Oil and gas pipeline safety inspection and positioning method

Abstract

The invention relates to an oil and gas pipeline safety inspection and positioning method, which comprises the following steps: acquiring image data of the surface and surrounding environment of an oil and gas pipeline, and constructing a data set; constructing an original detection model by using a convolutional neural network and a feature pyramid network; training the original detection model by using the data set to obtain a final oil and gas pipeline safety detection model; and performing real-time detection and potential safety hazard positioning on the oil and gas pipeline by using the oil and gas pipeline safety detection model. The system and the method have remarkable beneficial effects in the aspects of improving the inspection efficiency and accuracy, realizing real-time monitoring and early warning, improving the emergency response capability, reducing the manual inspection risk, intelligently managing and optimizing and the like.

Description

Technical Field

[0001] The present invention relates to the technical field of oil and gas pipeline monitoring, and in particular to a safety inspection and positioning method for oil and gas pipelines. Background Art

[0002] As an important means of energy transportation, the safety of oil and gas pipelines is directly related to the safety of people's lives and property. The traditional inspection method of oil and gas pipelines mainly relies on manual observation on foot, which has the disadvantages of heavy workload, low efficiency, long time consumption, and susceptibility to subjective factors. In addition, with the continuous expansion and complexity of pipeline networks, traditional inspection methods can no longer meet the growing safety monitoring needs.

[0003] In recent years, the rapid development of drone technology and deep learning technology has provided new solutions for oil and gas pipeline safety inspections. However, most existing drone inspection solutions rely on manual observation and analysis of captured images, which have problems such as poor real-time performance, high missed detection rate, and low processing efficiency. Therefore, it is of great practical significance and application value to develop a method that can automatically analyze images, identify and locate safety hazards of oil and gas pipelines in real time. Summary of the invention

[0004] The purpose of the present invention is to solve the problems existing in the above-mentioned prior art, to automatically analyze the images of the pipeline periphery taken by the drone, to identify the behaviors endangering the safety of the pipeline in real time, and to accurately locate the hidden dangers, thereby improving the inspection efficiency and accuracy.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for safety inspection and positioning of oil and gas pipelines, comprising:

[0007] Collect image data of the oil and gas pipeline surface and surrounding environment to build a data set;

[0008] Construct the original detection model using convolutional neural network and feature pyramid network;

[0009] Using the data set to train the original detection model to obtain a final oil and gas pipeline safety detection model;

[0010] The oil and gas pipeline safety detection model is used to perform real-time detection of the oil and gas pipeline and locate safety hazards.

[0011] Optionally, constructing a dataset includes:

[0012] Use drones equipped with image acquisition devices to conduct regular or irregular inspections and photography around oil and gas pipelines;

[0013] Preprocess the collected image data, including denoising, contrast enhancement, cropping, and data enhancement;

[0014] The potential safety hazards in the preprocessed image data are manually annotated to generate the data set for model training.

[0015] Optionally, the original detection model includes: a feature extraction module, a target detection module, a positioning calculation module and a result fusion module;

[0016] The feature extraction module uses a convolutional neural network to perform deep feature extraction on the input image and generate a high-level feature map;

[0017] The target detection module performs target recognition on the feature map and outputs detection information; wherein the detection information includes: the category, location and confidence level of the potential safety hazard;

[0018] The positioning calculation module performs coordinate conversion and calculation based on the output result of the target detection module and the GPS coordinates of the drone to obtain the location of the potential safety hazard on the map;

[0019] The result fusion module is used to fuse the target detection result and the positioning result to generate an inspection report; wherein the inspection report includes: hidden danger type, location, and timestamp information.

[0020] Optionally, an attention mechanism is added to the feature extraction module to dynamically adjust the network's attention to different areas in the input image.

[0021] Optionally, the target detection module adopts a feature pyramid network structure, selects a MobileNet network as a backbone network, and reduces the number of feature layers according to task requirements.

[0022] Optionally, the location calculation module obtains the location of the potential safety hazard on the map, including:

[0023] Use the camera's intrinsic and extrinsic matrix to transform the pixel coordinates in the image coordinate system into the camera coordinate system;

[0024] The coordinates in the camera coordinate system are converted into the geographic coordinate system by combining the GPS coordinates and attitude information of the drone to obtain the location of the safety hazard on the map.

