Geophone embedding quality intelligent analysis and evaluation method and system

Abstract

This invention provides a method and system for intelligent analysis and evaluation of geophone burial quality. The method includes: Step 1, collecting field photos of geophone burial sites and performing image preprocessing to construct a geophone image sample dataset; Step 2, labeling the geophone image sample dataset to create a label file; Step 3, training and optimizing the model on the geophone image sample dataset to construct a geophone background classification network model; Step 4, training and optimizing the network model on the geophone image sample dataset and the label file to construct a geophone burial status detection network model; Step 5, identifying the geophone burial location and burial status, and analyzing the geophone connectivity and tilt angle. This invention can replace the previous method of manually viewing geophone burial photos and subjectively evaluating the geophone burial quality, realizing intelligent analysis and evaluation of the burial quality of onshore cable seismic acquisition geophones.

Description

Technical Field

[0001] This invention relates to the field of seismic acquisition field geophone installation quality inspection and evaluation technology, and in particular to an intelligent analysis and evaluation method and system for geophone installation quality. Background Technology

[0002] Seismic exploration is one of the main technical means for oil and gas exploration and development. It is a geophysical exploration method that uses artificially generated and received seismic waves to analyze their propagation patterns underground, thereby inferring information about subsurface rock strata and geological structures. This provides crucial evidence for the discovery and evaluation of oil and gas reservoirs. In the seismic wave reception stage, the quality of the field geophone installation is a significant factor affecting the reception of seismic acquisition data.

[0003] To ensure the quality of seismic acquisition data, the state has set requirements for the installation of geophones. The relevant requirements for the quality of onshore geophone installation include: First, the center of the geophone assembly should be aligned with the station number, and the geophones should be installed according to the assembly pattern determined by the technical design or testing. For special terrain, the assembly pattern should be scaled down proportionally or placed along the terrain contour lines. Geophones in the same channel should have consistent installation conditions, good coupling with the ground surface, and meet the requirements of being "flat, stable, upright, straight, and tight." Geophones with damaged casings or without tail cones should not be used during construction. Second, geophone cables should not be suspended above tall poles or crops. Third, the time difference caused by differences in installation elevation within the same geophone assembly should be less than one-quarter of the apparent period of the reflected wave. China Petroleum & Chemical Corporation (Sinopec) has also made clear requirements for the installation of geophones. The requirements related to the quality of geophone installation on land include: First, geophones should be installed according to the combination pattern and number determined by design or testing. For special terrain, the combination pattern should be scaled down proportionally or placed along the terrain contour lines, and the installation conditions of geophones in the same channel should be kept as consistent as possible. Second, the center of the geophone combination should be aligned with the measurement mark, and the positioning difference between the center of the geophone combination and the measurement mark should not exceed 2 meters. If this limit is exceeded, actual measurement should be performed. Third, the coupling between the geophone and the ground should be ensured, achieving the requirements of "flat, stable, upright, straight, and tight." Fourth, the height difference of geophone combinations in the same channel should not exceed 2 meters.

[0004] Currently, a new round of single-point high-density seismic acquisition is underway in the eastern exploration areas of my country. Field operations mainly rely on single-point geophones. The quality inspection of geophone installation in the single-point high-density seismic acquisition project in the eastern exploration areas of my country should focus on four key aspects: First, whether the geophones are installed at the correct locations; second, whether the geophones are properly buried or tightly coupled to the ground; third, whether the geophones are vertically downward; and fourth, whether the geophones are connected to the instrument vehicle.

[0005] To ensure that each geophone is installed according to standards, the seismic team requires the surveying personnel to take a photo with their mobile phones after each geophone is installed. All photos are then compiled on the same day and inspected one by one by the geophone quality control personnel. The quality control personnel check the following: first, whether a photo has been taken for each geophone; second, whether each geophone has been installed on time; and third, whether the installation quality of each geophone meets the requirements of being "flat, stable, straight, upright, and tight." If any problems are found in these three aspects, the field surveying personnel are notified to rectify them the following day. Because more than 30,000 photos are taken in the field every day (based on 40 arrays of reception, with 800 channels per array), it puts a lot of pressure on the quality control personnel of the detectors. In particular, it is difficult to accurately identify whether the detectors are buried or fixed vertically downwards, and whether the detectors are deployed on time, through images. In addition, long hours of manual inspection can easily lead to visual fatigue and errors. Furthermore, the work experience and inspection evaluation standards of different quality inspectors may be inconsistent, which can also reduce the reliability of the inspection results.

[0006] Chinese patent application CN114255388A discloses an AI-based automatic detection method for the burial quality of a seismic acquisition and receiving device, comprising the following steps: S1, capturing field images of the seismic acquisition and receiving device and selecting clear images; S2, using a label image tool to annotate the clear images with features, classifying the images as qualified or unqualified, and saving them as XML tag files; S3, building a YOLOv3 neural network, setting network parameters, performing iterative training, outputting the model after each round of training, and selecting the optimal prediction model according to the evaluation index of prediction accuracy.

[0007] S4. The optimal prediction model is used to classify and predict images of the seismic acquisition and receiving device to be predicted. This invention can quickly and effectively conduct construction quality inspections, improving the effectiveness and timeliness of quality control, reducing quality control costs, increasing construction efficiency, helping quality inspectors to quickly assess the quality of instrument installation, and ensuring the quality of seismic data acquisition. This application only uses a YOLOv3 neural network model to perform target detection on detector embedding photos, and the sample labeling granularity is relatively coarse, only dividing samples into "qualified" and "unqualified" labels. It lacks detailed labeling and classification of specific problem points, affecting the model's ability to locate and identify problems. This limitation also makes it difficult for the constructed network model to generalize to more complex actual construction scenarios. Furthermore, this method does not detect key quality inspection indicators such as the correct embedding of the detector, detector connectivity, and detector embedding tilt angle, making it unsuitable for actual production.

[0008] Chinese patent application CN106054239B discloses a microseismic fracturing monitoring method, comprising: (A) acquiring existing seismic data of the work area and establishing a geological model; (B) determining the position and order of the geophones of a deep-well microseismic observation system placed in the monitoring well based on the existing seismic data and the geological model; (C) determining the target well section involved in the target layer for observation using a surface microseismic observation system based on the placement position of the geophones of the deep-well microseismic observation system; and (D) determining the position of the geophones of the surface microseismic observation system for the target well section. This method can improve the accuracy of monitoring results, laying the foundation for determining the morphology of artificial fractures in real-time and evaluating the effect of reservoir volume modification. While the method in this application involves geophone placement and burial, it does not inspect the quality of geophone placement, and the method does not involve the field of artificial intelligence.

[0009] Chinese patent application CN111796327A discloses a microseismic ground monitoring device and a source location method, comprising a data acquisition and processing module, a signal transmission module, and a power supply module. The data acquisition and processing module includes a seismic detector for receiving microseismic simulation signals. The seismic detector is connected to a signal conditioning module, which in turn is connected to an A / D conversion module. The A / D conversion module is connected to a processor module. The processor module is connected to a data storage module and a global navigation satellite module. The processor module is also connected to the signal transmission module. The power supply module is connected to the data acquisition and processing module and provides power to it. This invention employs superimposed energy scanning imaging technology that does not require extraction of the initial arrival of signal events for ground source location. It can effectively determine the source location even with low requirements for signal source quality. Furthermore, it supports real-time microseismic location monitoring via a server and utilizes mobile 4G signals, offering the advantage of low signal transmission costs. The application describes the development of a microseismic ground monitoring device that involves burying a geophone underground to receive simulated microseismic signals, converting them into digital signals, and then using model-based ray tracing technology and the principle of energy superposition to estimate the location of the source point of a possible microseismic event. However, the patent in question only involves burying the geophone underground, without conducting quality testing on the geophone burial, and the method does not involve the field of artificial intelligence.

[0010] The existing technologies described above are significantly different from this invention. A search reveals no literature of the XY category, indicating the innovativeness of this invention. Since existing technologies lack solutions to the technical problems we seek to address, we have invented a novel intelligent analysis and evaluation method and system for detector embedding quality. Summary of the Invention

[0011] The purpose of this invention is to provide a highly efficient and automated intelligent analysis and evaluation method and system for the embedding quality of detectors that can be used in actual production.

[0012] The objective of this invention can be achieved through the following technical measures: an intelligent analysis and evaluation method for detector embedding quality, comprising:

[0013] Step 1: Collect photos of the field geophone installation sites and perform image preprocessing to construct a geophone image sample dataset;

[0014] Step 2: Label the detector image sample dataset and create a label file;

[0015] Step 3: Train the model and fine-tune the parameters on the detector image sample dataset to build a detector background classification network model;

[0016] Step 4: Train the network model and fine-tune the parameters of the detector image sample dataset and label file to build a network model for detecting the detector's burial status.

[0017] Step 5: Identify the location and status of the detector installation, and analyze the connectivity and tilt angle of the detector.

[0018] The objective of this invention can also be achieved through the following technical measures:

[0019] In step 1, images of the geophones buried at the construction site are collected and preprocessed, including denoising, data augmentation, and image enhancement, to construct a geophone image sample dataset.

[0020] Step 1 includes:

[0021] S101. Perform median filtering on the acquired image to remove noise interference and improve image smoothness;

[0022] S102. Perform data augmentation processing on the image after denoising in step S101 to reduce the occurrence of data sample imbalance.

[0023] S103. The image after data enhancement in step S102 is improved by histogram equalization image enhancement technology to improve the image quality.

[0024] S104. Merge the images after data preprocessing in the above three steps S101, S102 and S103 with the original detector embedding images to construct an initial detector image sample dataset.

