Weld defect detection method and system fusing TOFD image waveform characteristics

Abstract

The invention provides a weld defect detection method and system fusing TOFD image waveform characteristics. The method comprises the following steps: carrying out adaptive noise reduction processing on TOFD image data and TOFD waveform data; performing feature extraction on the TOFD image data and the TOFD waveform data after adaptive noise reduction processing through a dual-network view angle to obtain image multi-scale features and waveform time sequence features; fusing the image multi-scale features and the waveform time sequence features by adopting a self-adaptive attention weighting mode to obtain fused features; and performing defect detection based on the fusion features. According to the method, film evaluation knowledge of'simultaneous analysis of wave maps' in the field of nondestructive TOFD detection is fused, TOFD image weld defect detection and identification are realized, and the accuracy and reliability of a defect detection result are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of nondestructive testing, and particularly relates to a welding defect detection method and system fusing TOFD image waveform features. BACKGROUND

[0002] The TOFD (Time of Flight Diffraction) technology is widely applied to welding defect detection due to its high defect detection rate and digital storage advantages. The TOFD detection images are large in quantity, and are currently mainly evaluated in an artificial manner, which is long in working cycle and high in labor intensity. Therefore, the intelligent defect recognition technology based on the TOFD images has become a research hotspot.

[0003] However, the existing welding defect recognition technology based on the TOFD images finds that, in the use process, the traditional intelligent method only relies on the TOFD image data, and does not combine the phase and amplitude change law of the defects in the waveform signal (for example, the phase of the defect endpoint diffraction wave is opposite to that of the straight-through wave / bottom wave), so that it is difficult to distinguish the defects from the noises. The existing deep learning model does not simulate the process of "wave chart simultaneous analysis" in the artificial evaluation, and cannot effectively utilize the physical characteristics (for example, the defect waveform amplitude is significantly higher than the noise) of the waveform data, so that the model lacks reliability. The parameter setting of the existing pretreatment method (for example, the standard Gaussian filter and the wavelet denoising) depends on the artificial experience, is not dynamically optimized according to the detection requirements, and is easy to cause noise interference or weaken the defect features due to the fixed parameters. In view of the above problems, there is an urgent need for a defect detection method fusing the waveform features, combining the field knowledge and being capable of self-adaptive optimization of the pretreatment parameters, so as to improve the defect detection precision and reliability. SUMMARY

[0004] In order to solve the problems of low detection accuracy and poor reliability of the existing welding defect recognition method and the problem of failing to fully fuse the field knowledge, the application provides a welding defect detection method and system fusing TOFD image waveform features.

[0005] To achieve the above-mentioned purpose, the application provides the following technical scheme: The application provides a welding defect detection method fusing TOFD image waveform features, which comprises the following steps: Adaptive denoising processing is performed on the TOFD image data and the TOFD waveform data; Feature extraction is performed on the TOFD image data and the TOFD waveform data subjected to the adaptive denoising processing through double network perspectives, so as to obtain image multi-scale features and waveform time sequence features; The image multi-scale features and the waveform time sequence features are fused by using an adaptive attention weighting mode, so as to obtain fusion features; Defect detection is performed based on the fusion features.

[0006] Preferably, the adaptive denoising processing of the TOFD image data and the TOFD waveform data comprises: Based on the constructed preprocessing parameter prediction network Adaptive preprocessing of the TOFD image data and the TOFD waveform data, prediction of Gaussian processing parameters and wavelet threshold filtering parameters, and scaling of the Gaussian processing parameters and the wavelet threshold filtering parameters; Based on the Gaussian processing parameters, the TOFD image data is denoised using a standard Gaussian filter preprocessing method; Based on the wavelet threshold filtering parameters, the TOFD waveform data is denoised using a wavelet threshold filtering WF preprocessing method; Preferably, the Gaussian processing parameters include kernel size And standard deviation The linear scaling process of the Gaussian processing parameters is:

[0007] Wherein, TOFD image data, Scaling factor, ; The wavelet threshold filtering parameters include decomposition layer number And threshold The linear scaling process of the wavelet threshold filtering parameters is:

[0008] Wherein, TOFD waveform data, Scaling factor, .

