Ultra-light natural gas pipeline weld defect detection method and system

Abstract

The invention provides an ultralight natural gas pipeline weld defect detection method and system, and belongs to the field of natural gas pipeline detection. In order to solve the problem of overfitting caused by category imbalance of a weld defect data set; small targets are difficult to effectively identify, so that the accuracy is low and the missing report / false report rate is high; and the model parameter quantity is large and cannot be deployed on edge detection equipment. According to the method, a public data set is adopted, the data set is preprocessed, and a data enhancement module is added; and an MSLE-YOLO network model is constructed. According to the method, the accuracy and robustness of natural gas pipeline weld defect detection can be improved, meanwhile, the missing report rate and the false report rate of diagnosis are greatly reduced, the parameter quantity and the calculation complexity of the model are further reduced, and subsequent deployment to edge detection equipment is facilitated.

Description

Technical Field

[0001] The present invention relates to the technical field of natural gas pipeline detection, and in particular to an ultra-lightweight natural gas pipeline weld defect detection method and system. Background Art

[0002] In natural gas pipelines, welds, as the connecting parts of the pipelines, are subject to great internal and external pressures. Defects in welds may affect the fatigue strength and corrosion resistance of welding materials, which directly affect the safety and operational stability of the pipelines. In order to avoid major accidents such as natural gas leakage caused by weld defects, resulting in property losses and casualties, it is extremely important to detect weld defects. Traditional weld defect detection is generally carried out through non-destructive testing methods, including X-rays, traditional ultrasonic waves, TOFD ultrasonic diffraction time difference method, eddy current and magnetic powder. Although these methods can accurately identify multiple types of defects in welds, their identification process still relies on manual evaluation, so there are cases of missed judgments and misjudgments.

[0003] With the continuous development of computer vision technology, deep learning-based object detection technology has been widely used in the field of weld defect detection. Deep learning models automatically characterize and learn high-dimensional discriminant features of welds through deep neural networks, which can effectively eliminate the influence of subjective bias on detection results, thereby significantly improving the efficiency and accuracy of detection. Although deep learning has made important progress in weld defect detection, it still faces the following problems: (1) There is often a class imbalance problem in weld defect datasets, that is, the difference in the number of normal welds and defective welds leads to insufficient learning of deep learning models on defect categories, affecting the ability to identify rare or small defects; (2) Weld defects are usually manifested as tiny targets, such as microcracks and pores. These small targets are often difficult to clearly identify due to low resolution and background noise. Therefore, existing weld defect detection models often have a problem of reduced accuracy when processing small targets; (3) Existing weld defect detection models are usually based on deep convolutional neural networks or large-scale pre-trained models, which require a lot of computing resources and storage space, and are difficult to run efficiently on resource-limited edge devices, thus limiting the practical application of real-time detection. Summary of the invention

[0004] The technical problems to be solved by the present invention are: In order to solve the problems of imbalanced categories in weld defect datasets, which leads to overfitting; it is difficult to effectively identify small targets, resulting in lower accuracy and higher omission / false alarm rates; and the model parameters are too large to be deployed on edge detection equipment.

[0005] The present invention adopts the following technical solutions to solve the above technical problems: The present invention provides an ultra-lightweight natural gas pipeline weld defect detection method, comprising the following steps: S100, build a data set, obtain the public data set steeltube, perform random data enhancement on the data set to complete the expansion, and then process the data set through image enhancement; S200, build an ultra-lightweight network model MSLE-YOLO, including building a multi-semantic difference mitigation module SCSA based on a progressive channel self-attention mechanism, building a four-head enhanced fast pyramid architecture network EFPAN with an improved P2 small target layer, and using the Inner_CIoU loss function as the network optimization target; S300, using the ultra-lightweight network model MSLE-YOLO constructed in step S200 for training. During the training process, the model is regularly evaluated using a validation set, and the hyperparameters of the model are adjusted to ensure the stability of the training process and prevent overfitting. The trained ultra-lightweight network model MSLE-YOLO is used to detect weld defects in natural gas pipelines.