[0025] Optionally, using the data set to train the original detection model includes:

[0026] Use the labeled data set to train the model and continuously adjust the model parameters through the back-propagation algorithm;

[0027] During the training process, cross-validation is used to avoid overfitting, and the training effect is optimized by adjusting the learning rate and batch size hyperparameters.

[0028] Optionally, using the oil and gas pipeline safety detection model to perform real-time detection of the oil and gas pipeline and locate potential safety hazards includes:

[0029] Real-time collection of image data of the oil and gas pipeline surface and surrounding environment;

[0030] The image data collected in real time is input into the oil and gas pipeline safety detection model to obtain an inspection report including hidden danger type, location, and timestamp information.

[0031] The beneficial effects of the present invention are:

[0032] The present invention first collects image data of the surface of the oil and gas pipeline and the surrounding environment to construct a data set; then uses the data set to train the original detection model constructed based on the convolutional neural network and the feature pyramid network to obtain the oil and gas pipeline safety detection model; finally, uses the oil and gas pipeline safety detection model to perform real-time detection and safety hazard location on the oil and gas pipeline. The present invention can automatically analyze the pipeline surrounding images taken by the drone, identify behaviors that endanger pipeline safety in real time, and accurately locate the hidden dangers, thereby improving the inspection efficiency and accuracy; reducing the workload and risks of the inspection personnel, especially in harsh environments or dangerous areas, and avoiding the safety risks that may be caused by manual inspections. Through real-time monitoring and early warning, potential safety hazards can be discovered and dealt with in advance, improving the overall safety of the oil and gas pipeline. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0034] Figure 1 The present invention is a flowchart of a method for oil and gas pipeline safety inspection and positioning according to an embodiment of the present invention. DETAILED DESCRIPTION

[0035] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0036] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0037] like Figure 1 As shown, this embodiment proposes a method for safety inspection and positioning of oil and gas pipelines, including:

[0038] Collect image data of the oil and gas pipeline surface and surrounding environment to build a data set;

[0039] Construct the original detection model using convolutional neural network and feature pyramid network;

[0040] Use the data set to train the original detection model and obtain the final oil and gas pipeline safety detection model;

[0041] The oil and gas pipeline safety detection model is used to conduct real-time detection of oil and gas pipelines and locate safety hazards.

[0042] Furthermore, constructing the dataset includes:

[0043] Use drones equipped with image acquisition devices to conduct regular or irregular inspections and photography around oil and gas pipelines;

[0044] Preprocess the collected image data, including denoising, contrast enhancement, cropping, and data enhancement. Data enhancement is to increase the diversity and quantity of training data by rotating, scaling, cropping, adding noise, and other operations on the original image, thereby improving the generalization ability of the model.

[0045] Manually annotate the safety hazards in the preprocessed image data, including the location, category and other information of the safety hazards, to generate a data set for model training.

[0046] This embodiment uses a drone equipped with a high-resolution camera to photograph the vicinity of the oil and gas pipeline to obtain a large amount of image data. The image data is preprocessed, including denoising, contrast enhancement, cropping, etc., to improve the efficiency and accuracy of subsequent processing.

[0047] Furthermore, the original detection model includes: a feature extraction module, a target detection module, a positioning calculation module and a result fusion module;

[0048] The feature extraction module uses a convolutional neural network to perform deep feature extraction on the input image and generate a high-level feature map;

[0049] The target detection module performs target recognition on the feature map and outputs detection information; the detection information includes: the category, location and confidence level of the safety hazard;

[0050] The positioning calculation module performs coordinate conversion and calculation based on the output results of the target detection module and the GPS coordinates of the drone to obtain the location of the safety hazard on the map;

[0051] The result fusion module is used to fuse the target detection results and the positioning results to generate an inspection report; wherein the inspection report includes: hidden danger type, location, and timestamp information.