[0009] Preferably, the feature extraction of the TOFD image data and the TOFD waveform data after adaptive denoising processing is performed through a double network perspective, respectively, to obtain image multi-scale features and waveform time sequence features, comprising: Constructing a feature extraction network, i.e., an image feature extraction network and a waveform feature extraction network; Based on the image feature extraction network, the feature extraction of the TOFD image data after adaptive denoising processing is performed to obtain image multi-scale features; Based on the waveform feature extraction network, the feature extraction of the TOFD waveform data after adaptive denoising processing is performed to obtain waveform time sequence features.

[0010] Preferably, the process of performing feature extraction of the TOFD image data after adaptive denoising processing based on the image feature extraction network to obtain image multi-scale features is:

[0011] wherein, is the input adaptive noise reduction processed image data, is the extracted image multi-scale feature.

[0012] Preferably, the waveform feature extraction network based on the adaptive noise reduction processed TOFD waveform data is extracted, and the process of obtaining the waveform time sequence feature is:

[0013] wherein, is the input adaptive noise reduction processed waveform data, is the extracted waveform time sequence feature.

[0014] Preferably, the image multi-scale feature and the waveform time sequence feature are fused by using the adaptive attention weighting method to obtain the fusion feature, comprising: screening the image multi-scale feature, and retaining three layers of image multi-scale features; randomly extracting three layers of features in the waveform time sequence feature, and performing convolution processing to make the extracted three layers of features have the same dimension as the image multi-scale feature; splicing the corresponding feature maps of the waveform time sequence feature and the image multi-scale feature with the same dimension to obtain a feature representation, and then obtaining an attention map after convolution and activation ; splicing the corresponding feature maps of the waveform time sequence feature and the image multi-scale feature with the same dimension to obtain a feature representation, and then obtaining an attention map after convolution and activation and its complement respectively multiplied by two features Multipy to perform weighting, and the weighted features are concatenated to obtain the fusion feature .

[0015] The present application provides a kind of fusion TOFD image waveform feature weld defect detection system, applies the weld defect detection method of fusion TOFD image waveform feature described above, comprising: adaptive noise reduction processing module is configured to be used for adaptive noise reduction processing to TOFD image data and TOFD waveform data; feature extraction module is configured to be used for by double network perspective respectively to adaptive noise reduction processing after TOFD image data and TOFD waveform data are extracted, and obtain image multi-scale feature and waveform time sequence feature; feature fusion module is configured to be used for by adaptive attention weighting method to image multi-scale feature and waveform time sequence feature are fused, and obtain fusion feature; detection module is configured to be used for based on fusion feature and carries out defect detection.

[0016] The application provides an electronic device, including a memory, a processor and a computer program stored in the memory and executable in the processor, and the processor implements the steps of the weld defect detection method by fusing TOFD image waveform features when executing the computer program.

[0017] The application provides a computer readable storage medium, which stores a computer program, and the computer program implements the steps of the weld defect detection method by fusing TOFD image waveform features when executed by a processor.

[0018] Compared with the prior art, the application has the following beneficial technical effects: The application provides a weld defect detection method by fusing TOFD image waveform features, which reduces the influence of noise in TOFD data on defect detection performance through adaptive noise reduction processing of TOFD image data and TOFD waveform data; image multi-scale features and waveform time sequence features are extracted through double-network perspective feature extraction of the TOFD image data and the TOFD waveform data after adaptive noise reduction processing; the image multi-scale features and the waveform time sequence features are fused through an adaptive attention weighting method, and the fused features are used for defect detection, so that the method fuses the knowledge of "waveform-image simultaneous analysis" in the field of non-destructive TOFD detection, realizes TOFD image weld defect detection and recognition, and improves the accuracy and reliability of the defect detection result.

[0019] Further, in the adaptive preprocessing based on parameter self-learning, a parameter prediction network that can automatically learn and optimize according to the defect detection result is used to provide data enhancement parameters that are beneficial to defect detection for standard Gaussian filtering GF and wavelet threshold filtering WF preprocessing methods, and customized parameters are dynamically provided for each TOFD data.

[0020] Further, in the wave-image feature extraction based on double-network perspective, YOLOv8 and Transformer are used to extract features from TOFD images and waveforms in a way that is beneficial to capturing data information, and deeper and richer feature information is extracted from two perspectives.