[0025] S105. The initial detector image sample dataset is classified according to the detector burial background, into a detector image sample dataset with a field background and a detector image sample dataset with a hardened road background, collectively referred to as the detector image sample dataset, which is used for training the subsequent detector background classification network model and detector burial status detection network model.

[0026] Step 2 includes:

[0027] S201. Based on the characteristics of the detector's burial, define image label categories for two background backgrounds: farmland and hardened road surface.

[0028] Against a field background, the image label categories are set as follows: front-facing detector, back-facing detector, tilted detector, detector tail cone, and acquisition station (5 categories); against a hardened road background, the image label categories are set as follows: front-facing detector, back-facing detector, tilted detector, detector tail cone, acquisition station, and adhesive (6 categories).

[0029] S202. Label the samples under two different backgrounds: farmland and paved road surface, and output the label files respectively.

[0030] The tag file format includes: image width, height, depth, target category name, and target location coordinates.

[0031] In step 3, the detector image sample dataset formed in step 1 is loaded, and the ResNet-18 network model is trained and its parameters are tuned to construct the ResNet-18 detector background classification network model. The ResNet-18 network structure is used to refine the scene classification problem into a typical binary classification problem. The gradient vanishing problem in deep networks is solved by introducing residual learning units. The model has 18 layers of convolutional and residual blocks and 3 layers of fully connected parts, for a total of 21 layers.

[0032] Step 3 specifically includes:

[0033] S301: Load the detector image sample dataset formed in step 1, and divide it into training set, validation set and test set according to 8:1:1;

[0034] S302: To improve the training efficiency of the background classification network model, the training set images in the detector image sample dataset are uniformly scaled to a fixed size, and the pixel values ​​of the images are normalized.

[0035] S303: Construct a ResNet-18 detector background classification network;

[0036] S304: Perform model training and parameter tuning to obtain the ResNet-18 detector background classification network model.

[0037] In step 4, the YOLOv5 network model is trained and its parameters are tuned on the detector image sample dataset and label file to construct the YOLOv5 detector embedding status detection network model.

[0038] Step 4 includes:

[0039] S401: Load the detector image sample dataset constructed in step 1 and the label file formed in step 2, and divide the dataset into training set, validation set and test set according to 8:1:1;

[0040] S402: Construct the YOLOv5 detector embedding status detection network. The construction method is as follows:

[0041] A YOLOv5 backbone network is constructed to extract multi-scale features from images. A cross-stage partial residual structure is built to reduce computation while maintaining feature representation ability.

[0042] A feature pyramid is constructed to fuse features at different scales to detect targets of multiple sizes. At the same time, cross-scale residual connections are introduced into the feature pyramid for more efficient feature propagation. A transformation module between shallow and deep features is designed to achieve efficient connection, enhance the expressive power of the feature pyramid, and improve the detection capability of small targets.

[0043] A detection module is constructed, which uses a clustering algorithm to analyze the size of the target boxes in the data image, adjusts the anchor box mechanism to suit the detector state and size, and predicts the center point coordinates, width and height, confidence score, and class probability distribution of the output target boxes.

[0044] S403: Model training and tuning. Based on the performance of the server GPU card, after repeated testing, the training set is uniformly sized before model training. During the training process, the model learns how to detect and locate the detector in the image.

[0045] S404: Model Validation and Testing. Evaluate the model's detection performance on the validation set, calculate commonly used metrics such as mAP, and the precision and recall of various detection targets.

[0046] S405: Model testing. The model is tested on the test set to evaluate its potential performance on real data. Based on the test results, the model architecture and training parameters are repeatedly adjusted to finally obtain a YOLOv5 detector embedding status detection model with high recognition accuracy.

[0047] The intelligent analysis and evaluation method for detector burial quality also includes, after step 4, setting the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area, as well as the sampling interval and other construction parameters; loading the trained ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, the daily detector burial construction task file, the daily detector burial status inspection data file, and the on-site photos of each detector burial condition.

[0048] After step 4, when setting the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system, as well as the sampling interval, the specific steps include:

[0049] S501: Set seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area. The seven parameters include: translation distance in the x direction, translation distance in the y direction, translation distance in the z direction, rotation angle in the x direction, rotation angle in the y direction, rotation angle in the z direction, scale factor, and central meridian.

[0050] S502: Set the construction parameters for the work area, including: sampling interval, acquisition station model, minimum resistance, maximum resistance, tilt range, detector type, and the number and model of detectors.

[0051] After step 4, load the trained ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, the daily detector burial construction task file, the daily detector burial status inspection data file, and the on-site photos of each detector burial status, specifically including:

[0052] S601: Load the ResNet-18 detector background classification network model trained in step 3;

[0053] S602: Load the YOLOv5 detector embedding status detection network model obtained from step 4;

[0054] S603: Load the geophone installation task file for the day. The task file includes: serial number, line number, station number, east coordinate, north coordinate, geophone type, and geophone quantity information.

[0055] S604: Load the daily inspection data file of the geophone burial status for the day. The daily inspection file includes: serial number, acquisition station type, line number, station number, resistance value, and tilt angle information.

[0056] S605: Loads on-site photos of the burial status of each detector taken on the same day. The photo naming rule is: line number-point number-longitude-latitude-elevation. The photo shooting software takes photos with the following rules: the lower left corner of the photo should show the line number, point number, longitude, latitude, and elevation.

[0057] Step 5 identifies the detector's burial location and burial status, analyzes the detector's connectivity and tilt angle, and outputs the detector burial quality evaluation results; specifically including:

[0058] S701. Identify the location of the geophone burial site and determine if the geophone is installed on the correct location; intelligently identify the line number, station number, longitude, and latitude on the geophone burial photo and determine if they match the file name of the geophone burial photo. If they match, the geophone burial photo is taken and named in accordance with regulations; if not, the geophone burial photo is taken and named in accordance with regulations. For geophone burial photos with inappropriate names, output a list of geophone line numbers and station numbers, prompting the geophone quality control personnel to check and modify them in a timely manner.

[0059] Obtain the geophone burial construction task file and establish a one-to-one correspondence between the line numbers and point numbers intelligently identified in the geophone burial photos and the line numbers and point numbers in the geophone burial construction task file.

[0060] Obtain the seven parameters for the conversion between the WGS84 coordinate system and the Beijing 54 coordinate system, and convert the latitude and longitude coordinates in the WGS84 coordinate system identified in the detector embedding photo into the Beijing 54 coordinate system.

[0061] The difference between the theoretical design coordinates of the detector in the construction task file and the Beijing 54 coordinates obtained from the detector installation photo is calculated. If the difference is less than the preset value, the detector is considered to be installed on time. If the difference is greater than the preset value, the detector is considered to be installed on time.

[0062] S702. Identify the burial status of the detector; use the ResNet-18 detector background classification network model to intelligently identify whether the construction background of the detector burial image is farmland or hardened road surface.

[0063] If the detector is in a field background as identified above, then the detector burial quality inspection standard in the field background is activated, and the YOLOv5 detector burial status detection network model is used to identify the detector burial status and give the detector burial status identification result.

[0064] If the above identification result is a hardened road surface background, then the detector embedding quality inspection standard under hardened road surface background is activated, and the YOLOv5 detector embedding status detection network model is used to identify the detector embedding status and give the detector embedding status identification result.

[0065] There are two criteria for judging the quality of geophone burial in field background. If either one is met, it is considered qualified: First, the geophone tail cone is vertically inserted below the ground at a preset ratio; second, the geophone is buried underground and covered with a thin layer of soil, and the acquisition station is near the geophone.

[0066] There are four criteria for judging the quality of geophone burial in field background. If any one of them is met, it is considered unqualified: First, the geophone tail cone is not vertically inserted into the ground at the preset ratio; second, the geophone is placed horizontally on the ground; third, the geophone tail cone is vertically upward; fourth, there is no acquisition station near the geophone burial pit.

[0067] The criteria for judging the quality of geophone burial under hardened road surface background is: the geophone tail cone is removed and firmly glued to the ground with adhesive;

[0068] There are six criteria for judging the quality of geophone installation in hardened road surfaces as unqualified. Meeting any one of these criteria constitutes a failure: 1. The geophone tail cone is not removed and is inserted above a sewer; 2. The geophone tail cone is not removed and is placed horizontally on the ground; 3. The geophone tail cone is not removed and is inverted vertically upwards; 4. The geophone tail cone is not removed and is inserted obliquely in a corner; 5. The geophone tail cone has been removed but not glued to the ground; 6. The geophone tail cone has been removed and was initially glued to the ground, but later the geophone detached or tipped over.

[0069] In the process of detector burial status identification, the detector burial status detection model predicts the bounding box position, confidence level, and class probability of each feature point. Each feature point can predict multiple bounding boxes. During the prediction process, multiple candidate bounding boxes are generated, which may include overlapping boxes. In order to reduce the number of redundant boxes, the non-maximum suppression method is used to retain the box with the highest confidence level and filter out other boxes with large overlap with it, and finally obtain a detector burial status identification result with high accuracy.

[0070] S703. Analyze whether the detector connection status and detector tilt angle status are qualified.

[0071] Obtain the set threshold range for the resistance value and the threshold range for the detector tilt angle in the work area;

[0072] Sequentially acquire the resistance and tilt angle status values ​​of each detector in the daily inspection file of the instrument. If the resistance or tilt angle status of the detector is not within the predetermined threshold range, the detector is determined to be in a non-connected state or the tilt angle status of the detector is unqualified.

[0073] S704. Classify the analysis results obtained from S701, S702 and S703 according to the detector line number and point number, and output the analysis and evaluation results of whether the detector burial status is qualified.

[0074] The objective of this invention can also be achieved through the following technical measures: an intelligent analysis and evaluation system for detector embedding quality, comprising:

[0075] The dataset construction module builds a sample dataset of detector images.

