A buried pipeline detection method based on three-component magnetic field curve image recognition

Abstract

The present invention relates to the field of pipeline detection technology, and specifically to a buried pipeline detection method based on three-component magnetic field curve image recognition. It includes: S1: acquiring detection data of buried pipelines based on non-contact metal magnetic memory detection; S2: converting the measured data into three-component magnetic field curve images, dividing them into defects, stress concentration areas, and normal, establishing database 1, and using convolutional neural network to train and generate classifier 1; S3: dividing the stress concentration area images in database 1 into the stress danger degree of the pipe section according to the intensity of the curve change, establishing database 2, and using convolutional neural network to train and generate classifier 2; S4: in actual detection, sending the newly detected three-component magnetic field curve image into classifier 1 for classification, sending the identified stress concentration area curve image into classifier 2, and judging its pipe section stress danger degree. The present invention can effectively detect defects or cracks in buried pipelines, and realize early diagnosis of defects to prevent them before they occur.

Description

Technical Field

[0001] The present invention relates to the technical field of pipeline detection, and in particular to a buried pipeline detection method based on three-component magnetic field curve image recognition. Background Art

[0002] With the development of metal magnetic memory detection technology, a non-contact detection method has been introduced to the field of buried pipeline detection. Compared with traditional internal detection or electrical detection methods, this method does not require excavation of the pipeline. At the same time, because this technology uses the natural magnetism of the pipeline metal, it does not need to emit any excitation source or any substance. The measured data can be directly converted into a three-component magnetic field curve using data processing software for observation, which can determine whether there are defects or cracks in the pipeline, or achieve early diagnosis of defects by judging the stress risk level of the pipe section.

[0003] For long-distance pipelines, manual observation of images not only takes a lot of time, but may also cause missed or wrong judgments due to the lack of experience of the staff. Therefore, it is urgent to propose an artificial intelligence method to replace manual judgment. Summary of the invention

[0004] In view of the above-mentioned shortcomings and deficiencies of the prior art, the present invention provides a buried pipeline detection method based on three-component magnetic field curve image recognition, which achieves the purpose of buried pipeline detection by obtaining a three-component magnetic field curve graph and using an artificial intelligence method to recognize the image.

[0005] In order to achieve the above object, the main technical solutions adopted by the present invention include:

[0006] The present invention provides a buried pipeline detection method based on three-component magnetic field curve image recognition, comprising the following steps:

[0007] S1, obtaining detection data of buried pipelines based on non-contact metal magnetic memory detection;

[0008] S2, converting the data measured in S1 into a three-component magnetic field curve image, and classifying them into three categories according to defects, stress concentration areas, and normal conditions, establishing a database 1, training the database 1 using a convolutional neural network, and generating a classifier 1;

[0009] S3, classify the stress danger degree of the pipe section according to the intensity of the curve change of the stress concentration area image in database 1, and divide it into three categories: severe, medium, and light, establish database 2, use convolutional neural network to train database 2, and generate classifier 2;

[0010] S4, during actual detection, the newly detected three-component magnetic field curve image is sent to classifier 1 for classification, and the identified stress concentration area curve map is sent to classifier 2 to determine the stress danger level of the pipe section during actual detection.

[0011] Furthermore, the method for detecting buried pipelines based on non-contact metal magnetic memory in S1 includes:

[0012] S11, determine the exact location and direction of the pipeline, mark it on the ground, and ensure data traceability;

[0013] S12, clear the ground obstacles and record the location of the metal objects that cannot be cleared;

[0014] S13, using a non-contact scanning detection device connected to an encoder, running at a constant speed above the pipeline, collecting data and recording the detection length, and connecting to a GPS positioning system to record location information;

[0015] S14, save the data every 100-200 meters of detection, and repeat until all buried pipelines are detected.

[0016] Furthermore, the database 1 in S2 converts the data measured in S1 into a three-component magnetic field curve image, which is divided into three categories: defect, stress concentration area, and normal through excavation verification and repeated comparison.