[0021] Further, in the fusion of image multi-scale features and waveform time sequence features, an attention map that can adaptively and dynamically adjust weights is used to weight and fuse image and waveform features, so as to integrate the knowledge in the field of "waveform-image simultaneous analysis", and combine the phase, amplitude change law of the waveform with the texture, morphological features of the image, to provide comprehensive evaluation information for the final defect detection. BRIEF DESCRIPTION OF DRAWINGS

[0022] Figure 1 A flowchart of a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 2 A data processing flowchart of a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 3 A TOFD image data adaptive noise reduction processing process in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 4 A TOFD waveform data adaptive noise reduction processing process in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 5 An image feature extraction network based on a YOLOv8 backbone network in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 6 A waveform feature extraction network based on a Transformer encoder in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 7 A data processing block diagram of a wave pattern feature fusion model based on adaptive attention weighting in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 8 A data processing block diagram of an adaptive attention weighting model in a weld defect detection method fusing TOFD image waveform features according to the present application is shown in the figure. Figure 9 A schematic diagram of a computer device according to an embodiment of the present application is shown in the figure. Figure 10 A block diagram of a chip according to an embodiment of the present application is shown in the figure. DETAILED DESCRIPTION

[0023] In the following, only certain exemplary embodiments are simply described. As those skilled in the art can recognize, the described embodiments can be modified in various different ways without departing from the spirit or scope of the present application. Therefore, the drawings and the description are considered to be essentially exemplary rather than limiting.

[0024] The embodiments of the present application are described in detail below with reference to the accompanying drawings.

[0025] The application provides a weld defect detection method fusing TOFD image waveform features.

[0026] The application provides a weld defect detection method fusing TOFD image waveform features, as shown in Figure 1 and Figure 2 , comprising the following steps: S1, adaptively denoising TOFD image data and TOFD waveform data; In order to integrate the TOFD detection field knowledge of "waveform-image simultaneous analysis" into the deep learning model, first, the corresponding waveform data is obtained from the TOFD image data, and the TOFD waveform data is preprocessed by selecting a suitable algorithm to reduce the influence of noise on the detection performance. In the field of TOFD data preprocessing, there are a large number of methods (such as filtering and differential processing technology), and all of them have significant noise reduction effect. However, the current preprocessing method mainly selects parameters to improve the image quality, and does not consider the influence of the processing result on the defect detection performance, and even may have side effects.

[0027] Therefore, an adaptive preprocessing module based on parameter self-learning is constructed, the traditional TOFD data preprocessing method is combined with the parameter prediction network, the model automatically optimizes the parameters in the reverse direction according to the defect detection result, adaptive preprocessing is realized, and noise interference is reduced. In addition, since the preprocessing methods of the TOFD image data and the TOFD waveform data are different, the network is used for parameter prediction, as shown in Figure 3 and Figure 4 .

[0028] Since the parameter prediction network can automatically propagate in the reverse direction according to the loss function of the weld defect detection, and continuously optimize and learn, the input data can be adaptively enhanced in the direction conducive to defect detection. In order to preprocess the image data and the waveform data, the image and waveform dynamic prediction parameters are obtained, respectively, a preprocessing parameter prediction network is constructed, and the preprocessing parameter prediction network The first pre-processing parameter prediction network and the second pre-processing parameter prediction network The first pre-processing parameter prediction network and the second pre-processing parameter prediction network have the same network structure, both of which are composed of 2 convolutional layers and 1 fully connected layer, each convolutional layer includes a convolution kernel and a max-pooling operation, and the output of the final fully connected layer is adjusted to a specified range after passing through a Sigmoid activation function.

[0029] The specific calculation method is as follows:

[0030] In the formula, is the TOFD image data, and is a convolutional layer, and is a max-pooling layer, is a fully connected layer, is an activation function.

[0031] For the TOFD image data, a standard Gaussian filter GF pre-processing method is used for noise reduction processing. The Gaussian filter is a linear smoothing filter technology that can be used for image denoising, image smoothing, etc. It is a filter based on a Gaussian function, which realizes smoothing processing by weighted average of the pixel points around each pixel point in the image; in Gaussian filtering, for each pixel point, the pixel points around it are weighted and averaged by a Gaussian kernel; the Gaussian kernel is a two-dimensional function, which is similar in shape to a bell-shaped curve, with the center point having the largest weight, and the farther away from the center point, the smaller the weight, until it approaches 0. The size and standard deviation of the Gaussian kernel are two important parameters in Gaussian filtering, which determine the shape and size of the Gaussian kernel.