[0076] The dataset sample annotation module is used to create sample labels for detector image sample datasets;

[0077] The detector background classification network model building module trains the detector image sample dataset to obtain the detector background classification network model.

[0078] The detector state detection network model construction module trains the detector image sample dataset to obtain the detector embedding state detection network model;

[0079] The basic parameter setting module for the work area allows you to set the seven coordinate transformation parameters and sampling interval for the construction work area.

[0080] The network model and detector burial construction task file input module loads the trained detector background classification network model and detector burial status detection network model, the detector burial photos to be analyzed and evaluated, the detector burial construction task file, and the detector burial status daily inspection data file, providing a data foundation for the subsequent detector burial quality evaluation module.

[0081] The intelligent analysis and evaluation module for geophone burial quality analyzes the seismic acquisition geophone images in the construction area based on the geophone background classification network model and the geophone burial status detection model, and outputs the evaluation results of whether the geophone burial quality is qualified.

[0082] The objective of this invention can also be achieved through the following technical measures:

[0083] The dataset construction module includes:

[0084] The raw dataset construction unit is used to construct the raw detector image sample dataset;

[0085] The image denoising preprocessing unit is used to perform median filtering denoising on the original detector image sample dataset;

[0086] The data augmentation preprocessing unit is used to perform data augmentation processing on the denoised image;

[0087] The image enhancement preprocessing unit is used to perform image enhancement processing on the data after data enhancement.

[0088] The detector image sample dataset construction unit is used to construct the detector image sample dataset.

[0089] The dataset sample annotation module includes:

[0090] The sample label category definition unit is used to define the marker feature category of whether the detector to be tested is qualified or unqualified.

[0091] Sample labeling unit, used for creating sample labels for detector image sample datasets;

[0092] The sample label file output unit is used to output the labeled sample label file.

[0093] The detector background detection model training module includes:

[0094] The detector image sample training set, validation set, and test set generation unit is used to load the detector image sample dataset and divide it according to an 8:1:1 ratio to form the training set, validation set, and test set.

[0095] The training set image preprocessing unit is used to uniformly scale the images to a fixed size and normalize the pixel values ​​of the images.

[0096] Detector background classification network building unit, used to build ResNet-18 detector background classification network;

[0097] The detector background classification network model training unit is used to train the detector image sample dataset to obtain the ResNet-18 detector background classification network model.

[0098] The detector background classification network model output unit is used to output the ResNet-18 detector background classification network model file obtained from the above training.

[0099] The detector state detection network model training module includes:

[0100] The label file loading unit is used to load the label file output by the dataset sample annotation module;

[0101] Detector status detection network construction unit, used to construct YOLOv5 detector status detection network;

[0102] The model training and tuning unit is used for preprocessing training set data and optimizing model parameters.

[0103] The model validation and testing unit is used to evaluate the detection performance of the model on the validation set, and adjust the model architecture or training parameters through human-computer interaction based on the test results.

[0104] The target detection unit is used to intelligently identify the buried status of the detector. When the identification accuracy reaches the expected target, it is determined as the final detector status detection network model.

[0105] The YOLOv5 detector state detection network model output unit is used to output the YOLOv5 detector state detection network model file obtained from the above training.

[0106] The basic parameter setting module for the work area includes:

[0107] The coordinate transformation seven-parameter setting unit for the construction area is used to set the seven parameters for the transformation from the WGS84 coordinate system to the Beijing 54 coordinate system of the construction area.

[0108] The sampling interval basic construction parameter setting unit for the construction area is used to set the basic construction parameters such as the sampling interval of the construction area.

[0109] The network model and detector embedding construction task file input module includes:

[0110] The ResNet-18 detector background classification network model loading unit is used to load the ResNet-18 detector background classification network model.

[0111] YOLOv5 detector embedding status detection network model loading unit, used to load the YOLOv5 detector embedding status detection network model;

[0112] The detector embedding photo loading unit is used to batch load the detector embedding photos to be analyzed and evaluated.

[0113] The detector embedding construction task file loading unit is used to load the detector embedding construction task file;

[0114] The detector embedding status daily inspection data file loading unit is used to load the detector embedding status daily inspection data file.

[0115] The detector embedding quality evaluation module includes:

[0116] The detector embedding location intelligent detection unit is used to identify the latitude and longitude coordinates in the detector image;

[0117] The coordinate transformation unit is used to convert the WGS84 coordinates identified in the detector image into Beijing 54 coordinates.

[0118] The on-time deployment judgment unit is used to calculate the difference between the Beijing 54 coordinates of the detector image and the theoretical coordinates of the detector in the construction task file, and to give the judgment result of whether the deployment is on time.

[0119] The intelligent detection unit for detector embedding quality is used to identify whether the detector embedding quality in the detector image is up to standard and to provide analysis and evaluation results.

[0120] The detector resistance analysis and evaluation unit is used to analyze whether the detector is connected and to give the evaluation result.

[0121] The detector tilt angle analysis and evaluation unit is used to analyze whether the detector is vertical and to give the evaluation result;

[0122] The detector embedding quality evaluation result output unit is used to summarize and output the detector line number, point number, and reasons for the non-compliance of the embedding quality of the detectors, providing data for field rectification.

[0123] The intelligent analysis and evaluation method and system for detector burial quality in this invention includes: collecting field photos of detector burial sites and performing image preprocessing to construct a detector image sample dataset; labeling the detector image sample dataset to create a label file; training and optimizing the parameters of a ResNet-18 network model on the detector image sample dataset to construct a ResNet-18 detector background classification network model; and training and optimizing the parameters of a YOLOv5 network model on the detector image sample dataset and label file to construct a YOLOv5 detector burial quality assessment network model. The invention employs a geophone burial status detection network model; sets seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area, as well as construction parameters such as sampling intervals; loads the ResNet-18 geophone background classification network model, the YOLOv5 geophone burial status detection network model trained above, the daily geophone burial construction task file, the daily geophone burial status inspection data file, and the photographs taken at each geophone burial site; intelligently identifies the geophone burial location and status, automatically analyzes the geophone connectivity and tilt angle, and automatically outputs the geophone burial quality evaluation results. This invention can replace the previous method of manually viewing geophone burial photographs and subjectively evaluating the geophone burial quality, realizing intelligent analysis and evaluation of the burial quality of geophones in terrestrial cable seismic acquisition.

[0124] The intelligent analysis and evaluation method and system for geophone installation quality in this invention can solve the problems of low efficiency, high error rate, and serious quality risks associated with manual indoor inspection of geophone installation quality during actual seismic acquisition. Furthermore, by incorporating methods from the field of artificial intelligence, it achieves automated detection and evaluation of geophone installation quality. This invention can replace the previous method of manually reviewing geophone installation photographs and subjectively evaluating the quality of geophone installation, enabling intelligent analysis and evaluation of the quality of geophone installations in terrestrial cable-stayed seismic acquisition.

[0125] Compared with the prior art, the present invention has the following technical advantages:

[0126] (1) This patented solution adopts multi-model fusion technology and simultaneously constructs two neural network models, ResNet-18 and YOLOv5, which are used for detector background classification and detector burial status detection, respectively. This multi-model fusion method can classify and identify and analyze the burial status indicators of the detector under different construction backgrounds, which is in line with actual production.

[0127] (2) The method in this patent implements data preprocessing techniques to construct more diverse and high-quality datasets that conform to actual production scenarios, which helps improve the accuracy of subsequent model training. In contrast, the solution in the patent search simply selects clear images, which is not conducive to training a network model that conforms to actual production scenarios.

[0128] (3) The target detection model constructed in this patent method is based on the YOLOv5 network model, and the parameters of the network model are optimized according to the characteristics of the detector image. At the same time, the multi-scale feature pyramid structure of YOLOv5 is used to fuse feature information at different levels, which can capture the overall outline of the detector and identify detailed features, thereby improving the detection accuracy. Compared with the YOLOv3 network model mentioned in the searched patent, the YOLOv5 detector burial status detection model constructed by this patent method has higher detection accuracy and speed, and can better meet the needs of actual production applications.

[0129] (4) This patent solution also utilizes geographic information data, detector burial construction task files and detector burial status daily inspection data files, combined with artificial intelligence target detection technology, to detect key quality indicators in the field such as whether the detector is fixed-point deployed, detector connectivity status, and detector tilt angle status, to ensure that the detector burial quality is qualified.

[0130] (5) The method of this patent also realizes a set of intelligent analysis system for detector embedding quality, which automatically identifies and analyzes the detector embedding quality and outputs the detector embedding quality analysis results. This automated output method can greatly improve the efficiency and practicality of detector embedding quality detection.

[0131] In summary, this patented solution has made breakthroughs and innovations in multi-model fusion, data preprocessing, network model accuracy, comprehensive information application such as geographic information, and automated output. It can better solve technical problems in practical applications, and these advantages lay the foundation for the effectiveness and operability of the patented method. Attached Figure Description

[0132] Figure 1 A flowchart of a specific embodiment of the intelligent analysis and evaluation method for detector embedding quality of the present invention;

[0133] Figure 2 This is a schematic diagram of the detector embedding status detection tag file mentioned in this invention;

[0134] Figure 3 This is a schematic diagram of the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system, as well as the sampling interval and other construction parameters, set in a specific embodiment of the present invention.

[0135] Figure 4 This is a schematic diagram of the daily geophone burial construction task file and the daily inspection data file of the geophone burial status loaded in a specific embodiment of the present invention.

[0136] Figure 5 This is a schematic diagram of the detector embedding location analysis and evaluation results obtained by intelligent analysis in a specific embodiment of the present invention.

[0137] Figure 6 This is a schematic diagram of the analysis results of the qualified and unqualified detector burial status obtained by intelligent analysis under two backgrounds: farmland and hardened road surface, in a specific embodiment of the present invention.