[0006] Furthermore, in step S100, the random data enhancement method includes horizontal flipping, vertical flipping, random rotation, random translation and scaling.

[0007] Furthermore, in step S100, the image enhancement method includes blur effect, median blur and limited contrast adaptive histogram equalization.

[0008] Further, in step S200, it includes: S210, construct a multi-semantic difference mitigation module SCSA based on the progressive channel self-attention mechanism, which processes the feature dependency of space and channel through the synergy of spatial attention and channel attention; The invention comprises using the Mobilenetv3 BLOCK module to construct a backbone network, including two 1*1 point-by-point convolutions and one K*K depth convolution; wherein the Mobilenetv3 BLOCK module is set to an inverted bottleneck structure, and the number of channels is expanded to enhance the feature extraction capability by combining the SCSA attention module; wherein the SCSA module extracts multi-semantic spatial information through shared multi-semantic spatial attention SMSA, spatial decoupling and lightweight convolution to generate a spatial attention map; then the progressive channel self-attention PCSA spatial prior information is used to calculate the channel self-attention, and the effective channel attention map is generated by gradual compression and convolution; The calculation formula is as follows: SMSA eigendecomposition: ; ; SMSA feature extraction: ; ; PCSA attention weight calculation: ; The overall calculation process of the Mobilenetv3 BLOCK_SCSA module is: ; Where T represents the input feature map, represents the i-th sub-feature, H represents the height dimension, W represents the width dimension, C represents the number of channels, K represents the number of sub-features, Attn represents the attention mechanism function, GN represents group normalization, Concat is used to concatenate feature maps of different dimensions, Sigmoid and Softmax represent activation functions, Q represents the query vector, K represents the key vector, V represents the value vector, D represents the scaling factor, Conv represents point convolution, and DW represents depthwise separable convolution.

[0009] Furthermore, in step S200, it also includes: S220. Design an improved four-head enhanced fast pyramid architecture network EFPAN with a P2 small target layer, and add an ultra-lightweight dynamic upsampling module DySample to the EFPAN framework to construct an enhanced fast pyramid architecture DyS-EFPAN with dynamic upsampling.

[0010] Furthermore, in step S200, it also includes: S230, using the Inner_CIoU loss function as the optimization target of the network. The Inner_CIoU loss function calculates the IoU intersection-over-union loss by introducing an auxiliary bounding box, where the size of the auxiliary box is controlled by a scale factor. The calculation formula is as follows: ; ; ; ; ; ; in, represents the Inner_CIoU loss function, and b represent the true box and the predicted box respectively, r and l Respectively represent the right and left boundaries of the box; t and b Represent the top and bottom boundaries of the box, respectively.in Indicates the overlap between the real box and the predicted box. un Represents the total coverage area of ​​the real box and the predicted box, ratio represents the scale factor, and Represent the height and width of the real frame respectively, h and w Represent the height and width of the prediction box respectively, p Represents the distance between the center point of the anchor box and the center point of the prediction box, c represents the diagonal length of the minimum bounding rectangle between the anchor box and the true box, a represents a positive trade-off parameter used to balance the weights of different loss terms, v Used to measure the consistency of the aspect ratio between the predicted box and the true box.

[0011] An ultra-lightweight natural gas pipeline weld defect detection system is provided. The system has a program module corresponding to the above steps and executes the steps in the above ultra-lightweight natural gas pipeline weld defect detection method when running.

[0012] A computer-readable storage medium stores a computer program, wherein the computer program is configured to implement the steps of an ultra-lightweight natural gas pipeline weld defect detection method when called by a processor.

[0013] Compared with the prior art, the present invention has the following beneficial effects: (1) The present invention designs the MSLE-YOLO model, which effectively improves the detection accuracy of weld defects by introducing the Mobilenetv3 BLOCK_SCSA module, the P2 small target layer and the Inner_CIoU loss function, especially in the detection of small targets, and can detect tiny weld defects more accurately.