[0032] The Gaussian filter formula is as follows:

[0033] In the formula, is the weight value in the Gaussian kernel, is the relative position index of the filter, is the standard deviation of the Gaussian function; By adjusting the size and standard deviation of the Gaussian kernel, different degrees of smoothing effect of the image can be obtained while preserving the edge information of the image. Therefore, the parameter prediction network is responsible for predicting the Gaussian processing parameters for the standard Gaussian filter pre-processing method, including the kernel size and the standard deviation two parameters, and linearly scaling the Gaussian processing parameters so that , .

[0034] In the specific scaling process:

[0035] wherein, is the TOFD image data, is the scaling factor,

[0036] For TOFD waveform data, a wavelet threshold filtering (WF) preprocessing method is selected for noise reduction processing. Wavelet filtering is a widely used technology in the field of signal processing, and is particularly suitable for the analysis and processing of non-stationary signals. Compared with the traditional Fourier transform, wavelet transform provides a localized frequency domain analysis method, which can provide signal information in both time and frequency dimensions, which has a significant advantage for waveform data denoising and feature extraction.

[0037] Specifically, the process of wavelet filtering usually includes the following steps: Wavelet decomposition: first, the TOFD waveform data is subjected to wavelet transform, and the TOFD waveform data is decomposed into a plurality of coefficient sequences at different scales. These coefficients reflect the characteristics of the signal at different frequency bands. The calculation formula is as follows:

[0038] wherein, is the signal to be denoised of the TOFD waveform data, is the wavelet basis function, is the scaling function, is the jth layer of detail coefficients, is the Jth layer of approximation coefficients, and J is the number of decomposition layers.

[0039] Threshold denoising: according to the noise characteristics and signal characteristics, the wavelet coefficients obtained by decomposition are subjected to threshold processing. Common threshold processing methods include soft threshold and hard threshold. In this embodiment, a soft thresholding operation is selected, and the calculation formula is as follows:

[0040] wherein, is the detail coefficient after thresholding, is the threshold value.

[0041] Wavelet reconstruction: using the wavelet coefficients after threshold processing, inverse wavelet transform is performed to recover the denoised signal, and the calculation formula is as follows:

[0042] wherein, is the reconstructed denoised signal.

[0043] The appropriate wavelet decomposition layer and threshold value are crucial to the effect of the wavelet threshold filtering WF, and therefore the parameter prediction network is responsible for predicting the wavelet threshold filtering parameters for the wavelet threshold filtering WF, including the decomposition layer and the threshold value Two parameters, and linearly scaling the wavelet threshold filtering parameters so that , .

[0044] The process of linear scaling of the wavelet threshold filtering parameters is as follows:

[0045] wherein, is the TOFD waveform data, is the scaling factor, .

[0046] S2, respectively, through the dual network perspective, the TOFD image data and TOFD waveform data after adaptive noise reduction processing are subjected to feature extraction, obtaining image multi-scale features and waveform time sequence features; The TOFD image data and TOFD waveform data after adaptive noise reduction processing are used for defect detection, and rich feature information contained therein needs to be extracted. The image feature extraction network and the waveform feature extraction network based on the dual network perspective are constructed, as shown in Figure 5 and Figure 6 The image feature extraction network and the waveform feature extraction network respectively use different networks to extract features from two perspectives according to the characteristics of the TOFD image data and the TOFD waveform data after adaptive noise reduction processing, providing more comprehensive feature information.

[0047] The image feature extraction network adopts the backbone network of YOLOv8 for extraction. YOLOv8 is a highly integrated target detection framework, in which the backbone network plays a crucial role in effectively capturing different levels of features in images, from basic shapes and structures to complex edges and textures. The backbone network of YOLOv8 is usually composed of multiple stages, each responsible for extracting features at a specific level. These stages include but are not limited to the following components: CBS module: This module integrates Convolution (convolution layer), Batch Normalization (batch normalization layer), and SILU (Swish-like activation function). The main function of the CBS module is to perform convolution operations when processing input images or feature maps, thereby reducing spatial resolution while increasing the number of channels, thereby enhancing feature expression capabilities. C2f module: This module is an improved version of the C3 structure in YOLOv5, designed to address the problem of gradient vanishing in deep networks by providing more abundant gradient flow paths, thereby optimizing the learning process of features. In this way, the C2f module helps to improve the overall training effect and feature extraction capability of the network. SPPF module: This is a fixed-size version of the spatial pyramid pooling, which is a technique designed to capture information at different scales. By using fixed-size spatial pyramid pooling, the SPPF module enables the network to have stronger understanding and adaptability when processing objects of various sizes, thereby enhancing the model's detection performance for multiple target sizes.

