequence based on the uncertainty distribution of the inversion module and the abnormal parameters obtained by the feature extraction, in order to maximize the detection sensitivity and positioning accuracy of the target area under a limited number of measurements, and obtain detection results with significantly improved measurement efficiency.

8. A method for fusing high-density DC current array-based detection data, characterized in that, The method applied to the high-density DC current array detection device according to any one of claims 1-7 includes the following steps: S1. Acquire raw potential signals from multiple injection / measurement sequences and different injection modes of the electrode array to obtain multiple sets of raw potential field data; S2. Based on the electrode position and global time reference, perform position and time registration on the multiple sets of original potential field data to obtain aligned potential data; S3. Perform contact impedance estimation and local baseline correction on the alignment potential data, and decompose the decoupled noise and coupling terms according to the spatial scale to obtain the purification potential field; S4. Perform inversion on the purification potential field at the partition or layer scale and record the uncertainty distribution of each inversion result to obtain several local conductivity maps with uncertainty. S5. Based on the local conductivity map with uncertainty, perform consistency correction and fusion at the spatial level to output the joint conductivity distribution and corresponding uncertainty map; S6. Based on the joint conductivity distribution and uncertainty map, generate positioning imaging and structured anomaly parameters, and output a priority list for subsequent adaptive measurement scheduling.

9. The high-density DC-DC array-based detection data fusion method according to claim 8, characterized in that, In step S5, a multi-scale weighted fusion strategy from coarse to fine is adopted. The weights in the weighted fusion strategy are jointly determined by the inversion uncertainty of each local conductivity map, the confidence level of the contact impedance estimation of the corresponding measurement, and the confidence level of the electrode position / attitude.

Images