Ground penetrating radar image automatic identification method based on forward modeling and actual measurement mixing

By combining forward simulation and measured data in ground penetrating radar image recognition, and using the YOLOX model for automatic identification, the problem of traditional methods relying on manual interpretation and noise interference is solved, and more efficient recognition accuracy and efficiency are achieved.

CN119006997APending Publication Date: 2024-11-22EAST CHINA JIAOTONG UNIVERSITY +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410730784.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-06-06
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Traditional ground-penetrating radar image recognition methods rely on manual interpretation, are inefficient and susceptible to human factors, and the complex underground media leads to noise interference and signal attenuation, resulting in difficulty in identification.

Method used

Using a method based on a mixture of forward and actual measurement, combined with deep learning technology, a heterogeneous model is constructed through elliptical autocorrelation function, adding diseases and setting excitation sources, and using the YOLOX model to train to achieve automatic recognition of ground penetrating radar images.

Benefits of technology

It improves the accuracy and efficiency of ground-penetrating radar image recognition, enhances the generalization ability of the model, and is suitable for ground-penetrating radar image recognition tasks under different geological conditions.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119006997A_ABST
    Figure CN119006997A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of geophysical exploration, and particularly discloses a ground penetrating radar image automatic identification method based on forward modeling and actual measurement mixing. Comprising the following steps: step 1, acquiring image data; step 2, constructing a heterogeneous model; step 3, adding a disease into the model in the step 2; step 4, setting an excitation source, step 5, preprocessing a forward modeling image set, step 6, preprocessing an actually measured image, step 7, training a YOLOX model, and step 8, calculating a result. According to the method, forward modeling data and actual measurement data are combined, a mixed data set is constructed, and the generalization ability of the model is improved; the recognition model is trained by using the deep learning technology, automatic recognition of the ground penetrating radar image is realized, and the recognition precision and efficiency are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of geophysical detection technology, and in particular to an automatic recognition method for ground penetrating radar images based on a mixture of forward modeling and actual measurement. Background Art

[0002] With the continuous development of science and technology, ground penetrating radar, as a highly efficient and non-destructive underground detection technology, has been widely used in geological exploration, archaeological excavation, tunnel detection and other fields.

[0003] Traditional ground-penetrating radar image recognition methods are usually based on signal processing and image analysis techniques, such as time domain analysis, frequency domain analysis, filtering, edge detection, threshold segmentation, etc. The principles of these methods are intuitive and easy for professionals to understand and operate. Technicians can flexibly select and adjust parameters based on their understanding of radar data characteristics to adapt to different detection environments and target characteristics. In addition, traditional ground-penetrating radar image recognition methods have relatively low data quality requirements and do not require a large amount of labeled data. Although this method is technically mature, intuitive and easy to use, and computationally efficient, it still has many shortcomings.

[0004] Traditional ground-penetrating radar image recognition methods often rely on manual interpretation, which is not only inefficient but also easily affected by human factors, resulting in inaccurate recognition results. In addition, due to the complexity and variability of underground media, radar images often have problems such as noise interference and signal attenuation, which makes image recognition difficult.

[0005] Therefore, it is particularly urgent to develop a more accurate method for automatic recognition of ground penetrating radar images based on a mixture of forward modeling and measured data. This method combines forward modeling data and measured data to construct a hybrid data set, uses deep learning technology to train the recognition model, and realizes automatic recognition of ground penetrating radar images, thereby improving recognition accuracy and efficiency. Summary of the invention

[0006] In order to overcome the above-mentioned defects of the prior art, an embodiment of the present invention provides a method for automatic recognition of ground penetrating radar images based on a mixture of forward modeling and actual measurement to solve the problems raised in the above-mentioned background technology.

[0007] To achieve the above-mentioned purpose, the present invention provides the following technical solution: a method for automatic recognition of ground penetrating radar images based on a mixture of forward modeling and actual measurement, comprising the following steps: step 1: image data acquisition, step 2: constructing a heterogeneous model, step 3: adding defects to the model of step 2, step 4: setting an excitation source, step 5: preprocessing the forward modeling image set, step 6: preprocessing the actual measured image, step 7: YOLOX model training, and step 8: calculating the results.

