Steel pipe surface defect detection method driven by dual-tree complex wavelet multi-layer information fusion

Abstract

This invention discloses a method for detecting surface defects in seamless steel pipes driven by dual-tree complex wavelet multilayer information fusion, comprising the following steps: S1: acquiring a sample database of surface defects in seamless steel pipes; S2: establishing an intelligent detection network for surface defects in seamless steel pipes; S3: training the intelligent detection network using the sample database obtained in step S1 combined with dual-tree complex wavelets to obtain a surface defect detection network model; S4: for any acquired surface defect image of a seamless steel pipe to be tested, inputting it into the surface defect detection network model obtained in step 3. If a defect exists in the image to be tested, the surface defect detection network model marks the corresponding defect location in the image to be tested. This invention uses surface images to directly perform intelligent identification of surface defects in seamless steel pipes, featuring automatic detection of surface defects in seamless steel pipes and high detection efficiency.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion. Background Technology

[0002] Due to their superior performance and weld-free characteristics, seamless steel pipes have been widely used in critical sectors such as oil and gas. However, how to quickly and effectively inspect the surface condition of seamless steel pipes has become a pressing issue for many seamless steel pipe manufacturers. Currently, many manufacturers rely on manual visual inspection for surface defect detection; however, this method cannot guarantee long-term consistency in inspection quality and requires further in-depth research. Summary of the Invention

[0003] The purpose of this invention is to provide a method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion. This invention can automatically detect surface defects in seamless steel pipes, and features high detection efficiency and accurate detection results.

[0004] The technical solution of this invention: a method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion, comprising the following steps:

[0005] S1: Obtain a sample database of surface defects in seamless steel pipes;

[0006] S2: An intelligent detection network for surface defects in seamless steel pipes is established based on dual-tree complex wavelets;

[0007] S3: Use the sample database obtained in step S1 and the dual-tree complex wavelet to train the intelligent detection network to obtain the surface defect detection network model.

[0008] S4: For any obtained seamless steel pipe surface defect test image, input it into the surface defect detection network model obtained in step 3. If there is a defect in the test image, the surface defect detection network model marks the corresponding defect position in the test image.

[0009] The above-mentioned dual-tree complex wavelet multi-layer information fusion driven steel pipe surface defect detection method, in step S1, establishes the sample database as follows:

[0010] S11: Fix the seamless steel pipe to be tested on the image acquisition platform and acquire a surface image for training on the surface defects of the seamless steel pipe;

[0011] S12: Convert the RGB of the surface image to YUV space;

[0012] S13: Set the Y channel in the YUV space to zero and convert it back to an RGB image;

[0013] S14: Convert the RGB image from step 13 back to a grayscale image;

[0014] S15: Mark the location of defects in the surface image pixel by pixel in the grayscale image to obtain a sample defect database.

[0015] The aforementioned method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion, in step S12, uses the following method for YUV space conversion:

[0016]

[0017] R, G, and B represent the RGB three-channel information of the surface image, while Y, U, and V represent the channel information converted to YUV space.

[0018] In the aforementioned dual-tree complex wavelet multi-layer information fusion-driven steel pipe surface defect detection method, the RGB image conversion method in step S13 is as follows:

[0019]

[0020] R, G, and B represent the information of the three channels of the converted RGB image, while U and V represent the information of the two channels in the YUV space.

[0021] In the aforementioned steel pipe surface defect detection method driven by dual-tree complex wavelet multi-layer information fusion, in step S2, the intelligent detection network is provided with a CBAM-U-Net block composed of a CBAM module and a U-Net network; the CBAM module has a channel attention mechanism and a spatial attention mechanism, and the CBAM module enhances the input features through the channel attention mechanism and the spatial attention mechanism, and the enhanced features are passed to the U-Net network for the extraction of the feature state of this layer.

[0022] The aforementioned method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion, wherein the method for establishing the surface defect detection network model is as follows:

[0023] S21: Perform dual-tree complex wavelet decomposition on the input grayscale image to obtain the components of each layer of the dual-tree complex wavelet;

[0024] S22: Transfer the components of each layer of the dual-tree complex wavelet to the CBAM-U-Net block respectively;

[0025] S23: The results of each layer of CBAM-U-Net are summed to form a seamless steel pipe surface defect neural network, which is used as an intelligent detection network.