[0014] (2) The present invention designs an ultra-lightweight network architecture, including a Mobilenetv3 BLOCK module with an inverted bottleneck structure and a channel expansion strategy, and an EFPAN enhanced fast pyramid architecture with fewer convolutional layers and a lower number of channels combined with a DySample dynamic upsampling module; compared with the existing YOLOv8 model, the number of parameters of MSLE-YOLO is reduced by 56.3% and the GFLOPS is reduced by 18.5%; therefore, the present invention can run efficiently on edge devices with limited computing resources; this enables real-time application of weld defect detection on natural gas pipeline sites, which not only improves detection efficiency but also reduces hardware requirements, and has strong practicality and commercial value.

[0015] (3) The present invention accelerates the convergence speed of the network and improves the positioning accuracy by optimizing the loss function (Inner_CIoU). Compared with the traditional CIoU loss function, Inner_CIoU performs better when processing targets of different scales, effectively improving the training efficiency and the robustness of the network.

[0016] (4) The present invention adopts multi-level data enhancement technology, including blur effect, median blur and CLAHE, to improve the adaptability of the model under different environmental conditions; in addition, experimental results show that MSLE-YOLO performs well in both precision and recall, effectively reducing missed detection and false detection, and ensuring the reliability of the detection results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 These are eight types of steel pipe weld defect images in the public data set steeltube in the embodiment of the present invention, where (a) is air-hole, (b) is bite-edge, (c) is broken arc, (d) is crack, (e) is overlap, (f) is slag inclusion, (g) is unfused, and (h) is hollow bead. Figure 2 : is a block diagram of the MSLE-YOLO network structure in an embodiment of the present invention, wherein Backbone represents the backbone, Neck represents the neck, Head represents the head, mn_conv represents point-by-point convolution, SPPF represents the pyramid pooling operation, Concat represents the feature map used to splice different scales, Detect represents the detection head, INPUT represents the input image, and image enhance represents image enhancement; Figure 3 It is a structural diagram of the Mobilenetv3 BLOCK_SCSA module in an embodiment of the present invention, wherein MN Conv represents point-by-point convolution, DW Conv represents depth-wise convolution, AvgPool represents average pooling, Group Norm represents group normalization, Sigmoid represents activation function, Q, K, and V represent query, key, and value, respectively, which are used to calculate attention weights, CA-SHSA represents channel-level single-head self-attention mechanism, and ○ represents element-by-element multiplication; Figure 4 A structural diagram of an enhanced fast pyramid architecture (EFPAN) in an embodiment of the present invention; Figure 5is a flow chart of the DySample dynamic upsampling module in an embodiment of the present invention, wherein linear represents a linear layer, 0.25 represents an adjustable range factor, pixel shuffle represents pixel rearrangement, O represents an offset, g represents an original grid, and grid sample represents grid sampling; Figure 6 : This is a comparison chart of standard evaluation indicators of the MSLE-YOLO model and the YOLOv8 model in eight categories of steel pipe weld defects in an embodiment of the present invention, where (a) is the PR (precision-recall) curve of MSLE-YOLO, and (b) is the PR (precision-recall) curve of YOLOv8. In (a) and (b), the area under the PR curve is AP (mean average precision), and mAP is the average value of AP of all categories. DETAILED DESCRIPTION