[0052] Specifically, the deep learning model proposed in this embodiment adopts a multi-stage architecture, including a feature extraction module, a target detection module, a positioning calculation module, and a result fusion module;

[0053] The feature extraction module uses convolutional neural network (CNN) to perform deep feature extraction on the input image and generate a high-level feature map.

[0054] The target detection module performs target recognition based on the feature map and outputs information such as the category, location (bounding box) and confidence level of the safety hazard.

[0055] The positioning calculation module performs coordinate conversion and calculation based on the output results of the target detection module and the GPS coordinates of the drone to obtain the precise location of the safety hazard on the map.

[0056] The result fusion module fuses the target detection results and positioning results to generate an inspection report containing hidden danger type, location, timestamp and other information.

[0057] Furthermore, an attention mechanism is added to the feature extraction module to dynamically adjust the network's attention to different areas in the input image, and depthwise separable convolution is used instead of traditional standard convolution.

[0058] Specifically, in this embodiment, the feature extraction module is improved as follows:

[0059] Introducing attention mechanism: Based on the convolutional neural network (CNN), attention mechanism modules (such as SENet, CBAM, etc.) are added. These mechanisms can dynamically adjust the network's attention to different areas in the input image, making the model more focused on learning the characteristics of safety hazards and improving the accuracy and efficiency of feature extraction.

[0060] Depthwise separable convolution: Using depthwise separable convolution instead of traditional standard convolution reduces the number of model parameters and computation while maintaining or improving the performance of feature extraction. This helps deploy models on edge computing devices and improves the real-time performance of inspections.

[0061] Taking the addition of SENet as an example, the improved feature extraction module includes:

[0062] Basic convolutional blocks: First, the feature extraction module consists of multiple convolutional blocks, each of which usually contains multiple convolutional layers, activation functions (such as ReLU), and possible pooling layers;

[0063] SE module insertion: After each convolutional block, a SE module is inserted to recalibrate the importance of channel features.

[0064] Squeeze operation: Perform global average pooling on the feature map output by the convolutional block to compress the feature map of each channel into a real number, which is the global feature description of the channel. The purpose of this step is to embed global spatial information into the channel descriptor.

[0065] Excitation operation: The weight of each channel is learned through two fully connected layers (the first FC layer is followed by ReLU activation, and the second FC layer is followed by Sigmoid activation). The first FC layer usually performs dimensionality reduction to reduce the amount of computation, and the second FC layer restores the original number of channels. The Sigmoid function normalizes the output to the [0,1] interval to indicate the importance of each channel.

[0066] Scale operation: Apply the weights obtained by the Excitation operation to the corresponding channels of the original feature map, and recalibrate the channel features of the feature map through multiplication operations. In this way, important features will be enhanced, while unimportant features will be suppressed.

[0067] Furthermore, the target detection module adopts a feature pyramid network structure, selects the MobileNet network as the backbone network, and reduces the number of feature layers according to task requirements.

[0068] Specifically, in this embodiment, the optimization of the target detection module is:

[0069] Multi-scale feature fusion: The feature pyramid network (FPN) structure is introduced to fuse feature maps at different levels to improve the model's ability to detect multi-scale targets. This helps to accurately identify safety hazards of different sizes and shapes around oil and gas pipelines.

[0070] Lightweight network design: For target detection tasks, lightweight network structures (such as MobileNetV3, EfficientNet, etc.) are designed or selected to reduce model complexity, increase inference speed, and improve the efficiency of oil and gas pipeline inspections while ensuring detection accuracy.

[0071] Anchor box optimization: The size and proportion of anchor boxes are optimized through cluster analysis and other methods to make them more suitable for the target distribution in oil and gas pipeline inspection scenarios and improve the recall and accuracy of detection.

[0072] Feature Pyramid Networks (FPN): FPN is an effective architecture for multi-scale object detection. It fuses feature maps of different depths (usually from different layers of convolutional neural networks) through top-down paths and lateral connections. Shallow feature maps contain more spatial detail information, which is beneficial for the detection of small objects; while deep feature maps contain richer semantic information, which is more beneficial for the detection of large objects.

