Deep training learning method for ground penetrating radar image automatic identification based on AI
Through the deep training and learning method based on AI, a single-item and comprehensive database of soil layers is established, and the automatic recognition and diagnosis of ground penetrating radar images is realized, solving the problems of low recognition accuracy and efficiency caused by relying on personal experience in the existing technology. Improved identification efficiency and accuracy.
Patent Information
- Application Number
- CN202411087231.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-09
- Publication Date
- 2024-12-06
AI Technical Summary
The existing ground penetrating radar image feature recognition depends on the engineer's personal experience, resulting in low accuracy and efficiency, and cannot meet the timeliness of actual engineering.
Using deep training and learning methods based on AI, a single-item sample database and AI database of soil layer are established, and trained and optimized through deep learning integration algorithms to form a deep learning integration algorithm for automatic recognition and diagnosis of ground penetrating radar images.
It improves the recognition efficiency of ground penetrating radar images and the accuracy of diagnostic results, and can quickly and accurately identify the location and characteristics of underground objects, meeting the timeliness during construction.
Smart Images

Figure CN119091191A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of engineering non-destructive geophysical detection, and in particular to a deep training learning method for automatic recognition of ground penetrating radar images based on AI. Background Art
[0002] Urban development is closely linked to the construction of municipal engineering projects, which are the infrastructure that supports the normal operation of cities and the daily lives of residents. Since the construction of municipal engineering projects will inevitably affect the surrounding areas, geological disasters such as road settlement and underground cavities caused by underground pipeline leakage are common. These problems may not only cause pipeline leakage and road collapse, but also pose a threat to people's lives and property safety. Therefore, timely detection and discovery of these underground pipelines, soil settlement, and soil cavities is extremely important for ensuring public safety.
[0003] In recent years, ground penetrating radar technology has been widely used in geological exploration of municipal engineering due to its non-destructive detection and high efficiency. Ground penetrating radar uses the propagation characteristics of electromagnetic waves in underground media to determine the location and characteristics of underground objects by analyzing the reflection, scattering or absorption of signals. At present, the recognition of ground penetrating radar image features mainly relies on the personal experience of engineers. This method has the following defects: 1) Because it relies too much on personal judgment, its accuracy and recognition speed cannot be guaranteed; 2) When faced with a large amount of data, relying on manual analysis and judgment is difficult to meet the timeliness requirements of actual projects.
[0004] In view of the deficiency of the prior art that the recognition of ground penetrating radar image features relies on personal experience, those skilled in the art have been looking for solutions. Summary of the invention
[0005] The purpose of the present invention is to provide a deep training learning method for automatic recognition of ground penetrating radar images based on AI to solve the problems of low accuracy and low efficiency in ground penetrating radar image recognition.
[0006] In order to solve the above technical problems, the present invention provides a deep training and learning method for automatic recognition of ground penetrating radar images based on AI, and the deep training and learning method for automatic recognition of ground penetrating radar images based on AI includes:
[0007] S1. Establish a single sample database of three types of soil layers; the three types of soil layers include: abnormal soil layer, soil layer stratification, soil layer voids and loose bodies;
[0008] S2. Based on the single sample database of the three soil layers and intelligent image analysis technology, an AI database of the three soil layers is established;
[0009] S3. Train and optimize the AI database of the three soil layers to form single algorithm learning results for the three soil layers;
[0010] S4, integrating the learning results of the individual algorithms of the three soil layers to form a deep learning integration algorithm;
[0011] S5. Establish an AI comprehensive geophysical exploration results database;
[0012] S6. Based on the deep learning integration algorithm, deep learning is performed on the AI comprehensive geophysical exploration results database, and training is performed on inspection recognition, feature recognition, and classification recognition respectively to form the output of AI comprehensive geophysical exploration results diagnosis results;
[0013] S7. Perform data compatibility and algorithm parameter adjustment on the output of AI comprehensive geophysical exploration results diagnosis to form an AI diagnostic model for radar image data. The AI diagnostic model for radar image data can recognize the input ground penetrating radar image to generate a diagnostic image of the ground penetrating radar image.
