Petroleum pipeline defect size measurement method based on HMCNet

Abstract

The invention discloses a petroleum pipeline defect size measurement method based on HMCNet, and the method comprises the steps: S1, initializing parameters of petroleum pipeline shooting equipment, carrying out the multi-angle shooting of a petroleum pipeline, and obtaining a plurality of groups of image information of different parts of the petroleum pipeline; s2, inputting the acquired image information into a deep learning network HMCNet which is trained by using a large number of known defect samples in advance; s3, dividing the features of the input image into a plurality of regions through HMCNet, enhancing local attention to the target image by using a region sensing module, and identifying and positioning defect information; and S4, obtaining a processing result of the HMCNet, and outputting the position, the type and the defect size of the petroleum pipeline defect. According to the invention, rapid and accurate detection of the position, the type and the size of the petroleum pipeline defect is realized, the detection efficiency and the anti-interference capability are greatly improved, and the operation is simpler and more convenient.

Description

Technical Field

[0001] The present invention relates to the technical field of oil pipeline detection, and specifically provides a method for measuring the size of oil pipeline defects based on HMCNet. Background Art

[0002] As a key infrastructure for oil transportation, the safety of oil pipelines is of crucial importance. Once a defect occurs in the pipeline, resulting in oil leakage, it will not only cause huge economic losses but also lead to serious consequences such as environmental pollution and safety accidents. Accurately measuring the size of oil pipeline defects is of great significance for timely evaluating the safety status of the pipeline and formulating a reasonable maintenance plan. Therefore, it is urgent to develop a high-precision and efficient method for measuring the size of oil pipeline defects.

[0003] Traditional techniques for measuring the size of oil pipeline defects mainly include ultrasonic testing, magnetic particle testing, radiographic testing, etc. Ultrasonic testing uses the propagation characteristics of ultrasonic waves in different media to detect defects, but there are certain errors in judging the shape and position of defects, and it is difficult to detect defects in complex structural parts; magnetic particle testing is only applicable to ferromagnetic materials, with great limitations; although radiographic testing has high detection accuracy, it has radiation hazards, complex operation and high costs. These traditional techniques often require professional operators, with low detection efficiency and are difficult to meet the needs of large-scale and rapid detection of modern oil pipelines.

[0004] With the development of computer technology and image processing technology, oil pipeline defect detection techniques based on image processing have gradually emerged. Some existing detection methods based on deep learning have improved the accuracy and efficiency of defect detection to a certain extent. However, these methods still have problems with insufficient detection accuracy when dealing with tiny defects in complex backgrounds. Moreover, most existing techniques can only detect the existence of defects and are difficult to accurately measure the size of defects. In addition, existing techniques lack in multi-modal information fusion and local feature extraction, and cannot make full use of various information in pipeline images, resulting in inaccurate judgment of defect types and sizes.

[0005] Therefore, neither traditional detection techniques nor existing detection methods based on deep learning can fully meet the high-precision, high-efficiency and intelligent requirements for measuring the size of oil pipeline defects. Therefore, this application proposes a method for measuring the size of oil pipeline defects based on HMCNet, which overcomes the deficiencies of the existing technology and provides a more reliable guarantee for the safe operation of oil pipelines. Summary of the Invention

[0006] Based on the above, a method for measuring the size of oil pipeline defects based on HMCNet specifically includes the following steps:

[0007] S1. Initialize the parameters of the oil pipeline imaging device, take multi-angle images of the oil pipeline, and obtain multiple groups of image information of different parts of the oil pipeline;

[0008] S2. Input the obtained image information into the deep learning network HMCNet that has been pre-trained with a large number of known defect samples;

[0009] S3. Divide the input image features into multiple regions through HMCNet, use the region perception module to enhance the local attention to the target image, and identify and locate the defect information;

[0010] S4. Obtain the processing results of HMCNet and output the position, type, and defect size of the oil pipeline defects.

[0011] Preferably, in S1, an intelligent pan-tilt handheld device integrating an inertial measurement unit and a laser range finder is used. Based on the central axis of the pipeline, the laser range finder is used to measure the distance in real time, and the integrated inertial measurement unit is used to monitor the attitude. The pipeline is moved around and photographed according to a spiral trajectory to obtain multi-view image information.