[0008] Step 1: Image data acquisition: forward modeling is used to obtain image data and ground penetrating radar image data of roads is obtained by ground penetrating radar;

[0009] Step 2: Construct a heterogeneous model: Use the elliptic autocorrelation function to construct a heterogeneous model;

[0010] Step 3: Add diseases to the model in step 2: Randomly add diseases to the model in step 2 and change the surrounding conditions;

[0011] Step 4: Set the excitation source: During the forward simulation, set the excitation source.

[0012] Step 5: Preprocessing the forward modeling image set: performing data preprocessing on the forward modeling image set acquired in step 1;

[0013] Step 6: Preprocessing the measured images: performing data preprocessing on the measured image set acquired in step 1;

[0014] Step 7: YOLOX model training: Mix the two data and train them using the YOLOX model;

[0015] Step 8: Calculation results: Based on the evaluation index calculation results, evaluate the effect of the forward modeling dataset as a supplementary dataset for the measured data.

[0016] Preferably, the specific steps of constructing the heterogeneous model using the elliptic autocorrelation function in step 2 are as follows:

[0017] Step 2.1: Define the elliptical autocorrelation function: The elliptical autocorrelation function is used to describe the spatial correlation structure of geological variables. It is based on the shape of the ellipse and is defined by the length and direction angle of the major and minor axes.

[0018] Step 2.2: Determine the model parameters: including the main axis, the secondary axis length, the direction angle, the correlation range, and the shape parameters;

[0019] Step 2.3: Generate random field: Use the elliptical autocorrelation function and model parameters to generate a random field that satisfies the autocorrelation function through a random process, thereby obtaining a simulated heterogeneous underground medium model;

[0020] Step 2.4: Verify and adjust the model: The generated random field model is compared with the measured geological data for verification. If the difference between the two is significant, the parameters of the elliptical autocorrelation function should be adjusted, and then the random field should be reconstructed;

[0021] Step 2.5: Apply the model: In the forward simulation of the ground penetrating radar, the heterogeneous model generated based on the elliptical autocorrelation function is used as the input of the physical property distribution of the underground medium to calculate the propagation response of the electromagnetic wave in the medium, thereby generating a more realistic GPR image.

[0022] Preferably, the added defects refer to faults and folds; and the changed surrounding conditions include filling the interior of the defects with different media including soil, rock or artificial materials.

[0023] Preferably, the step 5 preprocesses the forward image set including data augmentation, filtering, and labeling.

[0024] Preferably, the step 6 pre-processes the measured image including filtering and labeling.

[0025] Preferably, the specific steps of training using the YOLOX model are as follows:

[0026] Step 7.1: Data preparation: Use the LabelImg annotation tool to annotate the dataset. In addition, randomly divide the radar images of various road damage into a training dataset and a test set used after training at a ratio of 9:1;

[0027] Step 7.2: Configuration file: Create a YOLOX configuration file to define the model structure and training parameters;

[0028] Step 7.3: Data preprocessing: Scale all images to 416×416 pixels, either long-side scaling or short-side scaling, and perform zero padding.

[0029] Step 7.4: Data enhancement;

[0030] Step 7.5: Model training: Use the training script to perform model training;

[0031] Step 7.6: Model prediction: Use the trained model to predict new images; YOLOX generates all bounding boxes and corresponding confidence scores at once, and then selects the final detection result through non-maximum suppression;

[0032] Step 7.7: Result visualization and analysis: Visualizing the detection results of step 7.6 can visually check the model's ability to identify the target and further analyze the performance of the model;

[0033] Step 7.8: Parameter optimization.

[0034] Preferably, the evaluation indicators include accuracy and recall.

[0035] Preferably, the calculation formula of the accuracy is as follows:

[0036] A=(TP+TN) / (TP+FN+FP+TN), where A represents the accuracy rate, TP represents a true positive example, the model predicts a positive result and the actual result is also positive; FN represents a false negative example, the model predicts a negative result but the actual result is positive; FP represents a false positive example, the model predicts a positive result but the actual result is negative; TN represents a true negative example, the model predicts a negative result and the actual result is also negative.