[0017] Furthermore, the convolutional neural network training adopts a convolutional neural network model for training, and the convolutional neural network model includes six convolutional layers and four fully connected layers. An attention GhostNet module and a maximum pooling layer of size 2×2 are added after each convolutional layer.

[0018] Furthermore, the attention GhostNet module is composed of a SENet and a GhostNet through a Concat function.

[0019] Furthermore, the SENet first performs a Squeeze operation on the features obtained by convolution to obtain global features, and then performs an Excitation operation on the global features to obtain weights of different channels, which are multiplied by the original features to finally obtain the final features.

[0020] Furthermore, the GhostNet first generates features through convolution operations, then performs a series of prior operations on each channel of the features to artificially generate redundant information, and finally concats this redundant information with the previous features.

[0021] Furthermore, the stress hazard level of the pipe section is: severe: the pipe is seriously damaged and should be replaced immediately; moderate: it can be used, but regular inspection is required; light: it can be used safely.

[0022] The beneficial effects of the present invention are as follows: the present invention provides a buried pipeline detection method based on three-component magnetic field curve image recognition, which can effectively detect defects or cracks in buried pipelines and realize early diagnosis of defects to prevent them from happening. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] Figure 1 This is a flow chart of a buried pipeline detection method based on three-component magnetic field curve image recognition according to the present invention;

[0024] Figure 2 A convolutional neural network structure diagram provided by the present invention;

[0025] Figure 3 A structural diagram of the attention GhostNet module provided by the present invention;

[0026] Figure 4 This is a flow chart of an image classification method provided by the present invention. DETAILED DESCRIPTION

[0027] In order to better explain the present invention and facilitate understanding, the present invention is described in detail below through specific implementation modes in conjunction with the accompanying drawings.

[0028] Embodiment 1:

[0029] like Figure 1 As shown, this embodiment provides a buried pipeline detection method based on three-component magnetic field curve image recognition. The method includes the following steps:

[0030] S1, obtaining detection data of buried pipelines based on non-contact metal magnetic memory detection;

[0031] S2, converting the data measured by S1 into a three-component magnetic field curve image, and classifying it into three categories according to defects, stress concentration areas, and normal conditions, establishing database 1, training database 1 using a convolutional neural network, and generating classifier 1;

[0032] S3, classify the stress danger degree of the pipe section according to the intensity of the curve change of the stress concentration area image in database 1, and divide it into three categories: severe, medium, and light, establish database 2, use convolutional neural network to train database 2, and generate classifier 2;

[0033] S4, during actual detection, the newly detected three-component magnetic field curve image is sent to classifier 1 for classification. If it is normal, it is used normally. If it is a defect, the actual position of the defect in the pipeline is marked, and it is verified through excavation whether it needs to be repaired. If it is a stress concentration area, the curve image is sent to classifier 2 to determine the stress danger level of the pipe section during actual detection.

[0034] Specifically, the method for detecting buried pipelines based on non-contact metal magnetic memory described in step S1 specifically includes:

[0035] S11. Determine the exact location and direction of the pipeline, mark it on the ground, and ensure data traceability;

[0036] S12. Clear obstacles on the ground and record the locations of metal objects that cannot be cleared;

[0037] S13, using a non-contact scanning detection device connected to an encoder, running at a constant speed above the pipeline, collecting data and recording the detection length, and connecting to a GPS positioning system to record location information;

[0038] S14. Save the data every 100-200 meters of detection, and repeat until all buried pipelines are detected.

[0039] The model of the non-contact scanning detection device is: 11-12WB or 11-6WB. When using it, it needs to be connected to the TSC-5M-32 stress concentration magnetic detector host.

[0040] The database 1 described in step S2 converts the measured data into a three-component magnetic field curve image, and divides the image into three categories according to defects, stress concentration areas, and normal conditions.

[0041] The database 2 described in step S3 verifies the data of the measured stress concentration area through excavation, and divides it into three categories: severe, medium and mild according to the severity of the three-component magnetic field curve image.