[0138] Figure 7 This is a schematic diagram of the detector connectivity and tilt angle results obtained by intelligent identification in a specific embodiment of the present invention.

[0139] Figure 8 This is an automatically statistically outputted analysis and evaluation result of the detector embedding quality in a specific embodiment of the present invention.

[0140] Figure 9 This is a structural diagram of the intelligent analysis system for detector embedding quality according to the present invention. Detailed Implementation

[0141] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0142] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments of the present invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, and / or combinations thereof.

[0143] The intelligent analysis and evaluation method for detector embedding quality of the present invention includes:

[0144] Step 1: Collect images of the geophones buried at the construction site, perform preprocessing including denoising, data augmentation, and image enhancement, and construct a geophone image sample dataset to ensure the accuracy of subsequent model training;

[0145] Step 2: Label the detector image sample dataset and create a label file;

[0146] Step 3: Train and fine-tune the ResNet-18 network model on the detector image sample dataset to construct the ResNet-18 detector background classification network model;

[0147] Step 4: Train and optimize the YOLOv5 network model and parameters of the detector image sample dataset and label file to build a YOLOv5 detector embedding status detection network model.

[0148] Step 5: Set the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area, as well as construction parameters such as the sampling interval;

[0149] Step 6: Load the ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, the detector burial construction task file for the day, the detector burial status daily inspection data file, and the on-site photos of each detector burial status taken above.

[0150] Step 7: Intelligently identify the location and status of the geophone burial, automatically analyze the connection status and tilt angle of the geophone, and automatically output the geophone burial quality analysis and evaluation results.

[0151] The main advantages of this invention are: (1) It uses artificial intelligence technology to replace the traditional manual verification method, which improves the efficiency of geophone burial quality inspection, reduces labor and time costs, reduces the dependence on the work experience of quality control personnel, improves the quality of geophone burial quality inspection, and thus improves the quality of seismic acquisition data; (2) It collects a large number of images of different geophone burial scenarios in the eastern exploration area and preprocesses the images to construct a diverse geophone burial status dataset, ensuring the accuracy of subsequent model training and improving the accuracy of intelligent detection of geophone burial quality; (3) It optimizes the parameters of the ResNet-18 network model and the YOLOv5 network model to make them more suitable for different geophone burial scenarios in the eastern exploration area; (4) It combines the deep learning models of ResNet-18 and YOLOv5 to improve the reliability of intelligent detection results of geophone burial quality. First, the ResNet-18 network model is used to automatically classify the detector burial photos according to the field construction background. The detector burial photos after background classification are then used for automatic detection of the detector burial status. This method can effectively reduce the interference of environmental background on the detection of detector burial status, thereby improving the accuracy of detector burial quality detection. (5) The combined application of artificial intelligence technology and daily inspection data files of detector burial status to comprehensively analyze the detector burial quality further reduces the probability of identification errors that may be caused by using only artificial intelligence technology. (6) A set of intelligent analysis system for detector burial quality was studied, realizing the full-process automation and visualization from the import of detector burial photos into the system to the output of intelligent detection results, thereby realizing industrial application in field production. (7) After the completion of each construction area, the sample dataset is updated in a timely manner. The ResNet-18 detector background classification network model and YOLOv5 detector burial status detection model are continuously corrected according to the updated detector burial image sample dataset, continuously improving the model accuracy, thereby continuously improving the accuracy of detector burial quality analysis and evaluation.

[0152] Example 1

[0153] In a specific embodiment 1 of the present invention, the present invention is applied. Figure 1 This is a flowchart of the intelligent analysis and evaluation method for detector embedding quality according to the present invention, as shown below. Figure 1 As shown, a method and system for intelligent analysis and evaluation of detector embedding quality includes:

[0154] Step 1: Collect photos of the field detector installation sites (usually in JPEG, JPG, BMP, PNG, etc.) and perform preprocessing such as denoising, data augmentation, and image enhancement on the images to construct a detector image sample dataset;

[0155] Due to the influence of various factors such as shooting equipment and shooting environment, as well as the limitations of field construction conditions, there is a high probability of inconsistent sample quality and sample imbalance, which will affect the accuracy of automatic detection of detector embedding quality in the later stage. Therefore, it is necessary to preprocess the acquired images before constructing the dataset. The specific method is as follows:

[0156] S101. Perform median filtering on the acquired image to remove noise interference and improve image smoothness;

[0157] Median filtering is a non-linear filtering method. Its core idea is to smooth the image by replacing pixel values ​​while preserving edge details as much as possible. It is highly effective in handling impulse noise and salt-and-pepper noise, effectively protecting the image's edge information. The specific implementation method is as follows:

[0158] a) Selecting the filtering window: Select a filtering scale factor, denoted as k (k is a positive integer), then the corresponding filtering window size is (2k+1)×(2k+1). The value of k determines the smoothness of the image. Since the noise of the detector buried image in the eastern exploration area is relatively small and the noise density is low, the smoothness is too low and it is easy to produce singular value phenomena. The smoothness is too high and it is easy to cause image distortion. Through repeated comparative experiments, k=2 is selected in the method of this invention, that is, the filtering window size is 5×5.

[0159] b) Move the filter window pixel by pixel from the top left corner of the image, gradually covering the entire image. Each time the window moves, it will cover one pixel and its surrounding neighborhood.

[0160] c) Extract the pixel values ​​of all pixels within the coverage area of ​​the filter window, and put these pixel values ​​into a set, denoted as X, where X = {x1, x2, ..., x...} N} where N is the number of pixels within the filtering window. When the filtering window is complete, N = (2k + 1). 2 When the filtering window is located at the edge of the image, the filtering window is incomplete, so N is taken as the number of pixels actually within the filtering window. The pixel values ​​in set X are sorted in ascending order to obtain an ordered sequence, denoted as set X′, where X′ = {x}. (1) ,x (2) ,...,x (N)};

[0161] d) Select the value at the middle position from the ordered sequence X′ as the new value of the center pixel of the filter window, denoted as y. If N is odd, the median is x. ((N+1) / 2) If N is even, then the median is x. (N / 2) For each pixel n (n = 1, 2, ..., N), assuming the pixel position of pixel n in the original image is (i, j), then the pixel value of pixel (i, j) in the original image is denoted as x.(i,j) The pixel value in the new image after filtering is denoted as y. ( i ,j) The pixel value filtering processing method is as follows:

[0162] y (i,j) =median{x (i+m,j+n) -k≤m,n≤k

[0163] Where: x (i,j) y represents the pixel value at position (i,j) in the original image. (i,j) This represents the pixel value at position (i,j) in the filtered image, where m and n are the offsets of this pixel relative to the center pixel within the filtering window, and median{·} represents taking the median of the set.

[0164] S102. Perform data augmentation processing on the image after denoising in step S101 to reduce the occurrence of data sample imbalance.

[0165] The specific method is as follows:

[0166] Perform image flipping processing on the original image;

[0167] The original image is randomly cropped so that the target to be detected appears in different positions of the image at different proportions;

[0168] Adjust the brightness, contrast, saturation, and hue of the original image, and randomly change the image colors.

[0169] S103. The image after data enhancement in step S102 is improved by histogram equalization image enhancement technology to improve the image quality.

[0170] By redistributing the gray levels of image pixels, the gray level distribution of the image is made more uniform, thereby enhancing the visual effect of the image. The specific method is as follows:

[0171] Count the frequency of each gray level in the image to form a gray-level histogram, and calculate the r of each gray level in the original image. k frequency n k That is, the number of pixels that appear at that gray level, where k = 0, 1, ..., K-1, and K is the total number of gray levels in the image;

[0172] Calculate the probability density for each gray level, denoted as P. r (r k Assuming the total number of pixels in the original image is N, then the gray levels r k The probability density is:

[0173]

[0174] Where k = 0, 1, ..., K-1, K is the total number of gray levels in the image;

[0175] The cumulative distribution function is calculated from the histogram, which is the proportion of pixels with a gray level less than or equal to a certain value out of the total number of pixels in the image, denoted as c(r). k The calculation method is as follows:

[0176]

[0177] Where k = 0, 1, ..., K-1, K is the total number of gray levels in the image;

[0178] The cumulative distribution function is used to convert the gray levels r of the original image. k Mapped to new grayscale level s k This creates a new image:

[0179]

[0180] Where k = 0, 1, ..., K-1, K is the total number of gray levels in the image. This indicates the floor function.

[0181] S104. Merge the images after data preprocessing in the above three steps S101, S102 and S103 with the original detector embedding images to construct an initial detector image sample dataset.

[0182] S105. The initial detector image sample dataset is divided into two categories according to the detector burial background: a detector image sample dataset with a field background and a detector image sample dataset with a hardened road surface background. These are collectively referred to as the detector image sample dataset and are used for training the subsequent detector background classification network model and detector burial quality detection network model.

[0183] Step 2: Label the detector image sample dataset and create a label file;

[0184] S201. Based on the characteristics of geophone burial in the eastern exploration area, define image label categories for two backgrounds: farmland and hardened road surface.

[0185] Against a field background, the image label categories are set to 5 categories: front-facing detector, back-facing detector, tilted detector, detector tail cone, and acquisition station; against a hardened road background, the image label categories are set to 6 categories: front-facing detector, back-facing detector, tilted detector, detector tail cone, acquisition station, and adhesive.

[0186] S202. Label the samples under two different backgrounds: farmland and paved road surface, and output the label files respectively.

[0187] The tag file format includes: image width, height, depth, target category name, and target location coordinates, such as... Figure 2 As shown.