[0048] The specific calculation methods of the above modules are as follows:

[0049]

[0050]

[0051] wherein, is the input image, is the kernel size of the pooling operation ( =5,9,13) The input image After passing through the CBS layer, four CSP layers (composed of CBS modules and C2f modules), and the SPPF layer, a total of 5 layers of image features P1~P5 are obtained, i.e., image multi-scale features are obtained.

[0052] The specific calculation methods are as follows:

[0053] wherein, is the input image data after adaptive noise reduction processing, The extracted image multi-scale features, i.e., P1-P5.

[0054] The waveform feature extraction network adopts the encoder of the Transformer for extraction. The Transformer model is a brand-new method for processing sequence data, which is different from the traditional method based on the recurrent neural network or the convolutional neural network. The Transformer completely relies on the attention mechanism to process the dependency relationship in the input sequence, which makes the Transformer have faster speed and better parallelization capability when processing long sequence data. The encoder of the Transformer is composed of multiple identical layers Layer. Each layer includes: multi-head self-attention (MHSA): by performing multiple self-attention mechanisms in parallel, the relationship between different parts of the input sequence is captured. Feed-forward neural network (FFN): used to further process and transform the output of the self-attention layer, enhancing the expression ability of the model. Residual connection (RES) and layer normalization (LayerNorm): the residual connection helps to alleviate the problem of gradient disappearance in deep networks, and the layer normalization helps to stabilize the training process. The specific calculation method is as follows:

[0055] Input waveform data is first linearly mapped into a feature vector, and then passes through the encoder layer , a total of 6 layers of time sequence features L1-L6 are obtained, i.e., waveform time sequence features are obtained. The specific calculation method is as follows:

[0056] wherein, is the input waveform data after adaptive noise reduction processing, is the extracted waveform time sequence feature, i.e., L1-L6.

[0057] S3, the image multi-scale features and the waveform time sequence features are fused by using the constructed adaptive attention weighting model to obtain the fused features; To realize the evaluation method of "waveform-image simultaneous analysis", the extracted image multi-scale features and waveform time sequence features need to be effectively fused, and a waveform-image feature fusion model based on adaptive attention weighting is constructed. The data processing process of the waveform-image feature fusion model is as shown in Figure 7 . For the different influences of different features on the model defect detection performance, the model proposes to use adaptive attention weight to dynamically adjust the weighting strategy of the input waveform-image features, so as to obtain the best fusion effect.

[0058] Firstly, the extracted image features P1-P5 and waveform features L1-L6 are selected. Since deeper image features have more abundant information, P3, P4 and P5 are retained. L1-L6 are obtained from different attention heads in the Transformer encoder, and each feature is extracted from a different perspective, so L2, L4 and L6 are randomly selected. Secondly, the selected features are grouped and fused into three groups: P3 and L2, P4 and L4, and P5 and L6. To facilitate feature fusion, the L layer feature dimension is adjusted using convolution pooling to make it the same as the P layer feature dimension.

[0059] The specific calculation method is as follows: ;

[0060]

[0061] wherein, is a convolution operation, is a max pooling operation.

[0062] The P layer and L layer features are fused using an adaptive attention weighting model to enhance the model's attention to important features. The final fused features F1, F2 and F3 are obtained. To avoid changing the subsequent detection network, F1, F2 and F3 have the same dimensions as P3, P4 and P5, respectively.

[0063] The most important structure in the wavegraph feature fusion is the adaptive attention weighting model. The data processing process of the adaptive attention weighting model is shown in Figure 8 . First, the two input feature maps are concatenated to obtain a comprehensive feature representation. After convolution and activation, the attention map is obtained. Then, the attention map and its complement are multiplied by the two features to perform weighting. Finally, the weighted features are concatenated to obtain the fused feature .