[0073] Implementation: In FPN, features are first extracted through convolutional neural networks (such as ResNet, VGG, etc.) to form feature maps of multiple scales. Then, the spatial resolution of the deep feature map is increased by upsampling, and fused with the shallow feature map through lateral connections. The fusion method usually includes simple element addition or splicing. Finally, each fused feature map is used as an independent detection layer to output the detection results of the corresponding scale.

[0074] Through strategies such as multi-scale feature fusion, lightweight network design, and anchor frame optimization, the performance of the target detection module in oil and gas pipeline inspection scenarios can be significantly improved, achieving efficient and accurate identification of safety hazards.

[0075] Optionally, the location calculation module obtains the location of the potential safety hazard on the map including:

[0076] Use the camera's intrinsic and extrinsic matrix to transform the pixel coordinates in the image coordinate system into the camera coordinate system;

[0077] The coordinates in the camera coordinate system are converted into the geographic coordinate system by combining the GPS coordinates and attitude information of the drone to obtain the location of the safety hazard on the map.

[0078] Specifically, in the present embodiment, the positioning module is used to accurately locate the hidden danger. Based on the output results of the target detection module, the module can combine the location information in the image (such as pixel coordinates) and the location information of the drone (such as GPS coordinates) to obtain the precise location of the hidden danger on the map through coordinate conversion and calculation. In order to improve the positioning accuracy, map matching, multi-sensor fusion and other technologies can also be introduced. First, the pixel coordinates in the image coordinate system are converted to the camera coordinate system using the camera's intrinsic parameter matrix and extrinsic parameter matrix, and then the coordinates in the camera coordinate system are converted to the geographic coordinate system in combination with the drone's GPS coordinates and attitude information. Taking into account the influence of factors such as the angle and distance of the drone during shooting on the positioning accuracy, corresponding calibration and correction methods can also be used to improve the positioning accuracy.

[0079] Furthermore, using the data set to train the original detection model includes:

[0080] Use the labeled data set to train the model and continuously adjust the model parameters through the back-propagation algorithm;

[0081] During the training process, cross-validation is used to avoid overfitting, and the training effect is optimized by adjusting the learning rate and batch size hyperparameters.

[0082] Specifically, in this embodiment, the model is trained using a labeled data set, and the model parameters are continuously adjusted through a back-propagation algorithm to minimize the prediction error and improve the recognition accuracy.

[0083] During the training process, cross-validation, early stopping and other techniques can be used to avoid overfitting, and the training effect can be optimized by adjusting hyperparameters such as learning rate and batch size.

[0084] Use an independent test set to evaluate the trained model, and calculate indicators such as recognition accuracy, recall rate, and F1 score to evaluate model performance.

[0085] The model is further optimized and adjusted based on the evaluation results, such as improving the network structure and introducing the attention mechanism to improve recognition accuracy and real-time performance.

[0086] Furthermore, the oil and gas pipeline safety detection model is used to conduct real-time detection of oil and gas pipelines and locate safety hazards, including:

[0087] Real-time collection of image data of the oil and gas pipeline surface and surrounding environment;

[0088] The real-time collected image data is input into the oil and gas pipeline safety detection model to obtain an inspection report including hidden danger type, location, and timestamp information.

[0089] Specifically, in this embodiment, on-site deployment and application:

[0090] Deploy the trained deep learning model into the drone inspection system.

[0091] When performing inspection tasks, the drone takes real-time images of the surrounding area of ​​the pipeline and transmits them to the ground station or cloud server.

[0092] The ground station or cloud server uses deep learning models to analyze and process images and identify safety hazards and their location information.

[0093] Once a hidden danger is discovered, an alarm will be triggered immediately and the relevant departments will be notified for emergency treatment. At the same time, the hidden danger information will be recorded for subsequent tracking and evaluation.

