Ground penetrating radar data gain method based on EAGA
By adaptively adjusting the ground-penetrating radar signal gain using the EAGA algorithm, the problems of nonlinear signal distortion and gain oversaturation in complex media are solved, enabling efficient detection of targets such as termite nests and improving the data quality of deep learning models.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-25
- Publication Date
- 2026-04-14
AI Technical Summary
Existing ground-penetrating radar technology struggles to address issues such as nonlinear signal distortion and gain oversaturation in complex media, making it difficult to detect targets like termite nests. This is especially true in dam engineering where the concealed nesting characteristics make early detection challenging, and conventional gain methods cannot effectively improve the signal-to-noise ratio.
Employing an EAGA-based segmented gain control strategy, this algorithm adaptively adjusts the gain intensity of each signal segment through information entropy maximization and a dynamic step-size search mechanism, forming an adaptive and entropy-driven gain algorithm suitable for various ground-penetrating radar data types.
It significantly improves the information content and signal-to-noise ratio of ground-penetrating radar data, enhances the signal characteristics of targets such as termite nests, avoids oversaturation, and improves detection accuracy and the recognition effect of deep learning models.
Smart Images
Figure CN121856900A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of ground penetrating radar, and more particularly to a ground penetrating radar data gain method based on EAGA (Enhanced Adaptive Greedy Algorithm). Background Technology
[0002] Termites are a core biological threat to global dam engineering. The presence of termite tunnels causes seepage "short circuits" inside the dam body, raising the phreatic line, increasing pore water pressure, reducing effective stress, and significantly lowering shear strength. Ultimately, this leads to leaks and collapses under water level fluctuations (Li Y, Dong ZY, Pan DZ, Pan C H. Effects of subterranean termite nest architectures on earth embankment seepage and stability[J]. Paddy and Water Environment, 2020, 18: 109-118.). Their concealed nesting characteristics make early detection difficult, resulting in a large number of dam failures being directly related to termite diseases (Wang Li et al. Termite Hazard Assessment Model Based on Risk Management[J]. Water Resources and Hydropower Engineering, 2021, 52(9): 89-95.). For example, in the 2016 breach of the Beijiang River dike in Guangdong, over 60% of the damaged area had seepage channels formed by termite fungal garden erosion (Chen Qiang et al. Three-dimensional reconstruction and stability analysis of termite nests in dikes [J]. Rock and Soil Mechanics, 2018, 39(3): 877-885.). The honeycomb-like mud skeleton structure of termite nests significantly reduces the shear strength of the dike. Experiments show that when the volume of the fungal garden exceeds 15%, the critical instability height of the dike decreases by 42% (Liu Fang et al. Influence of mesoscopic structure of termite nests on soil mechanical properties [J]. Journal of Civil Engineering, 2009, 42(11): 145-151.).
[0003] Ground-penetrating radar (GPR) gain algorithms are the core technology for dynamic signal compensation, and their performance directly affects the accuracy of target feature extraction. In traditional gain methods, linear gain compensates for signal attenuation by fixing the slope (Zhang Hua et al. Research on GPR signal gain optimization algorithm [J]. Chinese Journal of Geophysics, 2018, 61(3): 1120-1130.), but cannot solve the problem of nonlinear signal distortion in complex media; exponential gain can improve the signal-to-noise ratio of deep targets (Li Wei et al. Adaptive gain model of GPR based on time-frequency analysis [J]. Geophysical and Geochemical Exploration Computation Technology, 2020, 42(5): 615-622.), but is prone to near-surface signal oversaturation due to initial parameter selection deviation. Deep learning gain networks rely on large-scale labeled data (Wang Y. et al. DCGAN-based gain enhancement for buried object detection [J]. IEEE Transactions on Geoscience and Remote Sensing, 2022, 60: 5200413.), which restricts their application in data-scarce scenarios. Summary of the Invention
[0004] To address the issues of nonlinear distortion and gain oversaturation in complex signals after amplification, this invention provides a ground-penetrating radar (GPR) data gain method based on EAGA. The core of this method, the EAGA gain algorithm, is to maximize information entropy as the optimization objective, employing a piecewise gain control strategy and a dynamic step-size search mechanism to achieve precise enhancement of the GPR signal. It is applicable to various GPR data types, and the amplified data can be used directly or for deep learning training and recognition. It is particularly suitable for detecting isolated targets such as termite nests on dams.