[0012] Preferably, the multi-view image information obtained in S1 includes the overall appearance image of the pipeline and the local close-up image of the pipeline; the overall appearance image of the pipeline includes images of the pipeline connection part and the support structure connection area; the local close-up image of the pipeline includes images of the weld, the pipeline surface texture, and small pits or protrusions.

[0013] Preferably, the deep learning network HMCNet in S2 includes a backbone network, a neck network, and a head network; the obtained image information is transmitted to the backbone network to extract multi-scale features of the image, the multi-scale features of the image extracted by the backbone network are fused through the neck network to extract semantic information, and the head network performs object detection based on the semantic information provided by the neck network to generate detection results.

[0014] Preferably, the backbone network adopts a multi-branch heterogeneous fusion architecture, and the multi-branch heterogeneous fusion architecture includes a lightweight branch and a deep convolution branch; the lightweight branch reduces the calculation amount through grouped convolution and pointwise convolution and extracts the basic features of the pipeline image; the deep convolution branch dynamically adjusts the weights of different convolution kernels according to the input image features to enhance the ability to capture complex defect features; the feature maps of the lightweight branch and the deep convolution branch are fused through a cross-channel interaction module, and the attention mechanism is used to reallocate the features in the channel dimension to obtain the multi-scale features of the image.

[0015] Preferably, the neck network adopts a cross-scale feature interaction architecture. By constructing a multi-scale feature fusion tree, different-scale feature maps output by the backbone network are used as nodes of the tree, and each node is connected to other nodes at different levels through skip connections to perform cross-scale feature interaction.

[0016] Preferably, the head network adopts a multi-modal fusion architecture to detect oil pipeline defects; the head network is composed of a target classification sub-network, a bounding box regression sub-network, and a key point detection sub-network in parallel; the target classification sub-network introduces a meta-learning mechanism. By pre-training a set of general classifier weights, when facing different types of pipeline defect images, the weights are adjusted, and multi-modal features are used for defect type classification;

[0017] The bounding box regression sub-network adjusts the scale and ratio of the anchor points dynamically according to the feature map output by the neck network through an adaptive anchor point generation strategy to adapt to defects of different sizes and shapes;

[0018] The key point detection sub-network adopts a graph convolutional network structure, takes the key points of the defect as nodes in the graph, and learns the spatial distribution characteristics of the key points through the connection relationship between the nodes;

[0019] The outputs of the target classification sub-network, the bounding box regression sub-network, and the key point detection sub-network are integrated through a dynamic fusion mechanism. According to the uncertainty of the input features and the confidence of each sub-network, the weights in the output results are adjusted, and the defect classification, bounding box, and key point information of the image are output.

[0020] Preferably, in step S3, the input image features are divided into multiple regions by HMCNet, and the local attention to the target image is enhanced through a region perception module. The formula is: Among them, is the feature after region perception, and R(·) is the region perception module; HMCNET divides the image features into multiple regions and then uses the region perception module to distinguish defect types.

[0021] Preferably, in step S3, the self-attention mechanism of Transformer is used to capture image information to locate defect information; the image features and target query information are processed by the encoder and decoder of Transformer respectively; the encoder process is: Z enc = TransformerEnc(F r , Z pos ), where Z enc is the feature processed by the encoder, is the feature after region perception, and Z pos is the position encoding; the decoder process is: Z dec= TransformerDec(Z enc , Z query ), where Z query is the target query vector, and Z dec is the result output by the decoder; the defect position is output by capturing the image information through the self-attention mechanism of the Transformer.

[0022] Preferably, in S4, after obtaining the processing result of HMCNet, the position, type, and defect size of the oil pipeline defect are output to the intelligent warning platform, the defect information is obtained in real time and warned, and the defect information is notified to relevant personnel for processing through different warning channels, including text messages, emails, or communication software.