[0014] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, S1 includes:
[0015] S11, using a ground penetrating radar to obtain radar image data of three soil layers, and classifying and forming radar image data sets of the three soil layers; the radar image data sets of the three soil layers include: a radar image data set of soil layer anomaly, a radar image data set of soil layer stratification, and a radar image data set of soil layer voids and loose bodies;
[0016] S12. Preprocess the images and data of the radar image data sets of the three soil layers respectively, and establish a single sample database for each soil layer in a targeted manner.
[0017] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, in S11, the process of obtaining radar image data of three soil layers is as follows:
[0018] The ground penetrating radar is used to scan the reinforced concrete medium structure with an antenna center frequency of 800MHz, a sampling interval of 0.005m, and a time window of 20ns to obtain a ground penetrating radar image of the steel bar distribution;
[0019] The ground penetrating radar is used to scan the grouting effect behind the lining ring on the shield lining ring with the center frequency of the antenna being 600MHz, the sampling interval being 0.02m, and the time window being 30ns, and the ground penetrating radar image of the grouting soil layer distribution behind the lining ring is obtained;
[0020] The ground penetrating radar is used to scan the road surface on the urban road with the center frequency of the antenna being 400MHz, the sampling interval being 0.04m and the time window being 60ns to obtain the ground penetrating radar image of the loose soil layer and the voids under the road surface.
[0021] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, S12 includes:
[0022] Data preprocessing, including static correction removal, filtering and gain;
[0023] Image preprocessing, including image scaling, translation and cropping.
[0024] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, S2 includes:
[0025] S21, manually labeling soil layer information of the single sample databases of the three soil layers respectively, so as to establish a labeling library of the three soil layers; the labeling library of the three soil layers includes: a soil layer abnormal body labeling library, a soil layer stratification labeling library and a soil layer void and loose body labeling library;
[0026] S22. Import the annotated image libraries of the three soil layers into the AI database respectively, and establish the AI database of the three soil layers based on intelligent image analysis technology.
[0027] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, S3 includes:
[0028] S31, performing single-item reinforcement training and manual intervention reinforcement training on the AI database of each soil layer to correct the AI database of each soil layer;
[0029] S32, for each soil layer after correction, the AI database is used to adjust the data volume of the AI interpretation library, and the interpretation rate and accuracy are counted, and error data are collected to establish an error data collection library for each soil layer;
[0030] S33. Based on the data optimization algorithm of the collection library of each soil layer, the AI database of each soil layer is further improved and revised, and the interpretation rate and accuracy rate of the AI database of each soil layer are statistically analyzed until the statistical results of the interpretation rate and accuracy rate meet the predetermined standards. The AI database of each soil layer that meets the predetermined standards is regarded as the learning result of a single algorithm in the first stage. The cavity detection algorithm based on deep learning realizes the automatic identification and positioning of underground cavities; the anomaly detection algorithm based on deep learning realizes the automatic identification of the number and location of anomalies; the soil layer stratification detection algorithm based on deep learning realizes the automatic identification and stratification of soil layer structures.
[0031] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, in S32, the statistics of interpretation rate and accuracy based on adjusting the amount of AI interpretation library data include:
[0032] The data volume of the AI interpretation image library was adjusted to use 500 images, 1000 images, and 2000 images as nodes for interpretation rate and accuracy statistics.
[0033] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, in S32, the predetermined standard is: the interpretation rate is 95% and the accuracy rate is 90%.
[0034] Optionally, in the AI-based deep training and learning method for automatic recognition of ground penetrating radar images, S5 includes:
[0035] S51. Use ground penetrating radar to obtain comprehensive soil layer radar result image data;
[0036] S52, preprocessing the image and data of the comprehensive soil layer radar result image data to establish a comprehensive soil layer sample database;
[0037] S53, manually marking soil layer information on the comprehensive soil layer sample database to establish a comprehensive results marking library;
[0038] S54. Import the comprehensive results annotation library into the AI database, and establish an AI comprehensive geophysical exploration results database based on intelligent image analysis technology.