[0064] The specific calculation method is as follows:

[0065]

[0066] wherein, is an image feature, is a waveform feature, is an attention map, is a fused feature, i.e., F1-F3.

[0067] S4, defect detection based on the fusion features, the fusion features F1-F3 are input into a YOLOv8 detection head for processing, and a defect bounding box, a class and a confidence are output.

[0068] The method adopts a parameter self-learning adaptive preprocessing technology, dynamically optimizes Gaussian filter and wavelet filter parameters, and reduces noise interference; secondly, image multi-scale features are extracted through a YOLOv8 backbone network, and waveform time sequence features are extracted in combination with a Transformer encoder; finally, adaptive attention weighting modules are used to dynamically fuse the wave pattern features, enhance the expression ability of the defect key information, and use the fusion features for defect detection. The method is experimented through large pressure spherical tank weld TOFD data, and the results show that by integrating waveform features and adaptive preprocessing methods, the detection and recognition accuracy of the model for defects can be effectively improved, and the detection accuracy is increased by 1.5% compared with the basic model YOLOv8 method. The method improves the accuracy and reliability of defect detection by simulating the field knowledge of artificial "wave pattern simultaneous analysis", and is suitable for intelligent detection of large welded structures.

[0069] The method is experimented through large pressure spherical tank weld TOFD data, and the results show that by integrating waveform features and adaptive preprocessing methods, the detection and recognition accuracy of the model for defects can be effectively improved, and the detection accuracy is increased by 1.5% compared with the basic model YOLOv8 method.

[0070] The application provides a weld defect detection system fusing TOFD image waveform features, and applies the weld defect detection method fusing TOFD image waveform features. The adaptive noise reduction processing module is configured to perform adaptive noise reduction processing on TOFD image data and TOFD waveform data. The feature extraction module is configured to perform feature extraction on the TOFD image data and the TOFD waveform data after adaptive noise reduction processing through double network perspectives, and obtain image multi-scale features and waveform time sequence features. The feature fusion module is configured to fuse the image multi-scale features and the waveform time sequence features in an adaptive attention weighting manner, and obtain fusion features. The detection module is configured to perform defect detection based on the fusion features.

[0071] In still another embodiment of the present application, an electronic device is provided, which includes a processor and a memory for storing a computer program including program instructions, the processor being configured to execute the program instructions stored in the computer storage medium. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gates or transistor logic, discrete hardware components, etc., which are the computing core and control core of the terminal, and are suitable for implementing one or more instructions, and are particularly suitable for loading and executing one or more instructions to implement a corresponding method flow or a corresponding function; the processor in the embodiments of the present application can be used for the operation of a welding defect detection method fusing TOFD image waveform features, including: The TOFD image data and the TOFD waveform data are adaptively denoised; the adaptively denoised TOFD image data and the adaptively denoised TOFD waveform data are respectively extracted for features through double network perspectives, to obtain image multi-scale features and waveform time sequence features; the image multi-scale features and the waveform time sequence features are fused through an adaptive attention weighting manner to obtain fused features; and defect detection is performed based on the fused features.

[0072] In still another embodiment of the present application, the present application further provides a storage medium, specifically a computer readable storage medium (Memory), which is a memory device in a terminal device, and is used for storing programs and data. It can be understood that the computer readable storage medium herein can include an internal storage medium in the terminal device, and of course can also include an expansion storage medium supported by the terminal device. The computer readable storage medium provides a storage space, and the storage space stores an operating system of the terminal. Moreover, one or more instructions suitable for being loaded and executed by the processor are also stored in the storage space, and the instructions can be one or more computer programs (including program codes). It should be noted that the computer readable storage medium herein can be a high-speed RAM memory, or a non-volatile memory such as at least one disk memory.

[0073] The one or more instructions stored in the computer readable storage medium can be loaded and executed by the processor to implement the corresponding steps of the method for detecting a weld defect by fusing TOFD image waveform features in the above embodiments; the one or more instructions stored in the computer readable storage medium are loaded and executed by the processor to implement the following steps: Adaptive noise reduction processing is performed on the TOFD image data and the TOFD waveform data; image multi-scale features and waveform time sequence features are obtained by performing feature extraction on the TOFD image data and the TOFD waveform data subjected to the adaptive noise reduction processing through double network perspectives; the image multi-scale features and the waveform time sequence features are fused by using an adaptive attention weighting manner to obtain fused features; and defect detection is performed based on the fused features.