[0023] Compared with the prior art, the technical solution of the present application has the following technical effects:

[0024] By using the intelligent cloud platform handheld device integrating an inertial measurement unit and a laser rangefinder sensor, and taking multi-view image information by moving around the pipeline along a spiral trajectory with the pipeline central axis as the reference, the present invention solves the technical problem that it is difficult to comprehensively obtain images of different parts of the oil pipeline by traditional shooting methods, resulting in some defects being easily missed, and obtains the technical effect of being able to obtain multi-view image information including the overall appearance of the pipeline (such as connection parts, support structure connection areas) and local close-ups (such as welds, surface textures, and small pits or protrusions), providing a rich and comprehensive data basis for subsequent accurate detection and measurement of defects.

[0025] The present invention adopts the deep learning network HMCNet, whose backbone network adopts a multi-branch heterogeneous fusion architecture, the neck network adopts a cross-scale feature interaction architecture, and the head network adopts a multi-modal fusion architecture. This technical solution solves the technical problems that the existing deep learning-based detection methods have insufficient detection accuracy for tiny defects in complex backgrounds and are difficult to accurately measure the defect size, and obtains the technical effect of being able to effectively capture complex defect features, fuse multi-scale features and multi-modal information, thereby accurately identifying and locating the oil pipeline defect information, accurately measuring the defect size, and accurately judging the defect type.

[0026] By using the region perception module to enhance the local attention to the target image, process the image features, and using the self-attention mechanism of the Transformer to capture the image information to locate the defect, the present invention solves the technical problem that it is difficult to focus on tiny defects during the detection process and unable to accurately locate the defect position, and obtains the technical effect of being able to enhance the attention to specific regions of the target image, more accurately capture the defect information in the image, and then achieve the precise positioning of the oil pipeline defect position, providing an accurate position basis for subsequent defect evaluation and repair.

[0027] The technical solution of the present invention outputs the processing result of HMCNet to the intelligent early warning platform and notifies relevant personnel through different early warning channels such as text messages, emails or communication software, which solves the technical problem that information cannot be transmitted in time after detecting defects, resulting in the defects not being processed in time and posing potential safety hazards, and obtains the technical effect of being able to obtain defect information in real time and notify relevant personnel for processing in time, greatly improving the timeliness and safety of oil pipeline maintenance and reducing the pipeline operation risk.

[0028] The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, so that it can be implemented according to the content of the specification, and in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the following takes the preferred embodiments of this application and combines with the drawings to describe in detail as follows.

[0029] According to the following detailed description of the specific embodiments of this application in combination with the drawings, those skilled in the art will be more clear about the above and other purposes, advantages and features of this application. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of this application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to these drawings. In all the drawings, similar elements or parts are generally marked with similar reference numerals. In the drawings, the elements or parts do not necessarily draw according to the actual scale.

[0031] Figure 1 Flow chart of a method for measuring the size of oil pipeline defects based on HMCNet of the present invention;

[0032] Figure 2 Flow chart of the deep learning network HMCNet of a method for measuring the size of oil pipeline defects based on HMCNet of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0033] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Apparently, the described embodiments are some, but not all, of the embodiments of this application. In the following description, specific details such as specific configurations and components are provided only to assist in a comprehensive understanding of the embodiments of this application. Therefore, those skilled in the art should clearly understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Additionally, descriptions of known functions and configurations are omitted for clarity and conciseness in the embodiments.

[0034] It should be understood that the term "one embodiment" or "this embodiment" mentioned throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of this application. Therefore, the appearances of the term "one embodiment" or "this embodiment" throughout the specification do not necessarily refer to the same embodiment. Furthermore, these specific features, structures, or characteristics can be combined in one or more embodiments in any suitable manner.

[0035] In addition, this application may repeat reference numerals and / or letters in different instances. This repetition is for the purpose of simplicity and clarity and does not in itself indicate the relationship between the various embodiments and / or arrangements discussed.

[0036] The term "and / or" in this document is merely a description of the association relationship of associated objects, indicating that three relationships can exist. For example, A and / or B can represent: A exists alone, B exists alone, and both A and B exist simultaneously. The term " / and" in this document describes another association relationship of associated objects, indicating that two relationships can exist. For example, A / and B can represent: A exists alone, and both A and B exist. Additionally, the character " / " in this document generally indicates that the associated objects before and after are in an "or" relationship.