[0026] Compared with existing technologies, this invention acquires a sample database of surface defects in seamless steel pipes and then establishes an intelligent detection network for these defects based on dual-tree complex wavelets. The intelligent detection network is then trained using the acquired sample database to obtain a surface defect detection network model. This model can detect surface defects in any acquired image of a seamless steel pipe. If a defect is found in the image, the model marks the corresponding defect location. Therefore, this invention can directly identify surface defects in seamless steel pipes using surface images, avoiding the inaccuracies of manual visual inspection. It is also simple, convenient, efficient, and provides accurate results. Attached Figure Description

[0027] Figure 1 This is a flowchart of the present invention;

[0028] Figure 2 This is the CBAM structure used in this invention;

[0029] Figure 3 This is a schematic diagram of the surface defect detection network model used in this invention;

[0030] Figure 4 This is the RGB image three-channel information of the image A to be tested in this invention;

[0031] Figure 5 This is the YUV spatial three-channel information of the image A to be tested in this invention;

[0032] Figure 6 It is the grayscale image of the image A to be tested in this invention after conversion;

[0033] Figure 7 This is the recognition result of the image A to be tested in this invention;

[0034] Figure 8 This is the RGB image three-channel information of the image B to be tested in this invention;

[0035] Figure 9 This is the YUV spatial three-channel information of the image B to be tested in this invention;

[0036] Figure 10 It is the grayscale image of the image B to be tested in this invention after conversion;

[0037] Figure 11 This is the recognition result of the image B to be tested in this invention. Detailed Implementation

[0038] The present invention will be further described below with reference to the accompanying drawings and embodiments, but this should not be construed as limiting the present invention.

[0039] Example: A method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion, the flowchart of which is as follows. Figure 1 As shown, it includes the following steps:

[0040] S1: Obtain a sample database of surface defects in seamless steel pipes. The sample database includes the defect locations and corresponding training images. The method for establishing the sample database is as follows:

[0041] S11: Fix the seamless steel pipe to be tested on the image acquisition platform and acquire a surface image for training on the surface defects of the seamless steel pipe;

[0042] S12: Convert the RGB of the surface image to YUV space; the conversion method is as follows:

[0043]

[0044] Where R, G, and B represent the information of the three channels of the RGB image, and Y, U, and V represent the channel information converted to the YUV space.

[0045] S13: Set the Y channel in the YUV space to zero and convert it back to an RGB image; the method for converting to an RGB image is as follows:

[0046]

[0047] Where R, G, and B are the information of the three channels of the converted RGB image, and U and V are the information of the two channels in the YUV space;

[0048] S14: Convert the RGB image from step 13 back to a grayscale image;

[0049] S15: Mark the location of defects in the surface image pixel by pixel in the grayscale image to obtain a sample defect database.

[0050] S2: Establish an intelligent detection network for surface defects in seamless steel pipes; the intelligent detection network includes a CBAM-U-Net block composed of CBAM modules and a U-Net network; such as Figure 2 As shown, the CBAM module has a channel attention mechanism and a spatial attention mechanism. The CBAM module enhances the input features through the channel attention mechanism and the spatial attention mechanism. The enhanced features are then passed to the U-Net network for the extraction of the feature state of this layer.

[0051] S3: Using the sample database obtained in step S1, the intelligent detection network is trained using a dual-tree complex wavelet to obtain a surface defect detection network model; such as Figure 3 As shown, the method for establishing the surface defect detection network model is as follows:

[0052] S21: Perform dual-tree complex wavelet decomposition on the input grayscale image to obtain the components of each layer of the dual-tree complex wavelet. The dual-tree complex wavelet decomposition uses a two-path filter bank with a binary tree structure for signal decomposition and reconstruction. The first tree generates the real part, and the second tree generates the imaginary part. The low-pass filters for the real and imaginary trees are designed appropriately to satisfy the half-sampling delay condition and exhibit approximate translation invariance. The filters of the two trees have the same sampling frequency, but the delay between them is exactly one sampling interval. This ensures that the decimation of the first layer in the imaginary tree precisely captures the sampled values ​​lost in the decimation of the real tree. This achieves translation invariance of the complex wavelet transform while avoiding excessive computation and offering the advantage of easy implementation.