[0005] This method mainly includes the following steps: S1. Use ground-penetrating radar to measure abnormal areas that may contain termite infestations, and obtain the raw ground-penetrating radar signals from N sampling points. ; S2, regarding the original ground-penetrating radar signal Preprocessing is performed, and the optimal number of segments and optimal step size of the ground-penetrating radar signal are determined; S3. Based on the determined optimal number of segments and optimal step size, iteratively optimize the gain and information entropy values to obtain the optimal gain array; S4. Multiply the optimal gain array with the preprocessed ground-penetrating radar data from S2 to obtain the gain-processed ground-penetrating radar data, and then display it.
[0006] A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.
[0007] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.
[0008] A computer program product includes a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.
[0009] The beneficial effects of the technical solution provided by this invention are as follows: Based on the GA algorithm, the EAGA algorithm used in this invention forms an improved algorithm system through adaptive improvement and entropy maximization objective function optimization. This algorithm has been applied to termite nest detection operations on dams, highlighting the two innovative points of adaptive and entropy-based approaches. This invention can adaptively adjust the gain intensity of each signal segment according to the information entropy of the ground-penetrating radar data. Because the main material of the dam is relatively uniform, the information entropy of isolated body signals such as termite nests, rocks, and voids is much higher than that of noise. Therefore, the ground-penetrating radar data of termite nests on the dam after amplification by the method of this invention exhibits the characteristics of high information content and low noise. Compared with complex deep learning adaptive gain methods, the gain method proposed in this invention is based on mathematical classification, requires no prior environment setup, can be deployed on various platforms, and has a concise algorithm that can be implemented using various coding languages. Due to its concise code, it is convenient to combine with other machine learning, deep learning, and other ground-penetrating radar recognition models without affecting its computational efficiency. The gain method proposed in this invention is applicable to various ground-penetrating radar data types. The amplified data can not only be used directly but also used for deep learning training and recognition. Attached Figure Description
[0010] The present invention will be further described below with reference to the accompanying drawings and embodiments. In the accompanying drawings: Figure 1 This is a flowchart of a ground-penetrating radar data gain method based on EAGA in an embodiment of the present invention; Figure 2 This is a schematic diagram of a region suspected to contain termites in an embodiment of the present invention; Figure 3 This is a schematic diagram illustrating the use of ground-penetrating radar to collect data on suspicious areas in an embodiment of the present invention; Figure 4 This is a schematic diagram of the segment number-image entropy curve in an embodiment of the present invention; Figure 5 This is a schematic diagram of the iteration number-image entropy curve in an embodiment of the present invention; Figure 6This is a schematic diagram comparing the original data with the data obtained by the gain method in an embodiment of the present invention; Figure 7 This is a schematic diagram of digging out termite nests according to the waveform position in an embodiment of the present invention; Figure 8 This is a comparison chart of the effects of three gain methods in the embodiments of the present invention. Detailed Implementation
[0011] To provide a clearer understanding of the technical features, objectives, and effects of the present invention, specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.
[0012] Example 1 Please refer to Figure 1 , Figure 1 This is a flowchart of a ground-penetrating radar data gain method based on Adaptive Greedy Algorithm (EAGA) in an embodiment of the present invention, specifically including: S1. Use ground-penetrating radar to measure abnormal areas that may contain termite infestations, and obtain ground-penetrating radar signals from N sampling points. ; According to ground-penetrating radar signals , These represent the 1st, 2nd, 3rd, ..., Nth sampling points, respectively. Find the piecewise gain. Make:
[0013] in, Indicates the gain coefficient. x =1,2,3,...,N.
[0014] The constraints are:
[0015] Where H() is the image information entropy function. This indicates a piecewise dot product operation. G max This is a preset maximum gain limit to prevent signal overflow. For example, if x=[1,2,3] and y=[2,2,2], then the dot product is calculated as: x y=[2,4,6].
[0016] The key parameters for the performance of the EAGA algorithm are the number of segments M and the maximum number of iterations T. M determines the number of intervals into which the signal is divided, directly affecting the fineness of local optimization; T controls the depth of the search process, ensuring that the algorithm can approach the optimal solution within a finite number of steps. Once appropriate M and T are determined, the optimal gain array G can be calculated for a specific dam medium environment.
[0017] Step S1 is as follows: S1.1 Delineation of Abnormal Areas: Based on historical records of embankment damage, on-site signs of termite activity (such as mud lines, swarming holes), or preliminary survey results, accurately locate high-risk areas on the embankment where termite infestations exist (such as...). Figure 2 As shown in the diagram), the area is clearly marked on the dam plan as the key target area for ground-penetrating radar scanning.