[0037] The term "at least one" in this document is merely a description of the association relationship of associated objects, indicating that three relationships can exist. For example, at least one of A and B can represent: A exists alone, both A and B exist simultaneously, and B exists alone.

[0038] It should also be noted that in this document, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprises", "comprising", or any other variation thereof is intended to cover a non-exclusive inclusion.

[0039] Embodiment 1

[0040] This embodiment mainly describes a method for measuring the defect size of oil pipelines based on HMCNet, as Figure 1 shown, which includes the following steps:

[0041] S1. Initialize the parameters of the oil pipeline shooting device, shoot the oil pipeline from multiple angles, and obtain multiple groups of image information of different parts of the oil pipeline;

[0042] S2. Input the obtained image information into the deep learning network HMCNet that has been trained in advance using a large number of known defect samples;

[0043] S3. Divide the input image features into multiple regions through HMCNet, use the region perception module to enhance the local attention to the target image, and identify and locate the defect information;

[0044] S4. Obtain the processing result of HMCNet and output the position, type and defect size of the oil pipeline defect.

[0045] Furthermore, in S1, use an intelligent cloud platform handheld device integrating an inertial measurement unit and a laser rangefinder sensor. Based on the central axis of the pipeline, use the laser rangefinder sensor to measure the distance in real time, and use the integrated inertial measurement unit to monitor the attitude. Move around the pipeline along a spiral trajectory to take pictures and obtain multi-view image information.

[0046] The multi-angle shooting uses a handheld device equipped with an intelligent cloud platform. The intelligent cloud platform integrates a high-precision inertial measurement unit (IMU) and environmental perception sensors, such as a laser rangefinder sensor and a vision sensor. The laser rangefinder sensor measures the distance of the oil pipeline in real time. Combining with the pipeline contour information obtained by the vision sensor, the intelligent cloud platform automatically plans the shooting path based on the central axis of the pipeline according to the preset shooting strategy; during the shooting process, the IMU monitors the attitude change of the handheld device in real time to ensure that the angle deviation of each shooting is controlled within a very small range. Move the handheld device around the pipeline along a spiral ascending or descending trajectory, and take pictures at different axial positions and circumferential angles. The axial spacing and circumferential angle interval of each shooting are dynamically adjusted according to the pipe diameter of the pipeline and the preset accuracy requirements, so as to obtain high-resolution image information covering all parts of the oil pipeline, different perspectives.

[0047] Furthermore, the multi-view image information obtained in S1 includes the overall appearance image of the pipeline and the local close-up image of the pipeline; the overall appearance image of the pipeline includes the images of the pipeline connection part and the support structure connection area; the local close-up image of the pipeline includes the images of the weld, the pipeline surface texture and the small pits or protrusions.

[0048] Further, the deep learning network HMCNet in S2 includes a backbone network, a neck network, and a head network; the acquired image information is transmitted to the backbone network to extract multi-scale features of the image, the multi-scale features of the image extracted by the backbone network are fused through the neck network to extract semantic information, and the head network performs object detection based on the semantic information provided by the neck network to generate detection results.

[0049] Further, the backbone network adopts a multi-branch heterogeneous fusion architecture, and the multi-branch heterogeneous fusion architecture includes a lightweight branch and a deep convolution branch; the lightweight branch reduces the computational amount through grouped convolution and pointwise convolution to extract the basic features of the pipeline image; the deep convolution branch dynamically adjusts the weights of different convolutional kernels according to the input image features to enhance the ability to capture complex defect features; the feature maps of the lightweight branch and the deep convolution branch are fused through a cross-channel interaction module, and the attention mechanism is used to reallocate the features in the channel dimension to obtain the multi-scale features of the image.

[0050] Further, the neck network adopts a cross-scale feature interaction architecture. By constructing a multi-scale feature fusion tree, different-scale feature maps output by the backbone network are used as nodes of the tree, and each node is connected to other nodes at different levels through skip connections for cross-scale feature interaction.