[0037] Preferably, the calculation formula of the recall rate is as follows:

[0038] R = TP / (TP+FN), where R represents the recall rate, TP represents a true positive example, which is a positive example predicted by the model and is actually positive; FN represents a false negative example, which is a negative example predicted by the model but is actually positive.

[0039] Technical effects and advantages of the present invention:

[0040] 1. The present invention combines forward simulation data and measured data to construct a mixed data set, thereby improving the generalization ability of the model;

[0041] 2. The present invention uses deep learning technology to train the recognition model, realizes the automatic recognition of ground penetrating radar images, and improves the recognition accuracy and efficiency;

[0042] 3. The present invention is applicable to ground penetrating radar image recognition tasks under different geological conditions and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] The present invention is further described using the accompanying drawings, but the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative work.

[0044] Figure 1 A schematic diagram of the system structure of an embodiment of the present invention;

[0045] Figure 2 is a flow chart of a method according to an embodiment of the present invention;

[0046] Figure 3 This is a model training flow chart of an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] See also Figure 1As shown, the detection system based on the automatic recognition method of ground penetrating radar images mixed with forward modeling and actual measurement includes a data acquisition module, a preprocessing module, a YOLOX model training module, and a result evaluation module.

[0049] The output end of the data acquisition module is connected to the input end of the preprocessing module by telecommunication, the output end of the preprocessing module is connected to the input end of the YOLOX model training module by telecommunication, and the output end of the YOLOX model training module is connected to the input end of the result evaluation module by telecommunication.

[0050] Data acquisition module: used to collect image data of the road section to be detected and transmit the collected image data to the preprocessing module;

[0051] In this embodiment, it should be specifically explained that the acquisition of image data includes acquiring image data by forward simulation and acquiring road ground penetrating radar image data by ground penetrating radar;

[0052] Preprocessing module: used to preprocess the forward modeling image set and the measured image data set obtained by the data acquisition module, and transmit the preprocessed image data to the YOLOX model training module;

[0053] In this embodiment, it should be specifically explained that the preprocessing in the preprocessing module includes data augmentation, filtering, and labeling of the forward image set, and filtering and labeling of the measured image data set;

[0054] YOLOX model training module: used to import the forward image dataset, the measured image dataset, and the mixed dataset obtained by the preprocessing module into the YOLOX target detection model for training respectively;

[0055] Result evaluation module: evaluates the effect of the forward modeling dataset as a supplementary dataset for measured data based on the evaluation index calculation results;

[0056] See also Figure 2 As shown, the automatic recognition method of ground penetrating radar images based on the mixture of forward modeling and actual measurement includes the following steps: step 1: image data acquisition, step 2: building a heterogeneous model, step 3: adding diseases to the model of step 2, step 4: setting an excitation source, step 5: preprocessing the forward modeling image set, step 6: preprocessing the actual measured image, step 7: YOLOX model training, and step 8: calculating the results.

[0057] Step 1: Image data acquisition: forward modeling is used to obtain image data and ground penetrating radar image data of roads is obtained by ground penetrating radar;

[0058] Step 2: Construct a heterogeneous model: Use the elliptic autocorrelation function to construct a heterogeneous model;

[0059] In this embodiment, it should be specifically explained that the specific steps of constructing the heterogeneous model with the elliptic autocorrelation function in step 2 are as follows:

[0060] Step 2.1: Define the elliptical autocorrelation function: The elliptical autocorrelation function is used to describe the spatial correlation structure of geological variables. It is based on the shape of the ellipse and is defined by the length and direction angle of the major and minor axes.