[0018] S1.2 Equipment Configuration and Data Acquisition: ① Select a suitable ground-penetrating radar antenna (such as an antenna with a center frequency of 170MHz, which is suitable for deep-layer detection in dams).
[0019] ② Within the designated area, conduct a systematic scan according to pre-set survey lines (usually parallel to the dam axis or perpendicular to the suspected area). Ensure that the spacing between survey lines and the density of survey points are sufficient to cover the target area (i.e., the abnormal area) and meet the resolution requirements.
[0020] ③ Operate the ground-penetrating radar equipment to collect raw ground-penetrating radar echo data, such as Figure 3 As shown. During the data acquisition process, attention should be paid to the effects of factors such as antenna coupling and ground undulations, and records should be made as necessary for subsequent processing and correction.
[0021] ④ Collect the raw ground-penetrating radar signals (Usually in A-Scan or B-Scan format) After real-time transmission or storage, it is imported into the data processing host computer (such as a laptop or workstation).
[0022] S2, regarding the original ground-penetrating radar signal Preprocessing is performed, and the number of segments and dynamic step size for dividing the ground-penetrating radar signal are determined. Specifically: S2.1. Perform necessary preprocessing on the acquired raw data: First, perform direct wave (to eliminate strong surface reflection interference) clipping, background removal (to suppress system noise and background clutter), bandpass filtering (to retain effective frequency band information), etc., to provide a "cleaner" signal and a more effective gain range for subsequent gain processing.
[0023] S2.2 Determine the number of segments M Based on signal length Based on the expected processing accuracy, an initial range for the number of segments is set (e.g., M = 16, 32, 64, 128). The signal length... Determined by the number of sampling points, the expected processing accuracy includes medium accuracy for termite nest detection. The update logic for the one-dimensional gain array (gain_matrix) is to perform segmented local optimization, gradually adjusting to maximize image entropy by searching for a better value near the current best gain in each segment. The specific update process for the one-dimensional gain array is as follows: ① Set the initial number of segments M to divide the signal into equal-length intervals, and set the initial gain one-dimensional array A to be a one-dimensional matrix with 1 row, M columns, and all elements equal to 1 (rounded down):
[0024] Wherein, ΔL is the number of sampling points contained in each interval when the ground penetrating radar signal is divided into equal-length intervals according to the initial segmentation number M.
[0025] ② Determine the search range of the current segment: For the k-th segment, a search interval is generated with the current best gain (i.e., A(k)) of the segment as the center. The lower limit of the interval is the search step size (typically 0.1) and the upper limit is the initial step size (typically 0.5). The initial step size will decrease with iteration decay to ensure that the search range gradually decreases.
[0026] ③ Evaluate each candidate gain within the search range: Create a temporary gain one-dimensional array B to store the result of multiplying the gain of each round with the preprocessed ground-penetrating radar data, and then calculate the information entropy of the image as the evaluation index.
[0027] The formula for calculating information entropy H is: Suppose the image has LH different gray levels (for an 8-bit grayscale image, LH=256). Let H represent the probability of gray level i occurring. Then, the entropy H of the image is calculated using the following formula:
[0028]
[0029] in, is the number of times gray level i appears in the image, and N is the total number of pixels in the image.
[0030] The matrix with the largest information entropy value is used as the one-dimensional array of the best candidate gain.
[0031] ④ Update the one-dimensional gain array and the best gain record. After finding the optimal gain A(k) for the current segment, update the optimal gain for that segment. At this point, the k-th segment ends. If k is not the last segment, process the k+1 segment and subsequent segments according to the same process until the last segment ends.
[0032] This embodiment uses a measured signal from a specific levee location as an example to verify the effectiveness of the method. Based on the fact that the difference between the maximum and minimum sampling point values in the actual data is greater than 20 times, setting the initial step size and minimum step size to 0.5 and 0.1 respectively is sufficient. Since the number of iterations and the number of segments are independent, the number of iterations can be set to 4. Subsequently, the number of segments is checked sequentially, and the results are plotted as shown below. Figure 4The curve showing the relationship between M and H is shown. The gain of each segment is not reflected in this curve; the gain coefficient array needs to be determined according to EAGA, and its effect can be verified through linear gain and exponential gain. Figure 4 It can be seen that when the number of segments M=32, the information entropy of the amplified ground-penetrating radar image is better, and the number of segments is smaller, reducing the computational load. Since the number of segments is not directly related to the number of iterations, therefore... Figure 4 The judgment is sufficient; only one judgment is needed for each dike project.