[0051] Further, the head network adopts a multi-modal fusion architecture to detect oil pipeline defects; the head network is composed of a target classification sub-network, a bounding box regression sub-network, and a key point detection sub-network in parallel; the target classification sub-network introduces a meta-learning mechanism. By pre-training a set of general classifier weights, when facing different types of pipeline defect images, the weights are adjusted, and multi-modal features are used for defect type classification;

[0052] The bounding box regression sub-network dynamically adjusts the scale and ratio of the anchor points according to the feature map output by the neck network through an adaptive anchor point generation strategy to adapt to defects of different sizes and shapes;

[0053] The key point detection sub-network adopts a graph convolutional network structure, takes the key points of the defect as nodes in the graph, and learns the spatial distribution features of the key points through the connection relationship between the nodes;

[0054] The outputs of the target classification sub-network, the bounding box regression sub-network, and the key point detection sub-network are integrated through a dynamic fusion mechanism. According to the uncertainty of the input features and the confidence of each sub-network, the weights in the output results are adjusted to output the defect classification, bounding box, and key point information of the image.

[0055] Further, in S3, the input image features are divided into multiple regions through HMCNet, and the local attention to the target image is enhanced through a region perception module. The formula is: Among them, is the feature after region perception, and R(·) is the region perception module; HMCNET divides the image features into multiple regions and then uses the region perception module to distinguish the defect types.

[0056] Furthermore, in S3, to locate the defect information, the self-attention mechanism of Transformer is used to capture the image information for defect location; the image features and target query information are processed by the encoder and decoder of Transformer respectively; the encoder process is: Z enc = TransformerEnc(Fr, Z pos ), where Z enc is the feature processed by the encoder, is the feature after region perception, Z pos is the position encoding; the decoder process is: Z dec = TransformerDec(Z enc , Z query ), where Z query is the target query vector, and Z dec is the result output by the decoder; the self-attention mechanism of Transformer is used to capture the image information and output the defect location.

[0057] Furthermore, in S4, after obtaining the processing result of HMCNet, the location, type, and defect size of the oil pipeline defect are output to the intelligent early warning platform, the defect information is obtained in real time and early warning is carried out, and the defect information is notified to relevant personnel for processing through different early warning channels, including text messages, emails, or communication software.

[0058] This embodiment details how the present application uses an intelligent cloud platform handheld device integrating an inertial measurement unit and a laser range sensor to obtain multi-view image information of an oil pipeline, inputs it into the deep learning network HMCNet, combines a region perception module and a Transformer self-attention mechanism for defect recognition and location, and finally outputs the processing result to the intelligent early warning platform to notify relevant personnel, solving the problems of low accuracy, low efficiency, and difficulty in detecting defects in complex parts of traditional detection technologies, as well as the insufficient detection of micro-defects, inability to accurately measure sizes, and lack of timely early warning in existing deep learning detection methods.

[0059] Based on Embodiment 1, this embodiment describes the processing of oil pipeline images by the deep learning network HMCNet and outputs defect information, as Figure 2 shown, specifically:

[0060] Obtain the image information of the oil pipeline and transmit it to the backbone network of HMCNet. The backbone network adopts a multi-branch heterogeneous fusion architecture, including a lightweight branch and a deep convolutional branch. The lightweight branch reduces the computational complexity through grouped convolution and pointwise convolution, extracts the basic features of the pipeline image, reduces the computational burden while ensuring the extraction of basic features, and improves the processing efficiency. The deep convolutional branch dynamically adjusts the weights of different convolutional kernels according to the input image features, enhances the ability to capture complex defect features, and can effectively identify various complex defect situations. The feature maps generated by the lightweight branch and the deep convolutional branch will be fused through a cross-channel interaction module, and the attention mechanism is used to reallocate the features in the channel dimension, so as to obtain the multi-scale features of the image and provide rich feature information for subsequent processing.

[0061] The multi-scale features processed by the backbone network will enter the neck network. The neck network adopts a cross-scale feature interaction architecture. By constructing a multi-scale feature fusion tree, the feature maps of different scales output by the backbone network are used as the nodes of the tree. Each node is connected to other nodes at different levels through skip connections for cross-scale feature interaction, so that the features of different scales can complement and fuse each other, further extract more representative semantic information, and improve the ability to understand the image content.