[0061] In this embodiment, it should be specifically explained that the main axis is consistent with the main extension direction of the geological body; the secondary axis is perpendicular to the main axis; the direction angle represents the direction of the main axis;

[0062] Step 2.2: Determine the model parameters: including the main axis, the secondary axis length, the direction angle, the correlation range, and the shape parameters;

[0063] In this embodiment, it should be specifically explained that the shape parameter is used to control the shape of the autocorrelation function curve;

[0064] Step 2.3: Generate random field: Use the elliptical autocorrelation function and model parameters to generate a random field that satisfies the autocorrelation function through a random process, thereby obtaining a simulated heterogeneous underground medium model;

[0065] In this embodiment, it should be specifically noted that the random process is a Markov chain Monte Carlo method or FFT transformation;

[0066] Step 2.4: Verify and adjust the model: The generated random field model is compared with the measured geological data for verification. If the difference between the two is significant, the parameters of the elliptical autocorrelation function should be adjusted, and then the random field should be reconstructed;

[0067] In this embodiment, it should be specifically explained that the comparison method is visual comparison and statistical analysis;

[0068] The visual comparison is to visually compare the generated model with the actual geological data to observe whether their spatial structures and distributions are similar;

[0069] The statistical analysis is to perform statistical analysis on the generated model, calculate its statistical parameters, and compare them with the statistical parameters of actual data;

[0070] It should be specifically noted that the statistical parameters include mean, variance, and covariance;

[0071] Step 2.5: Apply the model: In the forward simulation of the ground penetrating radar, the heterogeneous model generated based on the elliptical autocorrelation function is used as the input of the physical property distribution of the underground medium to calculate the propagation response of the electromagnetic wave in the medium, thereby generating a more realistic GPR image;

[0072] Step 3: Add diseases to the model in step 2: Randomly add diseases to the model in step 2 and change the surrounding conditions;

[0073] In this embodiment, it should be specifically explained that the added damage refers to faults and folds; the changed surrounding conditions include filling the damage with different media including soil, rock or artificial materials;

[0074] Step 4: Set the excitation source: During the forward simulation, set the excitation source.

[0075] In this embodiment, it is necessary to specifically explain that the setting of the excitation source is to simulate the interference of the actual environment interference source;

[0076] Step 5: Preprocessing the forward modeling image set: performing data preprocessing on the forward modeling image set acquired in step 1;

[0077] In this embodiment, it should be specifically explained that the step 5 performs preprocessing on the forward image set, including data augmentation, filtering, and labeling, to obtain the effect of quality and diversity of image data. That is, the limitation of preprocessing in this embodiment is only to achieve the above effect, so no specific limitation is made on the specific actual process;

[0078] The data augmentation is a technique for increasing the diversity of a data set by transforming or perturbing the original data to generate new data samples. The methods for data augmentation include geometric transformation, noise addition, and data synthesis.

[0079] The geometric transformation includes rotation, translation, and scaling;

[0080] The noise addition refers to adding random noise to the image to simulate the interference existing in actual observation;

[0081] The data synthesis is to combine images or data fragments together to generate a new complex scene;

[0082] Step 6: Preprocessing the measured images: performing data preprocessing on the measured image set acquired in step 1;

[0083] In this embodiment, it is necessary to specifically explain that the step 6 performs preprocessing of the measured image, including filtering and labeling, in order to obtain the effect of improving the image data quality. That is, the limitation of the preprocessing in this embodiment is only to achieve the above effect, so no specific limitation is made on the specific actual process.

[0084] The filtering is a commonly used technology in image processing, which is used to improve image quality, highlight target features or eliminate interference; specific processing methods include Gaussian filtering, median filtering, and bilateral filtering;

[0085] It should be specifically noted that the Gaussian filter is a smoothing filter implemented by convolution of the Gaussian kernel function, which effectively reduces high-frequency noise;

[0086] The median filter is effective for salt and pepper noise and impulse noise, and can retain edge information;

[0087] The bilateral filtering can simultaneously consider the spatial proximity of pixels and the similarity of pixel values, and can perform smoothing while maintaining edges;

[0088] Step 7: YOLOX model training: Mix the two data and train them using the YOLOX model;

[0089] See also Figure 3 As shown, in this embodiment, it is necessary to specifically explain that the specific steps of training using the YOLOX model are as follows:

[0090] Step 7.1: Data preparation: Use the LabelImg annotation tool to annotate the dataset. In addition, randomly divide the radar images of various road damage into a training dataset and a test set used after training at a ratio of 9:1;

[0091] Step 7.2: Configuration file: Create a YOLOX configuration file to define the model structure and training parameters;

[0092] It should be noted that the configuration file contains the model's hyperparameters, dataset path, training batch size, and learning rate information;

[0093] Step 7.3: Data preprocessing: Scale all images to 416×416 pixels, either long-side scaling or short-side scaling, and perform zero padding.