[0053] S22: Transfer the components of each layer of the dual-tree complex wavelet to the CBAM-U-Net block respectively;

[0054] S23: The results of each layer of CBAM-U-Net are summed to form a seamless steel pipe surface defect neural network, which is used as an intelligent detection network.

[0055] S4: For any obtained seamless steel pipe surface defect test image, input it into the surface defect detection network model obtained in step 3. If there is a defect in the test image, the surface defect detection network model marks the corresponding defect position in the test image.

[0056] To verify the effectiveness of the present invention, the applicant used test image A and test image B for experiments. For test image A, its RGB image and YUV channel information are as follows: Figure 4 and Figure 5 As shown in the attached grayscale image. Figure 6 As shown; after passing through the surface defect detection network model, Figure 7 The location of the defect is marked. For image B to be tested, its RGB image and YUV channel information are shown in the attached figure. Figure 8 and Figure 9 As shown in the attached grayscale image. Figure 10 As shown; after passing through the surface defect detection network model, Figure 11 The location of the defect is marked in the text.

[0057] As can be seen, the present invention can directly identify surface defects of seamless steel pipes using surface images, avoiding the inaccuracies of manual visual identification. Moreover, it is simple, convenient, efficient, and produces accurate results.

Claims

1. A method for detecting surface defects in steel pipes driven by dual-tree complex wavelet multilayer information fusion, characterized in that: Includes the following steps: S1: Obtain a sample database of surface defects in seamless steel pipes; S2: Establish an intelligent detection network for surface defects in seamless steel pipes; S3: Use the sample database obtained in step S1 and the dual-tree complex wavelet to train the intelligent detection network to obtain the surface defect detection network model. S4: For any obtained seamless steel pipe surface defect test image, input it into the surface defect detection network model obtained in step 3. If there is a defect in the test image, the surface defect detection network model marks the corresponding defect position in the test image. In step S1, the method for establishing the sample database is as follows: S11: Fix the seamless steel pipe to be tested on the image acquisition platform and acquire a surface image for training on the surface defects of the seamless steel pipe; S12: Convert the RGB of the surface image to YUV space; S13: Set the Y channel in the YUV space to zero and convert it back to an RGB image; S14: Convert the RGB image from step 13 back to a grayscale image; S15: Mark the location of defects in the surface image pixel by pixel in the grayscale image to obtain a sample defect database; In step S2, the intelligent detection network includes a CBAM-U-Net block composed of a CBAM module and a U-Net network. The CBAM module has a channel attention mechanism and a spatial attention mechanism. The CBAM module enhances the input features through the channel attention mechanism and the spatial attention mechanism. The enhanced features are then passed to the U-Net network for the extraction of the feature state of this layer. The method for establishing the surface defect detection network model is as follows: S21: Perform dual-tree complex wavelet decomposition on the input grayscale image to obtain the components of each layer of the dual-tree complex wavelet; S22: Transfer the components of each layer of the dual-tree complex wavelet to the CBAM-U-Net block respectively; S23: The results of each layer of CBAM-U-Net are summed to form a seamless steel pipe surface defect neural network, which is used as an intelligent detection network.

2. The method for detecting surface defects of steel pipes driven by dual-tree complex wavelet multilayer information fusion according to claim 1, characterized in that: In step S12, the YUV space conversion method is as follows: ； R, G, and B represent the RGB three-channel information of the surface image, while Y, U, and V represent the channel information converted to YUV space.

3. The method for detecting surface defects of steel pipes driven by dual-tree complex wavelet multilayer information fusion according to claim 2, characterized in that: In step S13, the RGB image conversion method is as follows: ； R, G, and B represent the information of the three channels of the converted RGB image, while U and V represent the information of the two channels in the YUV space.

Images