[0062] The semantic information output by the neck network will be sent to the head network. The head network adopts a multi-modal fusion architecture, which is composed of a target classification sub-network, a bounding box regression sub-network, and a key point detection sub-network in parallel. The target classification sub-network introduces a meta-learning mechanism. By pre-training a set of general classifier weights, it adjusts the weights when facing different types of pipeline defect images, and uses multi-modal features for defect type classification, and can accurately judge the specific type of the defect. The bounding box regression sub-network adopts an adaptive anchor generation strategy, dynamically adjusts the scale and ratio of the anchor according to the feature map output by the neck network, adapts to defects of different sizes and shapes, and thus more accurately determines the boundary range of the defect. The key point detection sub-network adopts a graph convolutional network structure, takes the key points of the defect as the nodes in the graph, and learns the spatial distribution features of the key points through the connection relationship between the nodes. The outputs of the three sub-networks will be integrated through a dynamic fusion mechanism. According to the uncertainty of the input features and the confidence of each sub-network, the weights in the output results are adjusted, and the defect classification, bounding box, and key point information of the image are output.

[0063] HMCNet enhances the local attention to the target image through the region perception module. It divides the input image features into multiple regions, enabling the network to focus more on the defect regions and distinguish defect types. Moreover, it uses the self-attention mechanism of Transformer to capture image information for defect localization. By processing the image features and target query information through the encoder and decoder respectively, it finally accurately outputs the location information of the oil pipeline defects. Through this series of complex and orderly processing procedures, HMCNet can precisely process the oil pipeline images and output key information such as the location, type, and size of the defects.

[0064] For the region perception module, when processing the oil pipeline image, HMCNet first divides the input image features into multiple regions, and the region perception module then plays a role on this basis. By focusing on specific local regions of the target image, it strengthens the extraction and analysis of the features of these regions, thus better distinguishing different types of defects. For example, when processing an image containing defects such as tiny potholes or at the weld, the region perception module can make the network focus on the local areas with defects and ignore the interference of irrelevant background information, effectively enhancing the sensitivity and recognition ability to defect features, providing strong support for the subsequent accurate identification and localization of oil pipeline defects and accurate judgment of defect types, and ensuring that the entire detection process is more accurate and efficient.

[0065] This embodiment details the deep learning network HMCNet, which efficiently processes oil pipeline images through a multi-branch heterogeneous fusion backbone network, a cross-scale feature interaction neck network, a multi-modal fusion head network, as well as a region perception module and a Transformer self-attention mechanism. It accurately captures the multi-scale features of tiny defects in complex backgrounds from multi-view images, accurately identifies and locates defects through multi-modal information fusion, accurately judges defect types, and precisely measures defect sizes.

[0066] The above are only the preferred embodiments of the present invention, and it does not thereby limit the protection scope of the present invention. For those skilled in the art, the present invention can have various changes and modifications; all changes, modifications, substitutions, integrations, and parameter changes made to these embodiments within the spirit and principle of the present invention by means of conventional substitutions or capable of achieving the same functions without departing from the principle and spirit of the present invention fall within the protection scope of the present invention.

Claims

1. A method for measuring the defect size of a petroleum pipeline based on HMCNet, characterized in that: The following steps are involved: S1. Initialize the parameters of the oil pipeline shooting equipment, shoot the oil pipeline from multiple angles, and obtain multiple sets of image information of different parts of the oil pipeline; S2, input the acquired image information into the deep learning network HMCNet which is pre-trained with a large number of known defect samples; S3, divide the input image features into multiple regions through HMCNet, use the regional perception module to enhance the local attention of the target image, identify and locate the defect information; S4. Obtain the processing results of HMCNet and output the location, type and defect size of the oil pipeline defect.

2. The method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 1, characterized in that: In S1, an intelligent gimbal handheld device integrating an inertial measurement unit and a laser ranging sensor is used. The central axis of the pipeline is used as a reference, the laser ranging sensor is used to measure the distance in real time, and the integrated inertial measurement unit is used to monitor the posture. The device moves around the pipeline in a spiral trajectory to capture multi-view image information.