[0188] Step 3: Load the detector image sample dataset formed in Step 1, train and optimize the ResNet-18 network model, and build a ResNet-18 detector background classification network model suitable for the eastern exploration area.

[0189] The scene classification problem is refined into a typical binary classification problem using the ResNet-18 network structure. The vanishing gradient problem in deep networks is addressed by introducing residual learning units. The model consists of 18 layers (convolutional and residual blocks) and 3 fully connected layers, for a total of 21 layers. The specific method is as follows:

[0190] S301: Load the detector image sample dataset formed in step 1, and divide it into training set, validation set and test set according to 8:1:1;

[0191] S302: To improve the training efficiency of the background classification network model, the training set images in the detector image sample dataset are uniformly scaled to a fixed size of 224*224 (pixels), and the pixel values ​​of the images are normalized, scaling the pixel value range from [0,255] to [0,1].

[0192] S303: Construct a ResNet-18 detector background classification network adapted to the eastern exploration area;

[0193] The specific construction method is as follows:

[0194] Construct the input layer: Load the 224*224 pixel image generated in S302 above;

[0195] Constructing convolutional layers: Set a 7*7 convolutional kernel, 64 filters, stride of 2, and padding of 3 to initially extract low-level features. The output size is 112*112*64 (where 112 represents the number of pixels and 64 represents the number of channels).

[0196] Constructing a pooling layer: Set a 3*3 max pooling layer with a stride of 2. The output size is 56*56*64 (where 56 represents the number of pixels and 64 represents the number of channels).

[0197] Residual blocks are defined as follows: ResNet-18 contains 8 residual blocks, each consisting of two convolutional layers and one skip connection. The residual blocks are divided into four stages based on the input size and depth, with each stage containing two residual blocks.

[0198] Constructing a fully connected layer: Based on the characteristics of geophone burial in the eastern exploration area, farmland and hardened roads have different background features (such as texture and lighting). Especially when there are more interference factors (such as cracks and gravel) on hardened roads, the model needs to have stronger feature representation capabilities. Therefore, this invention adds an intermediate fully connected layer to gradually reduce the features from high dimension to low dimension, alleviating the information loss problem caused by direct compression and allowing the model to learn more detailed features in complex scenes. Therefore, to alleviate the "information bottleneck" problem, the method of this invention constructs two fully connected layers. The first layer creates a fully connected layer with 512 output features, retaining and recombining the original feature information. The second layer creates a fully connected layer with 2 output features, reducing the dimension from 512 to 2, and finally completing the classification task. The gradual compression method enables the model to extract more key information from diverse features.

[0199] Constructing the output layer: Using the softmax activation function, the output of the fully connected layer is transformed into a probability distribution for each class. Given an input vector Z, the softmax activation function is expressed as:

[0200]

[0201] Where z i It is the i-th element in the input vector Z. It involves performing an exponentiation operation on each input value, where n is the total number of elements in the input vector, i.e., the total number of categories. It is the sum of the exponents of all elements in the input vector, ensuring that the sum of all output values ​​is 1.

[0202] Based on the above, convolutional and pooling layers are used for initial feature extraction and spatial size compression; residual blocks retain input information through residual connections, avoiding the gradient vanishing problem and allowing the network to deepen; fully connected layers map deep features to class probabilities; the output layer is changed to a fully connected layer with 2 neurons, and the softmax activation function is used to output the probabilities of the fields and hardened surfaces, thus obtaining the predicted probability of each class.

[0203] S304: Model training and parameter tuning to obtain a ResNet-18 detector background classification network model suitable for the scenario of seismic acquisition detector burial in the eastern exploration area.

[0204] Define the loss function and optimizer, and use the binary classification cross-entropy loss function to measure the difference between the predicted result and the true label, where the cross-entropy loss function is expressed as:

[0205]

[0206] Where N is the sample size, y iFor the true label of sample i, Let be the predicted value of sample i (the probability obtained through softmax), representing the probability of belonging to the positive class, with a range of [0, 1].

[0207] The cross-entropy loss function measures the difference between the model's output probability and the true label, and optimizes it so that the model can accurately predict whether a sample belongs to farmland or a paved road.

[0208] The model is trained and its detection performance is evaluated on the validation set. The model is then tested on the test set to evaluate its performance on unseen real-world data. Based on the test results, the model architecture is repeatedly adjusted and the training parameters are optimized to finally obtain a ResNet-18 detector background classification network model with high recognition accuracy.

[0209] Step 4: Load the detector image sample dataset constructed in Step 1 and the label file formed in Step 2, train and optimize the YOLOv5 network model, and construct a YOLOv5 detector burial status detection network model suitable for the eastern exploration area.

[0210] The specific method is as follows:

[0211] S401: Load the detector image sample dataset constructed in step 1 and the label file formed in step 2, and divide the dataset into training set, validation set and test set according to 8:1:1;

[0212] S402: Construct a YOLOv5 geophone burial status monitoring network suitable for the eastern exploration area. The construction method is as follows:

[0213] A YOLOv5 backbone network (CSPDarknet53) was constructed to extract multi-scale features of images. A cross-stage partial residual structure was constructed to reduce the amount of computation while maintaining the feature expressive power.

[0214] A feature pyramid is constructed to fuse features at different scales to detect targets of multiple sizes. At the same time, cross-scale residual connections are introduced into the feature pyramid for more efficient feature propagation. A transformation module (1x1 convolution) is designed between shallow and deep features to achieve efficient connections, enhance the expressive power of the feature pyramid, and improve the detection capability of small targets (such as the slight tilt of the detector).

[0215] A detection module is constructed, which uses a clustering algorithm to analyze the size of the target box in the data image, adjusts the anchor box mechanism to suit the state and size of the detector, and predicts the center point coordinates, width and height, confidence score, and class probability distribution of the output target box (such as the detector not being embedded, the detector being tilted, etc.).

[0216] S403: Model training and tuning. Based on the performance of the server GPU card used in this study, after repeated testing, the training set was uniformly set to 604*604 images, the learning rate was set to 0.001, the batch size was set to 8, and the epochs were set to 200 for model training. During the training process, the model will learn how to detect and locate the position of the detector from the image.

[0217] S404: Model Validation and Testing. Evaluate the model's detection performance on the validation set, calculate commonly used metrics such as mAP, and the precision and recall of various detection targets.

[0218] S405: Model testing. The model is tested on the test set to evaluate its performance on actual unseen data. Based on the test results, the model architecture and training parameters are repeatedly adjusted to finally obtain a YOLOv5 detector embedding status detection model with high recognition accuracy.

[0219] Step 5: Set the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system, as well as the sampling interval and other construction parameters for the construction area;

[0220] S501: Set seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area. These values ​​are provided by the National Center for Surveying and Mapping. The seven parameters include: translation distance in the x-direction, translation distance in the y-direction, translation distance in the z-direction, rotation angle in the x-direction, rotation angle in the y-direction, rotation angle in the z-direction, scale factor, and central meridian. Because this value is classified as state secret data, it has been kept confidential in this invention. Figure 3-1 As shown;

[0221] S502: Set construction parameters for the work area, including: sampling interval, acquisition station model, minimum resistance, maximum resistance, tilt angle range, detector type, number of detectors, and detector model, etc. Figure 3-2 As shown.

[0222] Step 6: Load the ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, and the detector burial construction task file, detector burial status daily inspection data file, and on-site photos of each detector burial status obtained from the above training.

[0223] The specific content includes:

[0224] S601: Load the ResNet-18 detector background classification network model trained in step 3;

[0225] S602: Load the YOLOv5 detector embedding status detection network model obtained from step 4;

[0226] S603: Load the geophone installation task file for the day. The task file includes information such as: serial number, line number, station number, east coordinate, north coordinate, geophone type, and number of geophones. Figure 4-1 As shown;

[0227] S604: Load the daily inspection data file of the geophone burial status for the day. The daily inspection file includes information such as: serial number, acquisition station type, line number, station number, resistance value, and tilt angle. Figure 4-2 As shown;

[0228] S605: Loads on-site photos of the burial status of each detector taken on the same day. The photo naming rule is: line number-point number-longitude-latitude-elevation. The photo shooting software takes photos with the following rules: the lower left corner of the photo should show the line number, point number, longitude, latitude, and elevation.

[0229] Step 7: Intelligently identify the location and status of the geophone burial, automatically analyze the connectivity and tilt of the geophone, and automatically output the evaluation results of the geophone burial quality.

[0230] Specifically, the following steps are included:

[0231] S701: Intelligently identifies the location of the detector and determines whether the detector is installed on time.

[0232] Specifically:

[0233] The system intelligently identifies the line number, station number, longitude, and latitude on the geophone burial photos and determines whether they match the file name of the geophone burial photos. If they match, the geophone burial photos are taken and named in accordance with regulations; otherwise, the geophone burial photos are taken and named in accordance with regulations. For geophone burial photos with inappropriate names, this method automatically outputs a list of geophone line numbers and station numbers, prompting geophone quality control personnel to check and make timely corrections.

[0234] Obtain the detector burial construction task file imported in step 6, and establish a one-to-one correspondence between the line numbers and point numbers intelligently identified in the above detector burial photos and the line numbers and point numbers in the detector burial construction task file.

[0235] Obtain the seven parameters for the conversion between the WGS84 coordinate system and the Beijing 54 coordinate system set in step 5, and convert the latitude and longitude coordinates in the WGS84 coordinate system identified in the above detector burial photo into the Beijing 54 coordinate system.

[0236] The difference between the theoretical design of the detector in the construction task file and the Beijing 54 coordinates (i.e., east and north coordinates) and the Beijing 54 coordinates obtained from the above detector installation photos is calculated. If the difference is less than 5m, the detector is considered to be installed on time. If it is greater than 5m, the detector is considered to have failed to be installed on time.