[0074] Please refer to Figure 9 The terminal device is a computer device. The computer device 60 of this embodiment includes a processor 61, a memory 62, and a computer program 63 stored in the memory 62 and executable on the processor 61. When the computer program 63 is executed by the processor 61, the method for calculating fluid composition in a reservoir stimulation wellbore in the embodiment is implemented. To avoid repetition, details are not described here. Alternatively, when the computer program 63 is executed by the processor 61, the functions of each model / unit in the system for calculating fluid composition in a reservoir stimulation wellbore in the embodiment are implemented. To avoid repetition, details are not described here.

[0075] The computer device 60 can be a desktop computer, a notebook computer, a palm computer, a cloud server, and the like. The computer device 60 can include, but is not limited to, the processor 61 and the memory 62. Those skilled in the art can understand that the computer device 60 can further include other components, for example, an input / output device, a network access device, a bus, and the like. Figure 9 The computer device 60 is only an example and does not constitute a limitation on the computer device 60, and can include more or fewer components than those shown, or combine certain components, or different components, for example, the computer device can further include an input / output device, a network access device, a bus, and the like.

[0076] The processor 61 can be a central processing unit (CPU), and can also be other general-purpose processors, central processing units, graphics processing units, digital signal processors (DSPs), application specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs) or other programmable logic devices, discrete gates or transistor logic components, quantum computing-based data processing logic components, discrete hardware components, etc. The general-purpose processor can be a microprocessor or the processor can also be any conventional processor.

[0077] The memory 62 can be an internal storage unit of the computer device 60, such as a hard disk or a memory of the computer device 60. The memory 62 can also be an external storage device of the computer device 60, such as a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, etc.

[0078] Further, the memory 62 can include both an internal storage unit and an external storage device of the computer device 60. The memory 62 is used to store computer programs and other programs and data required by the computer device. The memory 62 can also be used to temporarily store data that has been output or will be output.

[0079] Any reference to memory, database, or other media herein includes at least one of volatile and non-volatile memory. Non-volatile memory can include read-only memory (ROM), tape, floppy disks, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, and the like. Volatile memory can include random access memory (RAM), external cache memory, and the like. By way of illustration, and not limitation, RAM can be SRAM, DRAM, or the like.

[0080] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a blockchain, and the like, without being limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, and the like, without being limited thereto.

[0081] Please refer to Figure 10 The terminal device is a chip, and the chip 600 of the embodiment includes a processor 622, the number of which can be one or more, and a memory 632 for storing a computer program executable by the processor 622. The computer program stored in the memory 632 can include one or more modules each corresponding to a set of instructions. In addition, the processor 622 can be configured to execute the computer program to perform the generalizable monocular absolute depth map estimation method described above.

[0082] In addition, the chip 600 can also include a power supply component 626 and a communication component 650, the power supply component 626 can be configured to perform power management of the chip 600, and the communication component 650 can be configured to realize communication of the chip 600, for example, wired or wireless communication. In addition, the chip 600 can also include an input / output interface 658. The chip 600 can operate based on an operating system stored in the memory 632.

[0083] The foregoing merely illustrates the principles of the application and application of its leading features. This application is not limited to the exact details shown above, and various modifications can be made to the embodiments described without departing from the spirit or scope of the application. Accordingly, the embodiments are to be considered as illustrative and not restrictive, and the scope of the application is to be determined not with reference to the above description but with reference to the appended claims, and their equivalents. No admission is made that any reference constitutes prior art. It is the combination of elements that is claimed. The scope of the technology is thus those limitations as indicated by the appended claims.

[0084] Furthermore, it should be appreciated that a single independent technical solution is not contained in each embodiment, and the description of the specification is only for the sake of clarity, and the skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can be combined appropriately to form other embodiments that can be understood by the skilled in the art. The above is only to illustrate the technical idea of the present application, and cannot limit the protection scope of the present application. Any modification made on the basis of the technical idea of the present application and technical solution falls within the protection scope of the claims of the present application.

Claims

1. A weld defect detection method integrating TOFD image waveform features, characterized in that: The following steps are involved: Adaptive noise reduction processing is performed on TOFD image data and TOFD waveform data; Through dual network perspectives, feature extraction is performed on TOFD image data and TOFD waveform data after adaptive noise reduction, and image multi-scale features and waveform time series features are obtained. Adaptive attention weighting is used to fuse the multi-scale features of the image and the waveform temporal features to obtain the fused features; Defect detection based on fused features.