[0033] S2.3. Determine the step size using an exponential decay step size mechanism:
[0034] In the formula, Indicates the current step size. The initial step size, Let t be the minimum step size and t be the current iteration number. This formula is used to update the current step size. Determining the step size is to find the optimal gain, as ground-penetrating radar data can only be compared for image entropy after being amplified. The step size corresponding to the optimal image entropy H is the optimal step size.
[0035] Verifying the optimal number of iterations: The initial step size and minimum step size are still set to 0.5 and 0.1 respectively. Given that for the ground-penetrating radar data of this location, a segment count of 32 results in optimal gain, we only need to repeatedly check the number of iterations to select the optimal number of iterations for the best gain. The iteration number-image entropy curve is shown below. Figure 5 As shown, by Figure 5 It can be seen that the image entropy converges when t=19, that is, the image entropy is optimal when the maximum number of iterations T=19, and the corresponding step size is the optimal step size.
[0036] This step determines the optimal number of segments and the optimal step size.
[0037] S3. Based on the determined optimal number of segments and optimal step size, iteratively optimize the gain and information entropy values to obtain the optimal gain array; Specifically: S3.1 For each segment k (k=1 to M): ① At the current gain g k (t-1) Based on this, try increasing and decreasing the current step size respectively. (i.e., candidate gain g) k (t-1) + and g k (t-1) - ), while ensuring that the candidate gain is within the constraint range [0.01, G max ]Inside.
[0038] ② Calculate the gain coefficient of the entire signal s when only the gain coefficient of the k-th segment is changed, while the gain coefficients of other segments remain unchanged. The information entropy H of G.
[0039] ③ Compare the current gain g k (t-1) And the information entropy values corresponding to the two candidate gains.
[0040] ④ Select the gain value with the largest information entropy value as the new gain g for this segment in the current iteration. k (t) :
[0041] in The partial gain coefficients (measured data of a dike in a certain area, which is a one-dimensional gain array that can be multiplied with the sampling points) are shown in Table 1.
[0042] Table 1 Partial Gain Coefficients
[0043] S3.2, Update Step Size: Calculate the next step size based on the exponential decay mechanism.
[0044] S3.3, Check the termination condition: Calculate the information entropy H of the entire signal after the current iteration. (t) .
[0045] ① If t>=T (reaching the maximum number of iterations) or |H (t) -H (t-1) If |<ε (the change in entropy is less than the threshold ε), then the iteration terminates.
[0046] ②Otherwise, let t = t + 1 and proceed to the next iteration.
[0047] S3.4 Output the optimal gain: When the iteration terminates, select the gain array G obtained from the last iteration. (t) As the optimal gain array G.
[0048] S4. Multiply the optimal gain array with the preprocessed ground-penetrating radar data from S2 to obtain the gain-processed ground-penetrating radar data (usually a B-Scan image), and then display it.
[0049] In raw ground-penetrating radar data, target signals are weak and difficult to identify. This method significantly and adaptively enhances the signal characteristics of the ant nest target area through EAGA gain, while maintaining background clarity and avoiding oversaturation, making the target outline and internal structure information easier to identify.
[0050] Engineers can use the enhanced, clear ground-penetrating radar images to precisely locate the spatial position (depth, horizontal range) of termite nests, guiding subsequent precise excavation and remediation work, such as... Figure 7 As shown.
[0051] Engineers can also use the high-quality ground-penetrating radar data after gain processing as training or input data for subsequent deep learning models (such as target detection and semantic segmentation networks), significantly improving the model's accuracy and robustness in identifying targets such as termite nests. This method outputs high-quality data, effectively solving the performance bottleneck problem of deep learning models caused by poor input data quality (low signal-to-noise ratio, oversaturation).
[0052] To better compare the advantages of the EAGA gain algorithm proposed in this invention, three gain methods were used to amplify the measured signal. Taking a dam as an example, the ground-based direct wave was removed; that is, the first row of data represents the dam slope, and the ground-penetrating radar center frequency was 170MHz. The gain effect is shown below. Figure 8 As shown in the image, the area within the black box is the main termite nest. Figure 8 It can be seen that while linear gain strengthens the bottom signal, it fails to effectively enhance the characteristic waveform of the ant nest; while exponential gain cannot adaptively adjust the gain coefficient according to the location of the ant nest, easily leading to oversaturation of the shallow or bottom layer signals. In contrast, the EAGA algorithm used in this invention can both adaptively identify the gain in complex backgrounds and accurately improve the gain amplitude in the effective signal range. (The text then repeats the description of the algorithm, which is redundant and can be omitted.) Figure 8 The image entropy of the five images was calculated, and the values were 2.0621, 5.0126, 4.7610, 5.1951 and 6.4963, respectively. It can be seen that the EAGA algorithm used in this invention can better enhance the ground penetrating radar, so that more hidden details in the ground penetrating radar image can be extracted.