[0237] Automatic analysis and evaluation results of detector burial location, such as Figure 5 As shown.

[0238] S702, intelligent identification of detector embedding status;

[0239] Specifically:

[0240] The ResNet-18 detector background classification network model is used to intelligently identify whether the construction background of the detector burial image is farmland or hardened road surface.

[0241] If the detector is in a field background as identified above, the detector burial quality inspection standard in the field background will be activated, and the YOLOv5 detector burial status detection network model will be used to identify the detector burial status and automatically provide the detector burial status identification result.

[0242] If the above identification result is a hardened road surface background, then the detector embedding quality inspection standard under hardened road surface background is activated, and the YOLOv5 detector embedding status detection network model is used to identify the detector embedding status and automatically give the detector embedding status identification result.

[0243] There are two criteria for judging the quality of geophone burial in field background. If either one is met, it is considered qualified: First, the tail cone of the geophone is vertically inserted into the ground for 2 / 3 of its length; second, the geophone is buried underground and covered with a thin layer of soil, and the acquisition station is near the geophone.

[0244] There are four criteria for judging the quality of geophone burial in field background. If any one of them is met, it is considered unqualified: First, the geophone tail cone is not inserted vertically into the ground for less than 2 / 3; second, the geophone is laid horizontally on the ground; third, the geophone tail cone is vertically upward; fourth, there is no acquisition station near the geophone burial pit.

[0245] The criteria for judging the quality of geophone burial under hardened road surface background is: the geophone tail cone is removed and firmly glued to the ground with adhesive;

[0246] There are six criteria for judging the quality of geophone installation in hardened road surfaces as unqualified. Meeting any one of these criteria constitutes a failure: 1. The geophone tail cone is not removed and is inserted above a sewer; 2. The geophone tail cone is not removed and is placed horizontally on the ground; 3. The geophone tail cone is not removed and is inverted vertically upwards; 4. The geophone tail cone is not removed and is inserted obliquely in a corner; 5. The geophone tail cone has been removed but not glued to the ground; 6. The geophone tail cone has been removed and was initially glued to the ground, but later the geophone detached or tipped over.

[0247] In the process of detector burial status identification, the detector burial status detection model predicts the bounding box position, confidence level, and class probability of each feature point. Each feature point can predict multiple bounding boxes. During the prediction process, multiple candidate bounding boxes are generated, which may include overlapping boxes. In order to reduce the number of redundant boxes, the non-maximum suppression (NMS) method is used to retain the box with the highest confidence level and filter out other boxes with large overlap (the threshold is set to 0.4 in the method of this invention). Finally, a detector burial status identification result with high accuracy is obtained.

[0248] Figure 6 This is a schematic diagram showing the analysis results of the qualified and unqualified detector burial status intelligently identified under two backgrounds: farmland and hardened road surface, as mentioned in this invention.

[0249] Figure 6-1 A schematic diagram of a geophone image showing a qualified geophone burial status for intelligent identification against a field background.

[0250] Figure 6-2 A schematic diagram of a detector with an unqualified burial status identified by intelligent identification against a field background;

[0251] Figure 6-3 A schematic diagram of a geophone image showing a qualified geophone embedding status for intelligent identification against a hardened road surface background.

[0252] Figure 6-4 This is a schematic diagram of a detector with an unqualified embedding status, identified by intelligent identification against a hardened road surface background.

[0253] S703. Automatically analyze whether the detector connectivity and detector tilt angle are qualified.

[0254] Specifically:

[0255] Obtain the threshold range of detector resistance and detector tilt angle for the work area set in step 5.

[0256] The system sequentially acquires the resistance and tilt angle values ​​of each detector in the daily instrument inspection report. If the resistance or tilt angle value of a detector is outside the predetermined threshold range, the system automatically determines that the detector is not connected or that its tilt angle is unqualified. Figure 7 As shown;

[0257] S704. Classify the analysis results obtained from S701, S702, and S703 according to the detector line number and point number, and output the analysis and evaluation results regarding whether the detector's burial status is qualified. Figure 8 As shown.

[0258] Example 2

[0259] In addition to including all the steps in Embodiment 1 above, this embodiment also includes the following steps:

[0260] After each construction zone is completed, the sample dataset is updated to obtain the updated detector embedding image dataset.

[0261] The ResNet-18 detector background classification network model and the YOLOv5 detector embedding status detection model were corrected based on the updated detector embedding image dataset.

[0262] Based on the modified ResNet-18 detector background classification network model and the YOLOv5 detector burial status detection model, the burial quality of seismic acquisition detectors in the new work area is automatically analyzed and evaluated. By continuously refining the detector image background detection model and the detector burial quality detection model, the model accuracy is improved, thereby enhancing the accuracy of the detector burial quality evaluation.

[0263] As can be seen from Examples 1 and 2, the main advantages of the present invention compared to the prior art are:

[0264] (1) Using artificial intelligence technology to replace the traditional manual verification method can improve the efficiency of geophone installation quality inspection, reduce labor and time costs, reduce reliance on the work experience of quality control personnel, improve the quality of geophone installation quality inspection, and thus improve the quality of seismic acquisition data.

[0265] (2) A large number of images of different geophone burial scenarios in the eastern exploration area were collected and the images were preprocessed to construct a diverse geophone burial status dataset, ensuring the accuracy of subsequent model training and improving the intelligent detection accuracy of geophone burial quality.

[0266] (3) Parameters of ResNet-18 network model and YOLOv5 network model were optimized to make them more suitable for different detector burial scenarios in the eastern exploration area;

[0267] (4) The reliability of intelligent detection results of detector burial quality is improved by jointly applying the deep learning models of ResNet-18 and YOLOv5. First, the ResNet-18 network model is used to automatically classify the detector burial photos according to the field construction background. Then, the detector burial photos after background classification are used to automatically detect the detector burial status. This method can effectively reduce the interference of environmental background on the detection of detector burial status, thereby improving the accuracy of detector burial quality detection.

[0268] (5) By combining artificial intelligence technology with daily inspection data files of detector burial status to comprehensively analyze the quality of detector burial, the probability of identification errors that may occur when only artificial intelligence technology is used is further reduced.

[0269] (6) A set of intelligent analysis system for detector embedding quality was studied, which realizes full-process automation and visualization from the import of detector embedding photos into the system to the output of intelligent detection results, and then realizes industrial application in field production;

[0270] (7) After each construction area is completed, the sample dataset is updated in a timely manner. Based on the updated detector burial image sample dataset, the ResNet-18 detector background classification network model and the YOLOv5 detector burial status detection model are continuously corrected to continuously improve the model accuracy and thus continuously improve the accuracy of detector burial quality evaluation.

[0271] Example 3

[0272] Figure 9 This is a structural diagram of the intelligent system for the installation quality of the seismic acquisition detector of this invention. Figure 9 As shown, an intelligent grading and evaluation system for the quality of seismic acquisition data includes:

[0273] Dataset construction module 1001 is used to construct a detector image sample dataset;

[0274] The dataset sample annotation module 1002 is used for creating sample labels for the detector image sample dataset.

[0275] The detector background classification network model building module 1003 is used to train the detector image sample dataset to obtain a ResNet-18 detector background classification network model suitable for the eastern exploration area.

[0276] The detector status detection network model construction module 1004 is used to train the detector image sample dataset to obtain a YOLOv5 detector burial status detection network model suitable for the eastern exploration area.

[0277] The basic parameter setting module 1005 for the work area is used to set the seven coordinate transformation parameters and sampling interval of the construction work area.

[0278] The input module 1006, which includes network models and detector installation tasks, is used to load the trained ResNet-18 detector background classification network model and YOLOv5 detector installation status detection network model, detector installation photos to be analyzed and evaluated, detector installation task files, and daily inspection data files of detector installation status, providing a data foundation for the subsequent detector installation quality evaluation module.

[0279] The intelligent analysis and evaluation module 1007 for geophone burial quality is used to automatically analyze the seismic acquisition geophone images in the construction area based on the ResNet-18 geophone background classification network model and the YOLOv5 geophone burial status detection model, and output the evaluation result of whether the geophone burial quality is qualified.

[0280] The dataset construction module 1001 specifically includes:

[0281] The raw dataset construction unit is used to construct the raw detector image sample dataset;

[0282] The image denoising preprocessing unit is used to perform median filtering denoising on the original detector image sample dataset;

[0283] The data augmentation preprocessing unit is used to perform data augmentation processing on the denoised image;

[0284] The image enhancement preprocessing unit is used to perform image enhancement processing on the data after data enhancement.

[0285] The detector image sample dataset construction unit is used to construct the detector image sample dataset.

[0286] The dataset sample annotation module 1002 specifically includes:

[0287] The sample label category definition unit is used to define the marker feature category of whether the detector to be tested is qualified or unqualified.

[0288] Sample labeling unit, used for creating sample labels for detector image sample datasets;

[0289] The sample label file output unit is used to output the labeled sample label file.

[0290] The detector background detection model training module 1003 specifically includes:

[0291] The detector image sample training set, validation set, and test set generation unit is used to load the detector image sample dataset and divide it according to an 8:1:1 ratio to form the training set, validation set, and test set.

[0292] The training set image preprocessing unit is used to uniformly scale the images to a fixed size and normalize the pixel values ​​of the images.

[0293] A detector background classification network building unit is used to construct a ResNet-18 detector background classification network suitable for the eastern exploration area.

[0294] The detector background classification network model training unit is used to train the detector image sample dataset to obtain a ResNet-18 detector background classification network model suitable for the eastern exploration area.

[0295] The detector background classification network model output unit is used to output the ResNet-18 detector background classification network model file obtained above and suitable for the eastern exploration area.