2. A weld defect detection method integrating TOFD image waveform features according to claim 1, characterized in that: The adaptive noise reduction processing of the TOFD image data and the TOFD waveform data includes: Based on the construction of preprocessing parameter prediction network Adaptively preprocessing the TOFD image data and TOFD waveform data, predicting Gaussian processing parameters and wavelet threshold filter parameters, and scaling the Gaussian processing parameters and wavelet threshold filter parameters; Based on Gaussian processing parameters, the TOFD image data is subjected to noise reduction using the standard Gaussian filter preprocessing method; Based on the wavelet threshold filtering parameters, the TOFD waveform data is denoised using the wavelet threshold filtering (WF) preprocessing method.

3. The weld defect detection method integrating TOFD image waveform features according to claim 2 is characterized in that: The Gaussian processing parameters include kernel size and standard deviation , the process of linear scaling of the Gaussian processing parameters is: in, is the TOFD image data, is the scaling factor, ; Wavelet threshold filtering parameters include the number of decomposition layers and threshold , the process of linear scaling of the wavelet threshold filter parameters is: in, is the TOFD waveform data, is the scaling factor, .

4. The weld defect detection method integrating TOFD image waveform features according to claim 2 is characterized in that: The TOFD image data and TOFD waveform data after adaptive noise reduction are respectively subjected to feature extraction through dual network perspectives to obtain image multi-scale features and waveform time series features, including: Construct feature extraction networks, namely image feature extraction networks and waveform feature extraction networks; Based on the image feature extraction network, feature extraction is performed on the TOFD image data after adaptive noise reduction to obtain multi-scale features of the image; Based on the waveform feature extraction network, the TOFD waveform data after adaptive noise reduction is subjected to feature extraction to obtain the waveform timing characteristics.

5. The weld defect detection method integrating TOFD image waveform features according to claim 4 is characterized in that: The process of extracting features from the TOFD image data after adaptive noise reduction based on the image feature extraction network to obtain multi-scale features of the image is as follows: in, is the input image data after adaptive noise reduction processing, is the extracted multi-scale features of the image.

6. The weld defect detection method integrating TOFD image waveform features according to claim 4 is characterized in that: The process of extracting features from the TOFD waveform data after adaptive noise reduction based on the waveform feature extraction network to obtain waveform timing features is as follows: in, is the input waveform data after adaptive noise reduction processing, is the extracted waveform timing characteristics.

7. The weld defect detection method integrating TOFD image waveform features according to claim 1 is characterized in that: The adaptive attention weighting method is used to fuse the multi-scale features of the image and the waveform time series features to obtain the fused features, including: Filter the multi-scale features of the image and retain the multi-scale features of the three layers of the image; Randomly extract three layers of features from the waveform time series features and perform convolution processing to make the extracted three layers of features have the same dimension as the image multi-scale features; The waveform timing features of the same dimension and the corresponding feature maps of the image multi-scale features are spliced ​​to obtain the feature representation, and then the attention map is obtained after convolution activation. ; Attention map and its complement Multiply the two features by Multipy to perform weighting, and concatenate the weighted features to obtain the fusion feature .

8. A weld defect detection system integrating TOFD image waveform features, applying the weld defect detection method integrating TOFD image waveform features according to any one of claims 1 to 7, characterized in that: include: An adaptive noise reduction processing module is configured to perform adaptive noise reduction processing on TOFD image data and TOFD waveform data; A feature extraction module is configured to extract features from the TOFD image data and TOFD waveform data after adaptive noise reduction through dual network perspectives to obtain multi-scale image features and waveform time series features; A feature fusion module is configured to fuse the image multi-scale features and the waveform temporal features using an adaptive attention weighting method to obtain a fused feature; The detection module is configured to perform defect detection based on fused features.

9. An electronic device, characterized in that: The invention comprises a memory, a processor, and a computer program stored in the memory and executable in the processor, wherein when the processor executes the computer program, the steps of a weld defect detection method integrating TOFD image waveform features as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of a weld defect detection method integrating TOFD image waveform features according to any one of claims 1 to 7 are implemented.

Images