3. The method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 2, characterized in that: The multi-view image information acquired in S1 includes an overall appearance image of the pipeline and a local close-up image of the pipeline; the overall appearance image of the pipeline includes images of the pipeline connection parts and the support structure connection areas; the local close-up image of the pipeline includes images of the welds, pipeline surface textures and tiny pits or protrusions.

4. The method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 2, characterized in that: The deep learning network HMCNet in S2 includes a backbone network, a neck network and a head network; the acquired image information is transmitted to the backbone network to extract multi-scale features of the image, the multi-scale features of the image extracted by the backbone network are fused through the neck network to extract semantic information, and the head network is used to perform target detection based on the semantic information provided by the neck network to generate detection results.

5. A method for measuring the defect size of a petroleum pipeline based on HMCNet according to claims 1 and 4, characterized in that: The backbone network adopts a multi-branch heterogeneous fusion architecture, which includes a lightweight branch and a deep convolution branch. The lightweight branch reduces the amount of calculation through grouped convolution and point-by-point convolution to extract the basic features of the pipeline image. The deep convolution branch dynamically adjusts the weights of different convolution kernels according to the input image features to enhance the ability to capture complex defect features. The feature maps of the lightweight branch and the deep convolution branch are fused through a cross-channel interaction module, and the attention mechanism is used to redistribute the features in the channel dimension to obtain multi-scale features of the image.

6. A method for measuring the defect size of a petroleum pipeline based on HMCNet according to claims 1 and 4, characterized in that: The neck network adopts a cross-scale feature interaction architecture. By constructing a multi-scale feature fusion tree, the different scale feature maps output by the backbone network are used as tree nodes. Each node is connected to other nodes at different levels through jump connections to perform cross-scale feature interaction.

7. A method for measuring the defect size of a petroleum pipeline based on HMCNet according to claims 1 and 4, characterized in that: The head network adopts a multimodal fusion architecture to detect oil pipeline defects; the head network is composed of a target classification subnetwork, a bounding box regression subnetwork and a key point detection subnetwork in parallel; the target classification subnetwork introduces a meta-learning mechanism, and by pre-training a set of general classifier weights, when facing different types of pipeline defect images, the weights are adjusted and multimodal features are used to classify defect types; The bounding box regression subnetwork uses an adaptive anchor point generation strategy to dynamically adjust the scale and proportion of the anchor points according to the feature map output by the neck network to adapt to defects of different sizes and shapes; The key point detection subnetwork adopts a graph convolutional network structure, takes the key points of defects as nodes in the graph, and learns the spatial distribution characteristics of the key points through the connection relationship between the nodes; The outputs of the target classification subnetwork, bounding box regression subnetwork, and key point detection subnetwork are integrated through a dynamic fusion mechanism. According to the uncertainty of the input features and the confidence of each subnetwork, the weights in the output results are adjusted to output the defect classification, bounding box, and key point information of the image.

8. The method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 1, characterized in that: In S3, the input image features are divided into multiple regions through HMCNet, and the local attention to the target image is enhanced through the region perception module. The formula is: in, is the feature after region perception, and R(·) is the region perception module. HMCNET divides the image features into multiple regions and uses the region perception module to distinguish the defect types.

9. A method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 1 or 8, characterized in that: The positioning defect information in S3 uses the Transformer's self-attention mechanism to capture image information to locate defects; the encoder and decoder of the Transformer process image features and target query information respectively; the encoder process is: Z enc = TransformerEnc(F r ,Z pos ), where Z enc is the feature processed by the encoder, is the feature after region perception, Z pos is the position code; the decoder process is: Z dec = TransformerDec(Z enc ,Z query ), where Z query is the target query vector, Z dec It is the result of the decoder output; the image information is captured through the Transformer's self-attention mechanism to output the defect location.

10. The method for measuring the defect size of a petroleum pipeline based on HMCNet according to claim 1, characterized in that: In S4, after obtaining the processing result of HMCNet, the location, type and defect size of the oil pipeline defect are output to the intelligent early warning platform, the defect information is obtained in real time, and an early warning is issued. The defect information is notified to relevant personnel for processing through different early warning channels, including SMS, email or communication software.

Images