[0018] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0019] Specific implementation plan 1: Combine Figures 1 to 5 As shown, the present invention provides an ultra-lightweight natural gas pipeline weld defect detection method, comprising the following steps: S100, construct a data set, obtain a public data set, perform random data enhancement on the data set to complete the expansion, and then process the data set through image enhancement, specifically including: Get the public dataset steeltube, combined with Figure 1 As shown in the figure, the dataset contains 3408 pictures of steel pipe weld defects in eight different categories. In order to avoid the overfitting problem caused by the imbalance of the weld defect dataset categories, the dataset is randomly enhanced. The data enhancement method includes horizontal flipping, vertical flipping, random rotation, random translation and scaling, and then the dataset is expanded to 4000 pictures (500 pictures per category). Then, multi-level image enhancement techniques are used, including blur effect, median blur and CLAHE (Contrast-Limited Adaptive Histogram Equalization) to process the training data. Apply blur effects to simulate the slight blur that occurs in images during actual inspections, including out-of-focus images caused by camera motion or ambient lighting changes; Apply median blur to effectively remove salt and pepper noise and small spots in the image, improve image quality, and enhance the model's robustness to noisy images; Applying limited contrast adaptive histogram equalization to enhance the local contrast of the image, making the weld defect features more obvious, and improving the detection ability of the model under different brightness and contrast conditions; After completing the above data set processing, the 4000 weld defect images were randomly redistributed in the ratio of 6:2:2 among the training set, validation set, and test set; S200, build the MSLE-YOLO (super lightweight network) model, including building a multi-semantic difference mitigation module (SCSA) based on the progressive channel self-attention mechanism, building a four-head enhanced fast pyramid architecture network (EFPAN) with improved P2 small target layer, and using the Inner_CIoU loss function as the optimization target of the network, including: S210, construct a multi-semantic difference mitigation module (SCSA) based on the progressive channel self-attention mechanism, which processes the feature dependencies of space and channel through the synergy of spatial attention and channel attention; In this process, the Mobilenetv3 BLOCK module is used to construct the backbone network, including two 1*1 point-by-point convolutions and one K*K depth convolution; the Mobilenetv3 BLOCK module is set to an inverted bottleneck structure, and the number of channels is expanded to enhance the feature extraction capability by combining the SCSA attention module; the SCSA module extracts multi-semantic spatial information through shared multi-semantic spatial attention SMSA, spatial decoupling and lightweight convolution to generate a spatial attention map; then the progressive channel self-attention PCSA spatial prior information is used to calculate the channel self-attention, and the effective channel attention map is generated by gradual compression and convolution; the calculation formula is as follows: SMSA eigendecomposition: ; ; SMSA feature extraction: ; ; PCSA attention weight calculation: ; The overall calculation process of the Mobilenetv3 BLOCK_SCSA module is: ; Among them, T is the input feature map, represents the i-th sub-feature, H represents the height dimension, W represents the width dimension, C represents the number of channels, K represents the number of sub-features, which is 4 in the present invention, Attn represents the attention mechanism function, GN represents the group normalization, Concat is used to concatenate feature maps of different dimensions, Sigmoid and Softmax are activation functions, Q represents the query vector, K represents the key vector, V represents the value vector, D represents the scaling factor, Conv represents the point convolution, and DW represents the depthwise separable convolution; S220, design and add P2 small target layer to improve the four-head enhanced fast pyramid architecture network (EFPAN), combined with Figure 4 As shown in the figure, the network has fewer convolutional layers and a lower number of channels, which can achieve more effective semantic information sharing. At the same time, the ultra-lightweight dynamic upsampling module DySample is added to the EFPAN framework to construct DyS-EFPAN (Enhanced Fast Pyramid Architecture with Dynamic Upsampling). This network can reduce the computational burden while retaining the necessary feature learning capabilities. S230, combined Figure 5 As shown in the figure, the Inner_CIoU loss function is used as the optimization target of the network. The Inner_CIoU loss function calculates the IoU (Intersection over Union) loss by introducing an auxiliary bounding box, where the size of the auxiliary box is controlled by the scale factor; the size of the auxiliary box will be dynamically adjusted according to different IoU samples, thereby accelerating the convergence process of the network and improving the overall network performance; the calculation formula is as follows: ; ; ; ; ; ; in, represents the Inner_CIoU loss function, and b represent the true box (GT) and the predicted box, respectively, r and l represent the right and left boundaries of the box, respectively; t and b represent the top and bottom boundaries of the box, respectively, in represents the overlapping part of the true box and the predicted box, un represents the total coverage area of ​​the true box and the predicted box, and ratio is the scaling factor, which is set to 1.2 in this experiment; and Represent the height and width of the real box, h and w represent the height and width of the predicted box, p represents the distance between the center point of the anchor box and the center point of the predicted box, c represents the diagonal length of the minimum circumscribed rectangle between the anchor box and the real box, a represents a positive trade-off parameter used to balance the weights of different loss items, and v is used to measure the consistency of the aspect ratio between the predicted box and the real box; S300, using the ultra-lightweight network model MSLE-YOLO constructed in step S200 for training. During the training process, the validation set is used to regularly evaluate the model and adjust the model's hyperparameters to ensure the stability of the training process and prevent overfitting. The trained ultra-lightweight network model MSLE-YOLO is used for training to detect weld defects in natural gas pipelines.