[0094] Step 7.4: Data enhancement;

[0095] It should be specifically noted that the data enhancement technology includes random rotation and random cropping to improve the robustness of the model;

[0096] Step 7.5: Model training: Use the training script to perform model training;

[0097] It should be specifically explained that the specific method of using the training script to execute the model training is: the training script loads the configuration file and the data set, and then executes the training cycle of the model; YOLOX adopts an iterative training strategy, first performs warm-up training, and then gradually increases the number of training times; during the training process, the model will gradually learn to detect the bounding box and category of the target, and adjust the network weights to improve performance; in order to reduce the training time, the pre-trained model uses large-scale data to obtain a model with wide applicability through pre-training; the official provides pre-trained models of multiple scales, but because there are no ground-penetrating radar images in the official pre-training data set, the role of the target recognition task of the ground-penetrating radar image is relatively small; at the same time, because there are many types of models used in the experiment, in order to ensure the consistency of experimental observations, the official pre-trained models are not used;

[0098] It should be noted that the experimental environment configuration table data used in this experiment is as follows: the system is Windows, the CPU model is AMD Ryzen 75800H, the memory is 16GB, the GPU model is VIDIA GeForce RTX 3060, the GPU video memory is 12.9GB, the GPU computing power value is 8.6, the programming language is Python3.7, the deep learning framework is Pytorch 2.0.1, and the dependent libraries mainly include Numpy 1.22.4, Tqdm4.65.0, Opencv 4.5.5.64, Pandas 2.0.1, and Pillow9.1.1;

[0099] Step 7.6: Model prediction: Use the trained model to predict new images; YOLOX generates all bounding boxes and corresponding confidence scores at once, and then selects the final detection result through non-maximum suppression;

[0100] In this embodiment, it is necessary to specifically explain that the working principle of non-maximum suppression is: first, all bounding boxes are sorted according to the confidence scores; second, the bounding box with the highest confidence is selected as a reference and other bounding boxes whose overlap exceeds a certain threshold are suppressed; finally, this process is repeated until all bounding boxes are processed;

[0101] It should be specified that the said suppression refers to deletion or reduction of weight;

[0102] Step 7.7: Result visualization and analysis: Visualizing the detection results of step 7.6 can visually check the model's ability to identify the target and further analyze the performance of the model;

[0103] In this embodiment, it should be specifically noted that the performance of the model includes precision, recall, and F1 score;

[0104] Step 7.8: Parameter optimization;

[0105] Step 8: Calculation results: Based on the calculation results of the evaluation indicators, evaluate the effect of the forward modeling dataset as a supplementary dataset for the measured data;

[0106] In this embodiment, it should be specifically noted that the evaluation indicators include precision and recall;

[0107] It should be specifically noted that the calculation formula of the accuracy is as follows:

[0108] A=(TP+TN) / (TP+FN+FP+TN), where A represents the accuracy rate, TP represents a true positive example, the model predicts it is positive and the actual result is also positive; FN represents a false negative example, the model predicts it is negative but the actual result is positive; FP represents a false positive example, the model predicts it is positive but the actual result is negative; TN represents a true negative example, the model predicts it is negative and the actual result is also negative;

[0109] It should be specifically noted that the calculation formula of the recall rate is as follows:

[0110] R = TP / (TP + FN), where R represents the recall rate, TP represents a true positive example, which is predicted by the model to be positive and is actually positive; FN represents a false negative example, which is predicted by the model to be negative but is actually positive;

[0111] Finally: 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 in the protection scope of the present invention.