[0094] This embodiment uses a deep learning model combined with technologies such as drones and the Internet of Things to achieve automated inspections of oil and gas pipelines, greatly reducing the cycle and cost of manual inspections. Through intelligent recognition of inspection images and videos using deep learning algorithms, abnormal conditions around the pipeline, such as corrosion, leakage, third-party damage, etc., can be accurately detected, thereby improving the accuracy of inspections. Not only can it identify abnormal conditions, but it can also quickly locate problem areas through data analysis to provide precise guidance for emergency response. The data analysis results based on deep learning can provide strong support for emergency decision-making and improve the efficiency and effectiveness of emergency response. Automated inspections reduce the workload and risks of inspectors, especially in harsh environments or dangerous areas, and avoid the safety risks that may be caused by manual inspections. Through real-time monitoring and early warning, potential safety hazards can be discovered and dealt with in advance, improving the overall safety of oil and gas pipelines.

[0095] The oil and gas pipeline safety inspection and positioning method proposed in this embodiment has significant beneficial effects in improving inspection efficiency and accuracy, achieving real-time monitoring and early warning, improving emergency response capabilities, reducing manual inspection risks, and intelligent management and optimization. These effects not only help to ensure the safe operation of oil and gas pipelines, but also provide more efficient and intelligent management methods for related enterprises.

[0096] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for safety inspection and positioning of oil and gas pipelines, characterized in that: include: Collect image data of the oil and gas pipeline surface and surrounding environment to build a data set; Construct the original detection model using convolutional neural network and feature pyramid network; Using the data set to train the original detection model to obtain a final oil and gas pipeline safety detection model; The oil and gas pipeline safety detection model is used to perform real-time detection of the oil and gas pipeline and locate safety hazards.

2. The oil and gas pipeline safety inspection and positioning method according to claim 1 is characterized in that: Building a dataset includes: Use drones equipped with image acquisition devices to conduct regular or irregular inspections and photography around oil and gas pipelines; Preprocess the collected image data, including denoising, contrast enhancement, cropping, and data enhancement; The potential safety hazards in the preprocessed image data are manually annotated to generate the data set for model training.

3. The oil and gas pipeline safety inspection and positioning method according to claim 1 is characterized in that: The original detection model includes: a feature extraction module, a target detection module, a positioning calculation module and a result fusion module; The feature extraction module uses a convolutional neural network to perform deep feature extraction on the input image and generate a high-level feature map; The target detection module performs target recognition on the feature map and outputs detection information; wherein the detection information includes: the category, location and confidence level of the potential safety hazard; The positioning calculation module performs coordinate conversion and calculation based on the output result of the target detection module and the GPS coordinates of the drone to obtain the location of the potential safety hazard on the map; The result fusion module is used to fuse the target detection result and the positioning result to generate an inspection report; wherein the inspection report includes: hidden danger type, location, and timestamp information.

4. The oil and gas pipeline safety inspection and positioning method according to claim 3 is characterized in that: An attention mechanism is added to the feature extraction module, and depthwise separable convolution is used instead of traditional standard convolution.

5. The oil and gas pipeline safety inspection and positioning method according to claim 3 is characterized in that: The target detection module adopts a feature pyramid network structure, selects the MobileNet network as the backbone network, and reduces the number of feature layers according to task requirements.

6. The oil and gas pipeline safety inspection and positioning method according to claim 3 is characterized in that: The location calculation module obtains the location of the potential safety hazard on the map by: Use the camera's intrinsic and extrinsic matrix to transform the pixel coordinates in the image coordinate system into the camera coordinate system; The coordinates in the camera coordinate system are converted into the geographic coordinate system by combining the GPS coordinates and attitude information of the drone to obtain the location of the safety hazard on the map.

7. The oil and gas pipeline safety inspection and positioning method according to claim 1 is characterized in that: Using the data set to train the original detection model includes: Use the labeled data set to train the model and continuously adjust the model parameters through the back-propagation algorithm; During the training process, cross-validation is used to avoid overfitting, and the training effect is optimized by adjusting the learning rate and batch size hyperparameters.

8. The oil and gas pipeline safety inspection and positioning method according to claim 1 is characterized in that: Using the oil and gas pipeline safety detection model, real-time detection of oil and gas pipelines and location of safety hazards include: Real-time collection of image data of the oil and gas pipeline surface and surrounding environment; The image data collected in real time is input into the oil and gas pipeline safety detection model to obtain an inspection report including hidden danger type, location, and timestamp information.

Images