[0042] like Figure 2 As shown, this embodiment provides a convolutional neural network model with a total of six convolutional layers and four fully connected layers, and each convolutional layer is followed by an activation function Rule and BN layer, an attention GhostNet module, and a maximum pooling layer of size 2×2.

[0043] Specifically, in one embodiment, the convolution kernels of the first five layers of the convolution neural network model are 5×5, the convolution kernel of the last layer is 3×3, and the structure of each layer is the same. The convolution layer is followed by the activation function Rule and BN layer, the attention GhostNet module, and the maximum pooling of size 2×2.

[0044] The 224×224×3 image is input into the model, and the first layer of convolution outputs 16 220×220 feature maps, which are then output after the attention GhostNet module and maximum pooling to 48 110×110 feature maps.

[0045] The second convolution layer outputs 32 106×106 feature maps, which are then passed through the attention GhostNet module and max pooling to output 96 53×53 feature maps.

[0046] The third convolution layer outputs 64 49×49 feature maps, which are then passed through the attention GhostNet module and max pooling to output 192 25×25 feature maps.

[0047] The fourth convolution layer outputs 128 21×21 feature maps, which are then passed through the attention GhostNet module and max pooling to output 384 11×11 feature maps.

[0048] The fifth convolution layer outputs 256 7×7 feature maps, which are then passed through the attention GhostNet module and max pooling to output 768 4×4 feature maps.

[0049] The sixth convolution layer outputs 512 2×2 feature maps, which are then passed through the attention GhostNet module and max pooling to output 1536 1×1 feature maps.

[0050] The features of the last layer are converted into a 1536-dimensional vector and passed to the fully connected layer, and finally sent to the softmax function for classification. The softmax function is also called the normalized exponential function, which is defined as

[0051]

[0052] Among them, Z i is the output value of the i-th node, C is the number of output nodes, that is, the classification category. The softmax function can be used to convert the multi-classification value into a probability distribution between [0,1] and the sum is 1.

[0053] The database 1 in step S2 is sent into the above-mentioned convolutional neural network model to train and generate classifier 1, and the softmax function will output a 3D vector.

[0054] The database 2 in step S3 is sent to the above-mentioned convolutional neural network model to train and generate classifier 2, and the softmax function will output a 3D vector.

[0055] like Figure 3As shown, this embodiment provides an attention GhostNet module, which is mainly composed of a SENet and a GhostNet through a Concat function.

[0056] SENet is a channel-dimensional attention that allows the model to pay more attention to channel features with high information content.

[0057] It first performs a Squeeze operation on the features obtained by convolution, that is, the input feature W×H×C (W refers to the width of the feature, H refers to the height of the feature, and C refers to the number of channels of the feature) is compressed to 1×1×C through global average pooling, and the information on one channel is compressed into a global feature, and a total of C global features are obtained. Then, the global features are subjected to an Excitation operation, and the features obtained after the Squeeze operation are subjected to a fully connected layer, Relu, a fully connected layer, and a sigmoid function operation, where the expression of the sigmoid function is

[0058]

[0059] This function can convert the binary classification value into a probability distribution between [0,1] and with a sum of 1. After the Excitation operation, the number of channels of the feature is compressed from C to C / r (the best effect is when r is 16) and then expanded to C. In this way, the weight coefficients of different channels can be obtained, and then the channel coefficients are multiplied by all the elements on the channel corresponding to the original feature to finally get the final feature.

[0060] A large number of redundant features will be generated in the convolution operation. The previous operation was to delete the redundant information as much as possible, but GhostNet artificially creates a large amount of redundant information, and the final model effect is better.

[0061] GhostNet first generates features through convolution operations, then performs a series of linear operations on each channel of the features to artificially generate redundant information, and finally concats this redundant information with the previous features, that is, stacking the two parts of the features together, so that a large number of features can be generated with only a small amount of calculation.

[0062] By fusing the features of the two, the network can obtain a large number of features and give higher weights to important features, which can effectively improve the effect of the model.

