Ultrasonic phased array weld defect intelligent identification method based on adaptive filtering

Abstract

The invention discloses an ultrasonic phased array weld defect intelligent identification method based on adaptive filtering, and belongs to industrial nondestructive detection image identification methods, the technical scheme is that the method comprises the following steps: sampling and sorting pipeline weld PAUT detection images; dividing the collected image samples into a training set and a test set; pixel classification and filtering noise reduction are carried out on the training set images through a hybrid adaptive filter; performing abnormal signal labeling on the preprocessed training set image and the unprocessed test set image respectively; training a target recognition network through the preprocessed training set; and collecting a new fan-shaped scanning image of the pipeline weld defect as network input, and identifying the weld defect. The ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering has the beneficial effect that the ultrasonic phased array weld defect intelligent identification method based on self-adaptive filtering is provided.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of industrial non-destructive testing image recognition methods, and particularly relates to an ultrasonic phased array weld defect intelligent recognition method based on adaptive filtering. BACKGROUND

[0002] The ultrasonic phased array technology is a technology commonly used for non-destructive testing and can be used for detecting weld defects. Currently, the discrimination of the ultrasonic phased array detection result can be determined by manual judgment or with the aid of existing intelligent analysis methods. For example, a patent with the patent name of a weld ultrasonic phased array detection data intelligent analysis method based on deep learning and the publication number of CN111060601A, however, has the following problems and shortcomings:

[0003] 1. The sample quantity of the ultrasonic phased array defect data set is low, and the ultrasonic phased array image itself has many noise and echo pseudo-defect signals. The pre-processing of the network training set in the patent is too simple. In the case of limited data set and pseudo-defect signal interference, the learning of the network on the defect characteristics is not sufficient enough, which affects the recognition accuracy.

[0004] 2. The Fast-RCNN algorithm used in the patent belongs to the two-stage category. This kind of algorithm has an additional pre-frame step, and its accuracy is higher than that of the algorithm without pre-frame, but its speed performance is poor.

[0005] 3. The Fast-RCNN algorithm has weak detection ability for small targets, and small defects / deficiency targets in the weld are easy to be missed. SUMMARY

[0006] The purpose of the present application is to provide an ultrasonic phased array weld defect intelligent recognition method based on adaptive filtering. Through the adaptive filter, the intelligent noise reduction of the fan-shaped scanning image is performed in the case that the weld ultrasonic phased array data sample quantity is low and is not conducive to deep learning, and the characteristics of the weld defect signal are enhanced, so as to improve the recognition accuracy of the small defect / deficiency of the deep learning algorithm and improve the defect detection speed.

[0007] The present application is realized through the following measures: an ultrasonic phased array weld defect intelligent recognition method based on adaptive filtering, characterized in that it comprises:

[0008] The PAUT detection image of the pipeline weld is sampled and arranged. The ultrasonic phased array detection is performed on the pipeline sample, and the fan-shaped scanning sampling is performed every 1 mm.

[0009] Data set division, the collected image samples are divided into training set and test set, generally divided according to the proportion of 7:3, ensure the reasonable distribution of defect type, uniform picture ratio, the specific proportion of training set and test set can be determined according to the actual situation, generally the number of training set is greater than the number of test set,

[0010] Image preprocessing, pixel classification and filter denoising of training set image are carried out through hybrid adaptive filter; effectively improve the quality of detection image, so as to improve the intelligent recognition and classification accuracy of weld defect.

[0011] Sample labeling, analyze the imaging characteristics of pipe weld PAUT detection image, label the position, size and category of abnormal signal (defect) of preprocessed training set image and unprocessed test set image respectively, and record the information of fan-shaped scanning and corresponding defects.

[0012] Through the training of the training set after preprocessing, the target recognition network is trained, the target detection framework based on YOLOv5 target recognition network model is adopted, the lightest YOLOv5n pre-training weight model is used to train the network; the output layer adopts loss function CIOU_LOSS and flexible non-maximum suppression SoftNMS.

[0013] Verify the target recognition network, load the unprocessed test set picture into the model, and output the recognition test result, and adjust the network parameters according to the result;

[0014] Collect new pipe weld defect fan-shaped scanning image as network input, load the trained YOLOv5 weight model file to recognize the weld defect.

[0015] Among them, image preprocessing, pixel classification and filter denoising of training set image are carried out through hybrid adaptive filter; specifically including:

[0016] Pixel classification, according to the gradient between the target pixel and the eight direction pixels around it and the threshold value, determine the number of large and small gradients of each pixel position, and combine the neighborhood structure characteristics of the pixel, the position of the gradient and the connectivity characteristics of the pixel, propose 8 rules, follow the rules to classify the pixel type of ultrasonic phased array fan-shaped scanning image, the specific rules are as follows:

[0017] Rule 1: calculate the gradient of the target pixel position p∈P in eight directions, if the gradient is higher than the set threshold value, the target pixel p is noise pixel, wherein higher than the set threshold value means high gradient.

[0018] Rule 2: calculate the gradient of the target pixel position p∈P in eight directions, if the gradient is lower than or equal to the set threshold value, the target pixel p is smooth area pixel, at this time the filter window size is 5x5, wherein lower than or equal to the set threshold value means low gradient.