[0020] Specific implementation scheme 2: The present invention provides an ultra-lightweight natural gas pipeline weld defect detection system, which has a program module corresponding to the above steps and executes the steps in the above ultra-lightweight natural gas pipeline weld defect detection method when running.

[0021] The other combinations and connection relationships of this embodiment are the same as those of the first embodiment.

[0022] Specific implementation scheme three: The present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of an ultra-lightweight natural gas pipeline weld defect detection method when called by a processor.

[0023] The other combinations and connection relationships of this embodiment are the same as those of the first embodiment.

[0024] Simulation experiment Based on the data set obtained in step S100, the constructed MSLE-YOLO model is trained according to the following path: Build the experimental environment. The experimental platform is Windows 11 operating system, CPU i5-12490f, graphics card is NVIDIA RTX 4060TI, 8g video memory. Environment configuration: Python version is 3.9.18, torch version is 2.2.1, cuda version is 11.8; Initialize network parameters, including, Input image size: 640×640 Batch size: 27 images are input each time to ensure that the model can handle enough batch information; Optimizer: The stochastic gradient descent (SGD) optimizer is used, the initial learning rate is set to 0.01, and the learning rate is decayed.

[0025] Training rounds: set to 300 rounds to ensure that the model can be fully trained; During the training process, the validation set is used to evaluate the model and adjust the model's hyperparameters (learning rate, batch size, weight decay coefficient, optimizer type, training rounds) to ensure the stability of the training process and prevent overfitting; The performance of MSLE-YOLO on weld defect detection tasks is evaluated by comparing it with existing models through standard evaluation indicators, including mAP, number of parameters, and GFLOPS, and the effectiveness of each module in the model is demonstrated through ablation experiments. After the model training is completed, the model is evaluated using the test set, and the performance is evaluated using the following indicators: mAP@0.5 (mean average precision): measures the average accuracy of the model over multiple categories.

[0026] Precision and recall: used to evaluate the model's ability to identify defects.

[0027] GFLOPS: Measures the computational complexity of the model and evaluates its running efficiency on edge devices.

[0028] Combination Figure 6 As shown in the figure, the experimental results show that the mAP@0.5 of MSLE-YOLO in weld defect detection reaches 89.9%, which is 3.5% higher than YOLOv8. At the same time, MSLE-YOLO has also significantly optimized the computational complexity, with a GFLOPS of 6.6, which is 18.5% lower than YOLOv8. Combined with the following Table 1, the contribution of each module to the model performance is verified through ablation experiments. The experimental results show that the Mobilenetv3 BLOCK_SCSA module, EFPAN architecture, DySample upsampling module, and Inner_CIoU loss function module have improved the accuracy and efficiency of the model to varying degrees.

[0029]

[0030] Table 1 Ablation experiment results of each module Although the present invention is disclosed as above, the protection scope of the present invention is not limited thereto. Those skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention, and these changes and modifications will fall within the protection scope of the present invention.

Claims

1. A method for detecting defects in ultra-lightweight natural gas pipeline welds, characterized in that: The following steps are involved: S100, build a data set, obtain the public data set steeltube, perform random data enhancement on the data set to complete the expansion, and then process the data set through image enhancement; S200, build an ultra-lightweight network model MSLE-YOLO, including building a multi-semantic difference mitigation module SCSA based on a progressive channel self-attention mechanism, building a four-head enhanced fast pyramid architecture network EFPAN with an improved P2 small target layer, and using the Inner_CIoU loss function as the network optimization target; S300, using the ultra-lightweight network model MSLE-YOLO constructed in step S200 for training. During the training process, the model is regularly evaluated using a validation set, and the hyperparameters of the model are adjusted to ensure the stability of the training process and prevent overfitting. The trained ultra-lightweight network model MSLE-YOLO is used to detect weld defects in natural gas pipelines.