[0296] The detector state detection network model training module 1004 specifically includes:

[0297] The tag file loading unit is used to load the tag file output by module 1002;

[0298] A detector condition monitoring network construction unit is used to construct a YOLOv5 detector condition monitoring network suitable for the eastern exploration area.

[0299] The model training and tuning unit is used for preprocessing training set data and optimizing model parameters.

[0300] The model validation and testing unit is used to evaluate the detection performance of the model on the validation set, and adjust the model architecture or training parameters through human-computer interaction based on the test results.

[0301] The target detection unit is used to intelligently identify the buried status of the detector. When the identification accuracy reaches the expected target, it is determined as the final detector status detection network model.

[0302] The YOLOv5 detector state detection network model output unit is used to output the YOLOv5 detector state detection network model file obtained from the above training, which is suitable for the eastern exploration area.

[0303] The work area basic parameter setting module 1005 specifically includes:

[0304] The coordinate transformation seven-parameter setting unit for the construction area is used to set the seven parameters for the transformation from the WGS84 coordinate system to the Beijing 54 coordinate system of the construction area.

[0305] The basic construction parameter setting unit for the construction area, such as the sampling interval, is used to set the basic construction parameters for the construction area.

[0306] The network model and detector installation task file input module 1006 specifically includes:

[0307] The ResNet-18 detector background classification network model loading unit is used to load the ResNet-18 detector background classification network model.

[0308] YOLOv5 detector embedding status detection network model loading unit, used to load the YOLOv5 detector embedding status detection network model;

[0309] The detector embedding photo loading unit is used to batch load the detector embedding photos to be analyzed and evaluated.

[0310] The detector embedding construction task file loading unit is used to load the detector embedding construction task file;

[0311] The detector embedding status daily inspection data file loading unit is used to load the detector embedding status daily inspection data file.

[0312] The detector embedding quality evaluation module 1007 specifically includes:

[0313] The detector embedding location intelligent detection unit is used to automatically identify the latitude and longitude coordinates in the detector image;

[0314] The coordinate transformation unit is used to convert the WGS84 coordinates identified in the detector image into Beijing 54 coordinates.

[0315] The on-time deployment judgment unit is used to calculate the difference between the Beijing 54 coordinates of the detector image and the theoretical coordinates of the detector in the construction task file, and to give the judgment result of whether the deployment is on time.

[0316] The intelligent detection unit for detector embedding quality is used to automatically identify whether the detector embedding quality in the detector image is up to standard and to provide analysis and evaluation results.

[0317] The detector resistance analysis and evaluation unit is used to analyze whether the detector is connected and to give the evaluation result.

[0318] The detector tilt angle analysis and evaluation unit is used to analyze whether the detector is vertical and to give the evaluation result;

[0319] The detector embedding quality evaluation result output unit is used to summarize and output the detector line number, point number, and reasons for the non-compliance of the embedding quality of the detectors, providing data for field rectification.

[0320] Specific examples have been used to illustrate the principles and implementation methods of this invention. The descriptions of the above embodiments are only for the purpose of helping to understand the method and core ideas of this invention. Furthermore, those skilled in the art will recognize that, based on the ideas of this invention, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this invention.

[0321] Except for the technical features described in the specification, all other technologies are known to those skilled in the art.

Claims

1. A method for intelligent analysis and evaluation of detector embedding quality, characterized in that, The intelligent analysis and evaluation method for the embedding quality of the detector includes: Step 1: Collect photos of the field geophone installation sites and perform image preprocessing to construct a geophone image sample dataset; Step 2: Label the detector image sample dataset and create a label file; Step 3: Train the model and fine-tune the parameters on the detector image sample dataset to build a detector background classification network model; Step 4: Train the network model and fine-tune the parameters of the detector image sample dataset and label file to build a network model for detecting the detector's burial status. Step 5: Identify the location and status of the detector installation, and analyze the connectivity and tilt angle of the detector.

2. The intelligent analysis and evaluation method for detector embedding quality according to claim 1, characterized in that, In step 1, images of the geophones buried at the construction site are collected and preprocessed, including denoising, data augmentation, and image enhancement, to construct a geophone image sample dataset.

3. The intelligent analysis and evaluation method for detector embedding quality according to claim 1, characterized in that, Step 1 includes: S101. Perform median filtering on the acquired image to remove noise interference and improve image smoothness; S102. Perform data augmentation processing on the image after denoising in step S101 to reduce the occurrence of data sample imbalance. S103. The image after data enhancement in step S102 is improved by histogram equalization image enhancement technology to improve the image quality. S104. Merge the images after data preprocessing in the above three steps S101, S102 and S103 with the original detector embedding images to construct an initial detector image sample dataset. S105. The initial detector image sample dataset is classified according to the detector burial background, into a detector image sample dataset with a field background and a detector image sample dataset with a hardened road background, collectively referred to as the detector image sample dataset, which is used for training the subsequent detector background classification network model and detector burial status detection network model.

4. The intelligent analysis and evaluation method for detector embedding quality according to claim 1, characterized in that, Step 2 includes: S201. Based on the characteristics of the detector's burial, define image label categories for two background backgrounds: farmland and hardened road surface. Against a field background, the image label categories are set as follows: front-facing detector, back-facing detector, tilted detector, detector tail cone, and acquisition station (5 categories); against a hardened road background, the image label categories are set as follows: front-facing detector, back-facing detector, tilted detector, detector tail cone, acquisition station, and adhesive (6 categories). S202. Label the samples under two different backgrounds: farmland and paved road surface, and output the label files respectively. The tag file format includes: image width, height, depth, target category name, and target location coordinates.

5. The intelligent analysis and evaluation method for detector embedding quality according to claim 1, characterized in that, In step 3, the detector image sample dataset formed in step 1 is loaded, and the ResNet-18 network model is trained and its parameters are tuned to construct the ResNet-18 detector background classification network model. The ResNet-18 network structure is used to refine the scene classification problem into a typical binary classification problem. The gradient vanishing problem in deep networks is solved by introducing residual learning units. The model has 18 layers of convolutional and residual blocks and 3 layers of fully connected parts, for a total of 21 layers.

6. The intelligent analysis and evaluation method for detector embedding quality according to claim 5, characterized in that, Step 3 specifically includes: S301: Load the detector image sample dataset formed in step 1, and divide it into training set, validation set and test set according to 8:1:1; S302: To improve the training efficiency of the background classification network model, the training set images in the detector image sample dataset are uniformly scaled to a fixed size, and the pixel values ​​of the images are normalized. S303: Construct a ResNet-18 detector background classification network; S304: Perform model training and parameter tuning to obtain the ResNet-18 detector background classification network model.

7. The intelligent analysis and evaluation method for detector embedding quality according to claim 6, characterized in that, In step 4, the YOLOv5 network model is trained and its parameters are tuned on the detector image sample dataset and label file to construct the YOLOv5 detector embedding status detection network model.

8. The intelligent analysis and evaluation method for detector embedding quality according to claim 7, characterized in that, Step 4 includes: S401: Load the detector image sample dataset constructed in step 1 and the label file formed in step 2, and divide the dataset into training set, validation set and test set according to 8:1:1; S402: Construct the YOLOv5 detector embedding status detection network. The construction method is as follows: A YOLOv5 backbone network is constructed to extract multi-scale features from images. A cross-stage partial residual structure is built to reduce computation while maintaining feature representation ability. A feature pyramid is constructed to fuse features at different scales to detect targets of multiple sizes. At the same time, cross-scale residual connections are introduced into the feature pyramid for more efficient feature propagation. A transformation module between shallow and deep features is designed to achieve efficient connection, enhance the expressive power of the feature pyramid, and improve the detection capability of small targets. A detection module is constructed, which uses a clustering algorithm to analyze the size of the target boxes in the data image, adjusts the anchor box mechanism to suit the detector state and size, and predicts the center point coordinates, width and height, confidence score, and class probability distribution of the output target boxes. S403: Model training and tuning. Based on the performance of the server GPU card, after repeated testing, the training set is uniformly sized before model training. During the training process, the model learns how to detect and locate the detector in the image. S404: Model Validation and Testing. Evaluate the model's detection performance on the validation set, calculate commonly used metrics such as mAP, and the precision and recall of various detection targets. S405: Model testing. The model is tested on the test set to evaluate its potential performance on real data. Based on the test results, the model architecture and training parameters are repeatedly adjusted to finally obtain a YOLOv5 detector embedding status detection model with high recognition accuracy.

9. The intelligent analysis and evaluation method for detector embedding quality according to claim 7, characterized in that, The intelligent analysis and evaluation method for detector burial quality also includes, after step 4, setting the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area, as well as the sampling interval and other construction parameters; loading the trained ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, the daily detector burial construction task file, the daily detector burial status inspection data file, and the on-site photos of each detector burial condition.

10. The intelligent analysis and evaluation method for detector embedding quality according to claim 9, characterized in that, After step 4, when setting the seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system, as well as the sampling interval, the specific steps include: S501: Set seven parameters for the transformation between the WGS84 coordinate system and the Beijing 54 coordinate system of the construction area. The seven parameters include: translation distance in the x direction, translation distance in the y direction, translation distance in the z direction, rotation angle in the x direction, rotation angle in the y direction, rotation angle in the z direction, scale factor, and central meridian. S502: Set the construction parameters for the work area, including: sampling interval, acquisition station model, minimum resistance, maximum resistance, tilt range, detector type, and the number and model of detectors.