[0112] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art who is familiar with the present technical field can easily think of changes or substitutions within the technical scope disclosed in the present application, which should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. An automatic recognition method of ground penetrating radar images based on mixed forward modeling and field measurement. It is characterized in that The following steps are involved: Step 1: Image data acquisition: forward modeling is used to obtain image data and ground penetrating radar image data of roads is obtained by ground penetrating radar; Step 2: Construct a heterogeneous model: Use the elliptic autocorrelation function to construct a heterogeneous model; Step 3: Add diseases to the model in step 2: Randomly add diseases to the model in step 2 and change the surrounding conditions; Step 4: Set the excitation source: During the forward simulation, set the excitation source. Step 5: Preprocessing the forward modeling image set: performing data preprocessing on the forward modeling image set acquired in step 1; Step 6: Preprocessing the measured images: performing data preprocessing on the measured image set acquired in step 1; Step 7: YOLOX model training: Mix the two data and train them using the YOLOX model; Step 8: Calculation results: Based on the calculation results of the evaluation indicators, evaluate the effect of the forward modeling dataset as a supplementary dataset for the measured data.

2. According to claim 1, a method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement, Features: The specific steps of constructing the heterogeneous model using the elliptical autocorrelation function in step 2 are as follows: Step 2.1: Define the elliptical autocorrelation function: Based on the shape of the ellipse, it is defined by the length and direction angle of the major and minor axes; Step 2.2: Determine the model parameters: including the main axis, the secondary axis length, the direction angle, the correlation range, and the shape parameters; Step 2.3: Generate random field: Use the elliptical autocorrelation function and model parameters to generate a random field that satisfies the autocorrelation function through a random process, thereby obtaining a simulated heterogeneous underground medium model; Step 2.4: Verify and adjust the model: The generated random field model is compared with the measured geological data for verification. If the difference between the two is significant, the parameters of the elliptical autocorrelation function should be adjusted, and then the random field should be reconstructed; Step 2.5: Apply the model: In the forward simulation of the ground penetrating radar, the heterogeneous model generated based on the elliptical autocorrelation function is used as the input of the physical property distribution of the underground medium to calculate the propagation response of the electromagnetic wave in the medium, thereby generating a more realistic GPR image.

3. According to claim 1, a method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement, Features: The added damage refers to faults and folds; the changed surrounding conditions include filling the interior of the damage with different media including soil, rock or artificial materials.

4. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The step 5 pre-processes the forward image set including data augmentation, filtering, and labeling.

5. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The step 6 pre-processes the measured image including filtering and labeling.

6. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The specific steps of training using the YOLOX model are as follows: Step 7.1: Data preparation: Use the LabelImg annotation tool to annotate the dataset. In addition, randomly divide the radar images of various road damage into a training dataset and a test set used after training at a ratio of 9:1; Step 7.2: Configuration file: Create a YOLOX configuration file; Step 7.3: Data preprocessing: Scale all images to 416×416 pixels, either long-side scaling or short-side scaling, and perform zero padding. Step 7.4: Data enhancement; Step 7.5: Model training: Use the training script to perform model training; Step 7.6: Model prediction: Use the trained model to predict new images; YOLOX generates all bounding boxes and corresponding confidence scores at once, and then selects the final detection result through non-maximum suppression; Step 7.7: Result visualization and analysis: Visualizing the detection results of step 7.6 can visually check the model's ability to identify the target and further analyze the performance of the model; Step 7.8: Parameter optimization.

7. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The evaluation indicators include precision and recall.

8. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The calculation formula of the accuracy is as follows: A=(TP+TN) / (TP+FN+FP+TN), where A represents the accuracy rate, TP represents a true positive example, the model predicts a positive result and the actual result is also positive; FN represents a false negative example, the model predicts a negative result but the actual result is positive; FP represents a false positive example, the model predicts a positive result but the actual result is negative; TN represents a true negative example, the model predicts a negative result and the actual result is also negative.

9. The method for automatic recognition of ground penetrating radar images based on mixed forward modeling and field measurement according to claim 1, Features: The calculation formula of the recall rate is as follows: R = TP / (TP+FN), where R represents the recall rate, TP represents a true positive example, which is a positive example predicted by the model and is actually positive; FN represents a false negative example, which is a negative example predicted by the model but is actually positive.