[0053] Example 2 A computer device includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps of the above-described method.
[0054] Example 3 A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above method.
[0055] Example 4 A computer program product includes a computer program or instructions that, when executed by a processor, implement the steps of the above-described method.
[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A ground-penetrating radar data gain method based on EAGA, characterized in that, include: S1. Use ground-penetrating radar to measure abnormal areas that may contain termite infestations, and obtain the raw ground-penetrating radar signals from N sampling points. ; S2, regarding the original ground-penetrating radar signal Preprocessing is performed, and the optimal number of segments and optimal step size of the ground-penetrating radar signal are determined; S3. Based on the determined optimal number of segments and optimal step size, iteratively optimize the gain and information entropy values to obtain the optimal gain array; S4. Multiply the optimal gain array with the preprocessed ground-penetrating radar data from S2 to obtain the gain-processed ground-penetrating radar data, and then display it.
2. The EAGA-based ground-penetrating radar data gain method as described in claim 1, characterized in that, In S1, based on historical records of dam damage, signs of termite activity on site, or preliminary survey results, abnormal areas on the dam that may have termite risks are located.
3. The EAGA-based ground-penetrating radar data gain method as described in claim 1, characterized in that, In S2, the preprocessing includes direct wave clipping, background removal, and bandpass filtering. The direct wave clipping is used to eliminate strong surface reflection interference, the background removal is used to suppress system noise and background clutter, and the bandpass filtering is used to retain effective frequency band information.
4. The EAGA-based ground-penetrating radar data gain method as described in claim 1, characterized in that, In S2, the process of determining the optimal number of segments M is as follows: (1) Based on signal length Based on the expected processing accuracy, common experience in segmenting ground-penetrating radar data for dams, and equipment computing power, the number of segments was manually defined as 5-100, with an initial segmentation of 5, to provide an initial iteration starting point for subsequent entropy value verification. (2) Initialize the gain array The iteration counter is initialized to t=1; (3) Calculate the information entropy H of the ground-penetrating radar image after gain at different M values, and take the M value corresponding to the optimal information entropy H as the determined number of segments; the formula for calculating information entropy H is: Assuming an image has LH distinct gray levels, the formula for calculating the image's information entropy H is as follows: in, Represents grayscale level i The probability of occurrence, where LH represents the number of different gray levels in the image. It is grayscale. i The number of times something appears in an image, where N is the total number of pixels in the image.
5. The EAGA-based ground-penetrating radar data gain method as described in claim 4, characterized in that, In S2, an exponential decay step size mechanism is used to determine the step size: in, Indicates the current step size. The initial step size, Let t be the minimum step size and t be the current iteration number. This exponentially decaying step size mechanism causes the step size to decrease exponentially with the number of iterations. When the decay value is less than t, the step size will decrease. When the step size is stable, This ensures both the efficiency of adjustments in the early iterations and avoids computational redundancy caused by excessively small step sizes in the later stages. The optimal step size corresponds to the optimal information entropy H of an image.
6. The EAGA-based ground-penetrating radar data gain method as described in claim 1, characterized in that, In S3, the iterative optimization process is as follows: S3.1 For each segment, select the gain value with the largest information entropy value as the new gain for that segment in the current iteration; S3.2 Calculate the next step size based on the exponential decay mechanism; S3.3 Calculate the information entropy H of the entire signal after the current iteration. (t) Set the iteration termination condition: If t>=T or |H (t) -H (t-1) If |<ε, then the iteration terminates. T represents the maximum number of iterations, t represents the current iteration number, and H... (t) H represents the information entropy value obtained in the t-th iteration. (t-1) Let represent the information entropy value obtained in the (t-1)th iteration, and ε represent the threshold. S3.4 When the iteration terminates, select the gain array G obtained from the last iteration. (t) As the optimal gain array G.
7. A computer device comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes a computer program to implement the steps of the EAGA-based ground-penetrating radar data gain method as described in any one of claims 1-6.
8. A computer-readable storage medium, characterized in that, The device contains a computer program that, when executed by a processor, implements the steps of the EAGA-based ground-penetrating radar data gain method as described in any one of claims 1-6.
9. A computer program product, characterized in that, Includes a computer program or instructions that, when executed by a processor, implement the steps of the EAGA-based ground-penetrating radar data gain method as described in any one of claims 1-6.