11. The intelligent analysis and evaluation method for detector embedding quality according to claim 9, characterized in that, After step 4, load the trained ResNet-18 detector background classification network model, YOLOv5 detector burial status detection network model, the daily detector burial construction task file, the daily detector burial status inspection data file, and the on-site photos of each detector burial status, specifically including: S601: Load the ResNet-18 detector background classification network model trained in step 3; S602: Load the YOLOv5 detector embedding status detection network model obtained from step 4; S603: Load the geophone installation task file for the day. The task file includes: serial number, line number, station number, east coordinate, north coordinate, geophone type, and geophone quantity information. S604: Load the daily inspection data file of the geophone burial status for the day. The daily inspection file includes: serial number, acquisition station type, line number, station number, resistance value, and tilt angle information. S605: Loads on-site photos of the burial status of each detector taken on the same day. The photo naming rule is: line number-point number-longitude-latitude-elevation. The photo shooting software takes photos with the following rules: the lower left corner of the photo should show the line number, point number, longitude, latitude, and elevation.

12. The intelligent analysis and evaluation method for detector embedding quality according to claim 1, characterized in that, Step 5 identifies the detector's burial location and burial status, analyzes the detector's connectivity and tilt angle, and outputs the detector burial quality evaluation results; specifically including: S701. Identify the location of the geophone burial site and determine if the geophone is installed on the correct location; intelligently identify the line number, station number, longitude, and latitude on the geophone burial photo and determine if they match the file name of the geophone burial photo. If they match, the geophone burial photo is taken and named in accordance with regulations; if not, the geophone burial photo is taken and named in accordance with regulations. For geophone burial photos with inappropriate names, output a list of geophone line numbers and station numbers, prompting the geophone quality control personnel to check and modify them in a timely manner. Obtain the geophone burial construction task file and establish a one-to-one correspondence between the line numbers and point numbers intelligently identified in the geophone burial photos and the line numbers and point numbers in the geophone burial construction task file. Obtain the seven parameters for the conversion between the WGS84 coordinate system and the Beijing 54 coordinate system, and convert the latitude and longitude coordinates in the WGS84 coordinate system identified in the detector embedding photo into the Beijing 54 coordinate system. The difference between the theoretical design coordinates of the detector in the construction task file and the Beijing 54 coordinates obtained from the detector installation photo is calculated. If the difference is less than the preset value, the detector is considered to be installed on time. If the difference is greater than the preset value, the detector is considered to be installed on time. S702. Identify the burial status of the detector; use the ResNet-18 detector background classification network model to intelligently identify whether the construction background of the detector burial image is farmland or hardened road surface. If the detector is in a field background as identified above, then the detector burial quality inspection standard in the field background is activated, and the YOLOv5 detector burial status detection network model is used to identify the detector burial status and give the detector burial status identification result. If the above identification result is a hardened road surface background, then the detector embedding quality inspection standard under hardened road surface background is activated, and the YOLOv5 detector embedding status detection network model is used to identify the detector embedding status and give the detector embedding status identification result. There are two criteria for judging the quality of geophone burial in field background. If either one is met, it is considered qualified: First, the geophone tail cone is vertically inserted below the ground at a preset ratio; second, the geophone is buried underground and covered with a thin layer of soil, and the acquisition station is near the geophone. There are four criteria for judging the quality of geophone burial in field background. If any one of them is met, it is considered unqualified: First, the geophone tail cone is not vertically inserted into the ground at the preset ratio; second, the geophone is placed horizontally on the ground; third, the geophone tail cone is vertically upward; fourth, there is no acquisition station near the geophone burial pit. The criteria for judging the quality of geophone burial under hardened road surface background is: the geophone tail cone is removed and firmly glued to the ground with adhesive; There are six criteria for judging the quality of geophone installation in hardened road surfaces as unqualified. Meeting any one of these criteria constitutes a failure:

1. The geophone tail cone is not removed and is inserted above a sewer; 2. The geophone tail cone is not removed and is placed horizontally on the ground; 3. The geophone tail cone is not removed and is inverted vertically upwards; 4. The geophone tail cone is not removed and is inserted obliquely in a corner; 5. The geophone tail cone has been removed but not glued to the ground; 6. The geophone tail cone has been removed and was initially glued to the ground, but later the geophone detached or tipped over. In the process of detector burial status identification, the detector burial status detection model predicts the bounding box position, confidence level, and class probability of each feature point. Each feature point can predict multiple bounding boxes. During the prediction process, multiple candidate bounding boxes are generated, which may include overlapping boxes. In order to reduce the number of redundant boxes, the non-maximum suppression method is used to retain the box with the highest confidence level and filter out other boxes with large overlap with it, and finally obtain a detector burial status identification result with high accuracy. S703. Analyze whether the detector connection status and detector tilt angle status are qualified. Obtain the set threshold range for the resistance value and the threshold range for the detector tilt angle in the work area; Sequentially acquire the resistance and tilt angle status values ​​of each detector in the daily inspection file of the instrument. If the resistance or tilt angle status of the detector is not within the predetermined threshold range, the detector is determined to be in a non-connected state or the tilt angle status of the detector is unqualified. S704. Classify the analysis results obtained from S701, S702 and S703 according to the detector line number and point number, and output the analysis and evaluation results of whether the detector burial status is qualified.

13. A smart analysis and evaluation system for detector embedding quality, characterized in that, The intelligent analysis and evaluation system for detector embedding quality includes: The dataset construction module builds a sample dataset of detector images. The dataset sample annotation module is used to create sample labels for detector image sample datasets; The detector background classification network model building module trains the detector image sample dataset to obtain the detector background classification network model. The detector state detection network model construction module trains the detector image sample dataset to obtain the detector embedding state detection network model; The basic parameter setting module for the work area allows you to set the seven coordinate transformation parameters and sampling interval for the construction work area. The network model and detector burial construction task file input module loads the trained detector background classification network model and detector burial status detection network model, the detector burial photos to be analyzed and evaluated, the detector burial construction task file, and the detector burial status daily inspection data file, providing a data foundation for the subsequent detector burial quality evaluation module. The intelligent analysis and evaluation module for geophone burial quality analyzes the seismic acquisition geophone images in the construction area based on the geophone background classification network model and the geophone burial status detection model, and outputs the evaluation results of whether the geophone burial quality is qualified.

14. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The dataset construction module includes: The raw dataset construction unit is used to construct the raw detector image sample dataset; The image denoising preprocessing unit is used to perform median filtering denoising on the original detector image sample dataset; The data augmentation preprocessing unit is used to perform data augmentation processing on the denoised image; The image enhancement preprocessing unit is used to perform image enhancement processing on the data after data enhancement. The detector image sample dataset construction unit is used to construct the detector image sample dataset.

15. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The dataset sample annotation module includes: The sample label category definition unit is used to define the marker feature category of whether the detector to be tested is qualified or unqualified. Sample labeling unit, used for creating sample labels for detector image sample datasets; The sample label file output unit is used to output the labeled sample label file.

16. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The detector background detection model training module includes: The detector image sample training set, validation set, and test set generation unit is used to load the detector image sample dataset and divide it according to an 8:1:1 ratio to form the training set, validation set, and test set. The training set image preprocessing unit is used to uniformly scale the images to a fixed size and normalize the pixel values ​​of the images. Detector background classification network building unit, used to build ResNet-18 detector background classification network; The detector background classification network model training unit is used to train the detector image sample dataset to obtain the ResNet-18 detector background classification network model. The detector background classification network model output unit is used to output the ResNet-18 detector background classification network model file obtained from the above training.

17. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The detector state detection network model training module includes: The label file loading unit is used to load the label file output by the dataset sample annotation module; Detector status detection network construction unit, used to construct YOLOv5 detector status detection network; The model training and tuning unit is used for preprocessing training set data and optimizing model parameters. The model validation and testing unit is used to evaluate the detection performance of the model on the validation set, and adjust the model architecture or training parameters through human-computer interaction based on the test results. The target detection unit is used to intelligently identify the buried status of the detector. When the identification accuracy reaches the expected target, it is determined as the final detector status detection network model. The YOLOv5 detector state detection network model output unit is used to output the YOLOv5 detector state detection network model file obtained from the above training.

18. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The basic parameter setting module for the work area includes: The coordinate transformation seven-parameter setting unit for the construction area is used to set the seven parameters for the transformation from the WGS84 coordinate system to the Beijing 54 coordinate system of the construction area. The sampling interval basic construction parameter setting unit for the construction area is used to set the basic construction parameters such as the sampling interval of the construction area.

19. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The network model and detector embedding construction task file input module includes: The ResNet-18 detector background classification network model loading unit is used to load the ResNet-18 detector background classification network model. YOLOv5 detector embedding status detection network model loading unit, used to load the YOLOv5 detector embedding status detection network model; The detector embedding photo loading unit is used to batch load the detector embedding photos to be analyzed and evaluated. The detector embedding construction task file loading unit is used to load the detector embedding construction task file; The detector embedding status daily inspection data file loading unit is used to load the detector embedding status daily inspection data file.

20. The intelligent analysis and evaluation system for detector embedding quality according to claim 13, characterized in that, The detector embedding quality evaluation module includes: The detector embedding location intelligent detection unit is used to identify the latitude and longitude coordinates in the detector image; The coordinate transformation unit is used to convert the WGS84 coordinates identified in the detector image into Beijing 54 coordinates. The on-time deployment judgment unit is used to calculate the difference between the Beijing 54 coordinates of the detector image and the theoretical coordinates of the detector in the construction task file, and to give the judgment result of whether the deployment is on time. The intelligent detection unit for detector embedding quality is used to identify whether the detector embedding quality in the detector image is up to standard and to provide analysis and evaluation results. The detector resistance analysis and evaluation unit is used to analyze whether the detector is connected and to give the evaluation result. The detector tilt angle analysis and evaluation unit is used to analyze whether the detector is vertical and to give the evaluation result; The detector embedding quality evaluation result output unit is used to summarize and output the detector line number, point number, and reasons for the non-compliance of the embedding quality of the detectors, providing data for field rectification.

Images