[0039] Optionally, in the AI-based deep training learning method for automatic recognition of ground penetrating radar images, S52 includes
[0040] Data preprocessing, including static correction removal, filtering and gain;
[0041] Image preprocessing, including image scaling, translation and cropping.
[0042] In the deep training and learning method for automatic recognition of ground penetrating radar images based on AI provided by the present invention, by establishing an AI database of three soil layers in the first stage of deep learning, the AI database of the three soil layers is trained and optimized to complete the interpretation requirements of each single feature image and the reasonable optimization of the data accuracy in the AI database, so as to effectively save the recognition time of the inspection and recognition, feature recognition and classification recognition of the ground penetrating radar image in the later second stage of deep learning AI comprehensive geophysical exploration results database, and effectively improve the recognition efficiency of the ground penetrating radar image and the accuracy of the diagnosis results. Based on the training of large-scale ground penetrating radar detection data, the AI big model can help the construction party and the construction party provide soil layer related information during the construction of underground projects, and provide an effective construction basis for subsequent construction, so as to improve the accuracy and efficiency of construction measures decision-making. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] The above and other objects, features and advantages of the present disclosure will become more apparent through a more detailed description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, wherein like reference numerals generally represent like components throughout the exemplary embodiments of the present disclosure.
[0044] Figure 1 It is a flow chart of a deep training and learning method for automatic recognition of ground penetrating radar images based on AI in one embodiment of the present invention;
[0045] Figure 2 It is a principle block diagram of a deep training learning method for automatic recognition of ground penetrating radar images based on AI in one embodiment of the present invention;
[0046] Figure 3 is a flow chart of single-item ground penetrating radar image recognition training in one embodiment of the present invention;
[0047] Figure 4 It is a flow chart of the depth training of ground penetrating radar image recognition in one embodiment of the present invention. DETAILED DESCRIPTION
[0048] The following is a further detailed description of the deep training learning method for automatic recognition of ground penetrating radar images based on AI proposed by the present invention in combination with the accompanying drawings and specific embodiments. The advantages and features of the present invention will become clearer according to the following description and claims. It should be noted that the drawings are all in a very simplified form and are not in precise proportions, and are only used to conveniently and clearly assist in explaining the purpose of the embodiments of the present invention.
[0049] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.
[0050] In the description of the invention, it is necessary to understand that the terms “center”, “longitudinal”, “lateral”, “length”, “width”, “thickness”, “up”, “down”, “front”, “back”, “left”, “right”, “vertical”, “horizontal”, “top”, “bottom”, “inside”, “outside”, “clockwise”, “counterclockwise”, etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the drawings, and are only for the convenience of describing the invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the invention.
[0051] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of the invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.
[0052] In the invention, unless otherwise clearly specified and limited, the terms "installed", "connected", "connected", "fixed" and the like should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two elements. For ordinary technicians in this field, the specific meanings of the above terms in the invention can be understood according to the specific circumstances.
[0053] Please refer to Figure 1 and Figure 2 The deep training learning method for automatic recognition of ground penetrating radar images based on AI includes one-stage deep learning and two-stage deep learning. The one-stage deep learning process includes S1-S3, and the two-stage deep learning process includes S4-S6. The specific method steps are as follows:
[0054] First, step S1 is executed to establish a single sample database of three types of soil layers; wherein the three types of soil layers include: abnormal soil body, stratified soil body, void soil body and loose soil body.