[0063] like Figure 4 As shown, this embodiment provides an image classification method, which includes:

[0064] Convert the measured data into a three-component magnetic field curve graph;

[0065] Send the three-component magnetic field curve graph to the classifier 1;

[0066] Mark the identified defects and find out their locations in the actual pipeline;

[0067] The image of the identified stress concentration area is sent to the classifier 2;

[0068] Mark the pipe sections with severe stress risk levels on the picture and find the actual location in the pipeline to replace them immediately.

[0069] Mark the pipe sections with medium stress hazard levels on the picture and at their actual locations in the pipeline and check them regularly.

[0070] After adopting the above method, the present invention has the following advantages over the prior art: the present invention detects the magnetic field intensity and converts it into a three-component magnetic field curve diagram, and adopts an artificial intelligence method to replace manual judgment, so as to more accurately and quickly identify pipeline defects, and realize early diagnosis of defects by judging the stress risk degree of the pipe section.

[0071] The above are preferred embodiments of the present invention, which do not limit the present invention. Those skilled in the art may make changes without departing from the technical scope of the present invention, but these changes are also within the protection scope of the present invention.

Claims

1. A buried pipeline detection method based on three-component magnetic field curve image recognition, characterized in that: The following steps are involved: S1, obtaining detection data of buried pipelines based on non-contact metal magnetic memory detection; S2, converting the data measured in S1 into a three-component magnetic field curve image, and classifying them into three categories according to defects, stress concentration areas, and normal conditions, establishing a database 1, training the database 1 using a convolutional neural network, and generating a classifier 1; S3, classify the stress danger degree of the pipe section according to the intensity of the curve change of the stress concentration area image in database 1, and divide it into three categories: severe, medium, and light, establish database 2, use convolutional neural network to train database 2, and generate classifier 2; S4, during actual detection, the newly detected three-component magnetic field curve image is sent to classifier 1 for classification, and the identified stress concentration area curve map is sent to classifier 2 to determine the stress danger level of the pipe section during actual detection; The convolutional neural network training adopts a convolutional neural network model for training, and the convolutional neural network model includes six convolutional layers and four fully connected layers. An attention GhostNet module and a maximum pooling layer of size 2×2 are added after each convolutional layer; The attention GhostNet module consists of a SENet and a GhostNet through a Concat function.

2. The buried pipeline detection method based on three-component magnetic field curve image recognition according to claim 1 is characterized in that: The method for detecting buried pipelines based on non-contact metal magnetic memory in S1 includes: S11, determine the exact location and direction of the pipeline, mark it on the ground, and ensure data traceability; S12, clear the ground obstacles and record the location of the metal objects that cannot be cleared; S13, using a non-contact scanning detection device connected to an encoder, running at a constant speed above the pipeline, collecting data and recording the detection length, and connecting to a GPS positioning system to record location information; S14, save the data every 100-200 meters of detection, and repeat until all buried pipelines are detected.

3. The buried pipeline detection method based on three-component magnetic field curve image recognition according to claim 1 is characterized in that: The database 1 in S2 converts the data measured in S1 into a three-component magnetic field curve image, which is divided into three categories: defect, stress concentration area, and normal through excavation verification and repeated comparison.

4. The buried pipeline detection method based on three-component magnetic field curve image recognition according to claim 1 is characterized in that: The SENet first performs a Squeeze operation on the features obtained by convolution to obtain global features, then performs an Excitation operation on the global features to obtain weights of different channels, multiply the weights by the original features, and finally obtain the final features.

5. The buried pipeline detection method based on three-component magnetic field curve image recognition according to claim 1 is characterized in that: The GhostNet first generates features through convolution operations, then performs a series of prior operations on each channel of the features to artificially generate redundant information, and finally concats this redundant information with the previous features.

6. The buried pipeline detection method based on three-component magnetic field curve image recognition according to claim 1 is characterized in that: The stress hazard levels of the pipe sections are: severe: the pipeline is severely damaged and should be replaced immediately; moderate: it can be used, but regular inspections are required; light: it can be used safely.

Images