2. The method for detecting weld defects of an ultra-lightweight natural gas pipeline according to claim 1, characterized in that: In step S100 , the random data enhancement method includes horizontal flipping, vertical flipping, random rotation, random translation and scaling.

3. The method for detecting weld defects of an ultra-lightweight natural gas pipeline according to claim 1, characterized in that: In step S100 , the image enhancement method includes blur effect, median blur and limited contrast adaptive histogram equalization.

4. The method for detecting weld defects of an ultra-lightweight natural gas pipeline according to claim 1, characterized in that: In step S200, it includes: S210, construct a multi-semantic difference mitigation module SCSA based on the progressive channel self-attention mechanism, which processes the feature dependency of space and channel through the synergy of spatial attention and channel attention; The invention comprises using the Mobilenetv3 BLOCK module to construct a backbone network, including two 1*1 point-by-point convolutions and one K*K depth convolution; wherein the Mobilenetv3 BLOCK module is set to an inverted bottleneck structure, and the number of channels is expanded to enhance the feature extraction capability by combining the SCSA attention module; wherein the SCSA module extracts multi-semantic spatial information through shared multi-semantic spatial attention SMSA, spatial decoupling and lightweight convolution to generate a spatial attention map; then the progressive channel self-attention PCSA spatial prior information is used to calculate the channel self-attention, and the effective channel attention map is generated by gradual compression and convolution; The calculation formula is as follows: SMSA eigendecomposition: ； ； SMSA feature extraction: ； ； PCSA attention weight calculation: ； The overall calculation process of the Mobilenetv3 BLOCK_SCSA module is: ； Where T represents the input feature map, represents the i-th sub-feature, H represents the height dimension, W represents the width dimension, C represents the number of channels, K represents the number of sub-features, Attn represents the attention mechanism function, GN represents group normalization, Concat is used to concatenate feature maps of different dimensions, Sigmoid and Softmax represent activation functions, Q represents the query vector, K represents the key vector, V represents the value vector, D represents the scaling factor, Conv represents point convolution, and DW represents depthwise separable convolution.

5. The method for detecting weld defects of an ultra-lightweight natural gas pipeline according to claim 4, characterized in that: In step S200, it also includes: S220. Design an improved four-head enhanced fast pyramid architecture network EFPAN with the addition of the P2 small target layer, and add an ultra-lightweight dynamic upsampling module DySample to the EFPAN framework to construct an enhanced fast pyramid architecture DyS-EFPAN with dynamic upsampling.

6. The method for detecting weld defects of an ultra-lightweight natural gas pipeline according to claim 5, characterized in that: In step S200, it also includes: S230, using the Inner_CIoU loss function as the optimization target of the network. The Inner_CIoU loss function calculates the IoU intersection-over-union loss by introducing an auxiliary bounding box, where the size of the auxiliary box is controlled by a scale factor. The calculation formula is as follows: ； ； ； ； ； ； in, represents the Inner_CIoU loss function, and b represent the true box and the predicted box respectively, r and l Respectively represent the right and left boundaries of the box; t and b Represent the top and bottom boundaries of the box, respectively. in Indicates the overlap between the real box and the predicted box. un Represents the total coverage area of ​​the real box and the predicted box, ratio represents the scale factor, and Represent the height and width of the real frame respectively, h and w Represent the height and width of the prediction box respectively, p Represents the distance between the center point of the anchor box and the center point of the prediction box, c represents the diagonal length of the minimum bounding rectangle between the anchor box and the true box, a represents a positive trade-off parameter used to balance the weights of different loss terms, v Used to measure the consistency of the aspect ratio between the predicted box and the true box.

7. An ultra-lightweight natural gas pipeline weld defect detection system, characterized by: The system has a program module corresponding to the steps described in any one of claims 1 to 6, and executes the steps in the above-mentioned ultra-lightweight natural gas pipeline weld defect detection method when running.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is configured to implement the steps of the ultra-lightweight natural gas pipeline weld defect detection method according to any one of claims 1 to 6 when called by a processor.

Images