[0055] Specifically, the S1 includes:
[0056] S11, using a ground penetrating radar to obtain radar image data of three soil layers, and classifying and forming radar image data sets of the three soil layers; the radar image data sets of the three soil layers include: radar image data sets of soil abnormal bodies, radar image data sets of soil stratification, and radar image data sets of soil voids and loose bodies; wherein the process of obtaining the radar image data of the three soil layers is as follows:
[0057] The ground penetrating radar is used to scan the reinforced concrete medium structure with an antenna center frequency of 800MHz, a sampling interval of 0.005m, and a time window of 20ns to obtain a ground penetrating radar image of the steel bar distribution;
[0058] The ground penetrating radar is used to scan the grouting effect behind the lining ring on the shield lining ring with the center frequency of the antenna being 600MHz, the sampling interval being 0.02m, and the time window being 30ns, and the ground penetrating radar image of the grouting soil layer distribution behind the lining ring is obtained;
[0059] The ground penetrating radar is used to scan the road surface on the urban road with the center frequency of the antenna being 400MHz, the sampling interval being 0.04m and the time window being 60ns to obtain the ground penetrating radar image of the loose soil layer and the voids under the road surface.
[0060] S12. Preprocess the images and data of the radar image data sets of the three soil layers respectively, and establish a single sample database for each soil layer in a targeted manner.
[0061] Next, step S2 is executed to establish an AI database of the three soil layers based on the single sample database of the three soil layers and the intelligent image analysis technology.
[0062] Specifically, step S2 includes:
[0063] S21, manually labeling soil layer information of the single sample databases of the three soil layers respectively, so as to establish a labeling library of the three soil layers; the labeling library of the three soil layers includes: a soil layer abnormal body labeling library, a soil layer stratification labeling library and a soil layer void and loose body labeling library;
[0064] S22. Import the annotated image libraries of the three soil layers into the AI database respectively, and establish the AI database of the three soil layers based on intelligent image analysis technology.
[0065] Next, step S3 is executed to train and optimize the AI databases of the three soil layers to form single algorithm learning results for the three soil layers.
[0066] Specifically, step S3 includes:
[0067] S31. Perform single-item reinforcement training and manual intervention reinforcement training on the AI database of each soil layer to correct the AI database of each soil layer; specifically, the number of single-item learning training needs to reach 10 rounds.
[0068] S32. For each corrected AI database of soil layer, the interpretation rate and accuracy are counted based on the adjusted AI interpretation library data volume, and error data are collected to establish an error data set library for each soil layer; wherein, the predetermined standard is: the interpretation rate is 95%, and the accuracy is 90%; the interpretation rate and accuracy statistics based on the adjusted AI interpretation library data volume include: adjusting the AI interpretation library data volume to perform interpretation rate and accuracy statistics with 500 images, 1000 images, and 2000 images as nodes, respectively, and the nodes are accepted with an interpretation rate of 95% and an accuracy of 90% (i.e., meeting the predetermined standard). Based on S32, intelligent and efficient interpretation of different types of soil layer detection can be achieved;
[0069] S33. Based on the collection library data of each soil layer, the algorithm is optimized to further improve the revised AI database of each soil layer, and the interpretation rate and accuracy rate of the AI database of each soil layer after the improvement are statistically analyzed until the statistical results of the interpretation rate and accuracy rate meet the predetermined standards. The AI database of each soil layer that meets the predetermined standards is used as the single algorithm learning results of the first stage, and the model algorithm is deployed. The cavity detection algorithm based on deep learning can realize the automatic identification and positioning of underground cavities; the anomaly detection algorithm based on deep learning can realize the automatic identification of parameters such as the number and location of anomalies; the soil layer stratification detection algorithm based on deep learning can realize the automatic identification and stratification of soil layer structures. Reasonable optimization of the algorithm based on the collection library data can effectively save the time of deep learning radar image inspection and recognition, feature recognition and classification recognition in the later stage, and further effectively improve the accuracy of AI diagnosis comprehensive radar data images.
[0070] Preferably, a single learning training reaches 10 rounds, and can complete 2,000 images or more with an interpretation rate of 95% and an accuracy rate of 90% for a single completion, and the model algorithm is deployed to form a single algorithm learning result.
[0071] Next, step S4 is executed to integrate the learning results of the individual algorithms of the three soil layers to form a deep learning integration algorithm.
[0072] Next, execute step S5 to establish an AI comprehensive geophysical exploration results database.
[0073] Specifically, step S5 includes:
[0074] S51. Use ground penetrating radar to obtain comprehensive soil layer radar result image data;
[0075] S52. Preprocess the image and data of the comprehensive soil layer radar result image data to establish a comprehensive soil layer sample database; here, the preprocessing of the image and data of the comprehensive soil layer radar result image data includes data preprocessing and image preprocessing. The data preprocessing includes: static correction removal, filtering and gain, etc.; the image preprocessing includes image scaling, translation and cropping (cropping to the size required for AI recognition), etc.
[0076] S53, manually annotating soil layer information of the comprehensive soil layer sample database to establish a comprehensive result annotation library;
[0077] S54. Import the comprehensive results annotation library into the AI database, and establish an AI comprehensive geophysical exploration results database based on intelligent image analysis technology.
[0078] Next, step S6 is executed to perform deep learning on the AI comprehensive geophysical exploration results database based on the deep learning integration algorithm, and to perform training of inspection and recognition, feature recognition and classification recognition respectively, so as to form the output of AI comprehensive geophysical exploration results diagnosis results. The inspection and recognition, feature recognition and classification recognition here can inspect, identify and classify the abnormal bodies, soil layer stratification, soil layer cavities and loose bodies required by the radar image data.
[0079] Next, execute step S7, perform data compatibility and algorithm parameter adjustment on the AI comprehensive geophysical exploration results diagnosis output, and deploy the large model algorithm to form a radar image data AI diagnostic model, which can identify the input ground penetrating radar image to generate a diagnostic image of the ground penetrating radar image. The output of the results can summarize the identified image information, and then obtain the corresponding radar image diagnostic image based on the data information analysis. The above-mentioned data compatibility module can output the diagnostic identification image file as a general JPG or PNG, and the model training label is in TXT format.
[0080] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the described order of actions, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.
[0081] In summary, in the deep training and learning method for automatic recognition of ground penetrating radar images based on AI provided by the present invention, an AI database of three soil layers is established in the first deep learning stage, and based on the training and optimization of the AI database of the three soil layers, the interpretation requirements of each single feature image and the reasonable optimization of the data accuracy in the AI database are completed, so as to effectively save the recognition time of inspection and recognition, feature recognition and classification recognition of ground penetrating radar images in the later second-stage deep learning AI comprehensive geophysical exploration results database, thereby effectively improving the recognition efficiency of ground penetrating radar images and the accuracy of diagnostic results.
[0082] The above description is only a description of the preferred embodiments of the present invention, and is not intended to limit the scope of the present invention. Any changes or modifications made by a person skilled in the art in the field of the present invention based on the above disclosure shall fall within the scope of protection of the claims.
Claims
1. A deep training and learning method for automatic recognition of ground penetrating radar images based on AI, characterized in that: include: S1. Establish a single sample database of three types of soil layers; the three types of soil layers include: abnormal soil layer, soil layer stratification, soil layer voids and loose bodies; S2. Based on the single sample database of the three soil layers and intelligent image analysis technology, an AI database of the three soil layers is established; S3. Train and optimize the AI database of the three soil layers to form single algorithm learning results for the three soil layers; S4, integrating the learning results of the individual algorithms of the three soil layers to form a deep learning integration algorithm; S5. Establish an AI comprehensive geophysical exploration results database; S6. Based on the deep learning integration algorithm, deep learning is performed on the AI comprehensive geophysical exploration results database, and training is performed on inspection recognition, feature recognition, and classification recognition respectively to form the output of AI comprehensive geophysical exploration results diagnosis results; S7. Perform data compatibility and algorithm parameter adjustment on the output of AI comprehensive geophysical exploration results diagnosis to form an AI diagnostic model for radar image data. The AI diagnostic model for radar image data can recognize the input ground penetrating radar image to generate a diagnostic image of the ground penetrating radar image.
2. The deep training learning method according to claim 1, characterized in that: The S1 includes: S11, using a ground penetrating radar to obtain radar image data of three soil layers, and classifying and forming radar image data sets of the three soil layers; the radar image data sets of the three soil layers include: a radar image data set of soil layer anomaly, a radar image data set of soil layer stratification, and a radar image data set of soil layer voids and loose bodies; S12. Preprocess the images and data of the radar image data sets of the three soil layers respectively, and establish a single sample database for each soil layer in a targeted manner.
3. The deep training learning method according to claim 2, characterized in that: In S11, the process of obtaining radar image data of three soil layers is as follows: The ground penetrating radar is used to scan the reinforced concrete medium structure with an antenna center frequency of 800MHz, a sampling interval of 0.005m, and a time window of 20ns to obtain a ground penetrating radar image of the steel bar distribution; The ground penetrating radar is used to scan the grouting effect behind the lining ring on the shield lining ring with the center frequency of the antenna being 600MHz, the sampling interval being 0.02m, and the time window being 30ns, and the ground penetrating radar image of the grouting soil layer distribution behind the lining ring is obtained; The ground penetrating radar is used to scan the road surface on the urban road with the center frequency of the antenna being 400MHz, the sampling interval being 0.04m and the time window being 60ns to obtain the ground penetrating radar image of the loose soil layer and the voids under the road surface.
4. The deep training learning method according to claim 2, characterized in that: The S12 includes: Data preprocessing, including static correction removal, filtering and gain; Image preprocessing, including image scaling, translation and cropping.
5. The deep training learning method according to claim 1, characterized in that: The S2 includes: S21, manually labeling soil layer information of the single sample databases of the three soil layers respectively, so as to establish a labeling library of the three soil layers; the labeling library of the three soil layers includes: a soil layer abnormal body labeling library, a soil layer stratification labeling library and a soil layer void and loose body labeling library; S22. Import the annotated image libraries of the three soil layers into the AI database respectively, and establish the AI database of the three soil layers based on intelligent image analysis technology.
6. The deep training learning method according to claim 1, characterized in that: The S3 includes: S31, performing single-item reinforcement training and manual intervention reinforcement training on the AI database of each soil layer to correct the AI database of each soil layer; S32, for each soil layer after correction, the AI database is used to adjust the data volume of the AI interpretation library, and the interpretation rate and accuracy are counted, and error data are collected to establish an error data collection library for each soil layer; S33. Optimize the algorithm based on the collection library data of each soil layer and further improve the revised AI database of each soil layer. Perform statistics on the interpretation rate and accuracy of the improved and revised AI database of each soil layer until the statistical results of the interpretation rate and accuracy meet the predetermined standards. The AI database of each soil layer that meets the predetermined standards is regarded as the single algorithm learning result of one stage.
7. The deep training learning method according to claim 6, characterized in that: In the S32, the statistics of interpretation rate and accuracy based on adjusting the amount of AI interpretation library data include: The data volume of the AI interpretation image library was adjusted to use 500 images, 1000 images, and 2000 images as nodes for interpretation rate and accuracy statistics.
8. The deep training learning method according to claim 7, characterized in that: In S32, the predetermined standard is: the interpretation rate is 95% and the accuracy rate is 90%.
9. The deep training learning method according to claim 1, characterized in that: The S5 includes: S51. Use ground penetrating radar to obtain comprehensive soil layer radar result image data; S52, preprocessing the image and data of the comprehensive soil layer radar result image data to establish a comprehensive soil layer sample database; S53, manually marking soil layer information on the comprehensive soil layer sample database to establish a comprehensive results marking library; S54. Import the comprehensive results annotation library into the AI database, and establish an AI comprehensive geophysical exploration results database based on intelligent image analysis technology.
10. The deep training learning method according to claim 9, characterized in that: The S52 includes data preprocessing, including: static correction removal, filtering and gain; Image preprocessing, including image scaling, translation and cropping.