[0019] Rule 3: Compute the gradient of the eight directions of the target pixel position p e P, if there are seven high gradients and one low gradient, then further judge the neighborhood of the low gradient direction pixel and its connectivity characteristics.

[0020] Rule 4: Compute the gradient of the eight directions of the target pixel position p e P, if there are six high gradients and two low gradients, then use the same method as rule 3 to compute the gradient of the neighborhood of the low gradient direction pixel.

[0021] Rule 5: Compute the gradient of the eight directions of the target pixel position p e P, if there are five high gradients and three low gradients, or four high gradients and four low gradients, then the target pixel p is an edge or detail pixel, at this time the filter window size is 4x4.

[0022] Rule 6: Compute the gradient of the eight directions of the target pixel position p e P, if there are three high gradients and five low gradients, then find the maximum and minimum difference of the high gradient pixel value set H of the neighborhood of the high gradient direction pixel, i.e. AbsolutVal = |max(H)-min(H)|. If AbsolutVal < 2T, determine that the target pixel p is an edge or detail pixel, and the filter window size is 4x4; otherwise, it is a smooth region pixel, and the filter window size is 3x3.

[0023] Rule 7: Compute the gradient of the eight directions of the target pixel position p e P, if there are two high gradients and six low gradients, then find the positions of the two high gradient direction pixels in the current neighborhood window, and then judge whether the two pixels are connected (eight connectivity). If connected, further judge the difference between the two pixels, if the difference is greater than T, determine that the target pixel p is a smooth region pixel, and the filter window size is 3x3. Otherwise, use the same method as rule 3 to calculate the gradient of each eight-connected pixel not in the current neighborhood window with the average value of the two pixels, and if the number of low gradients is less than 3, determine that the target pixel p is a smooth region pixel, and the filter window size is 3x3; otherwise, it is determined to be an edge or detail pixel, and the filter window size is 3x3. If the two pixels are not connected, determine that the target pixel p is an edge or detail pixel, and the filter window size is 3x3.

[0024] Rule 8: Compute the gradient of the eight directions of the target pixel position p e P, if there is one high gradient and seven low gradients, then the target pixel p is a smooth region pixel, and the filter window size is 3x3.

[0025] According to the pixel classification result, the image is filtered using three different filters. For noise pixels, a median filter is used for filtering; for edge or detail pixels, a bilateral filter is used for filtering; for smooth area pixels, a weighted mean filter is used for filtering. The filter window size of the bilateral filter and the weighted mean filter is adaptively selected according to the window size given in the pixel type determination rule, and the filter window size of the median filter is 3x3.

[0026] Further, the target recognition network output end explains that CIOU_LOSS is used as the loss function of the bounding box; CIOU adds an influence factor to consider the overlapping area, center point distance and aspect ratio of the predicted box and the GT box in the loss; when two defect targets are very close, softNMS is used to provide a slightly lower score to replace the original score, so that the low confidence will be gently suppressed by the high confidence box instead of being directly set to zero.

[0027] Further, new pipeline weld defect sector scan image is collected as network input, and the trained YOLOv5 weight model file is loaded for weld defect recognition, including: the network first automatically scales the verification image, and then sends it to the Backbone network for feature extraction; then the feature map output by the backbone network is input into the neck network to detect the defect object and its category, position and size information; finally, the network labels the defect category and defect position box to the un-scaled verification image and outputs and saves. At this time, the recognition of the ultrasonic phased array defect image of YOLOv5 is realized, which has practical application value for the intelligent detection and automatic evaluation of ultrasonic phased array defects to improve the current situation of manual defect evaluation.

[0028] The embodiment of the application further provides an electronic device, which can include a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory complete mutual communication through the communication bus.

[0029] The memory is used to store computer programs.

[0030] The processor is used to execute the computer programs stored on the memory to realize the adaptive filtering based ultrasonic phased array weld defect intelligent recognition method in the above-mentioned embodiment one.

[0031] The communication bus mentioned in the terminal can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0032] The communication interface is used for communication between the terminal and other devices.

[0033] The memory can include a Random Access Memory (RAM) and can also include a non-volatile memory, such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the aforementioned processor.

[0034] The processor mentioned above can be a general-purpose processor, including a Central Processing Unit (CPU), a Network Processor (NP), etc. It can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0035] In another embodiment provided in the application, a computer readable storage medium is also provided, and the computer readable storage medium stores instructions, when the instructions run on a computer, the computer executes the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in the above-mentioned embodiment one.

[0036] In another embodiment provided in the application, a computer program product containing instructions is also provided, when the instructions run on a computer, the computer executes the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in the above-mentioned embodiment.

[0037] The technical scheme provided by the embodiment of the application has the beneficial effects that: before and after intelligent adaptive filtering, the noise, echo and other pseudo defects in the ultrasonic phased array fan-shaped scanning image of the weld are filtered out, and the edge of the lack and defect signal is enhanced, and the learning effect of the deep learning network is improved.

[0038] Compared with other target detection frameworks (SSD, Fast R-CNN, etc.), the framework improves the defect recognition accuracy, classification accuracy, and detection speed. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the present application, the drawings used in the embodiments will be briefly introduced as follows. Obviously, the drawings listed below are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 is a flow chart of an ultrasonic phased array weld defect intelligent recognition method based on adaptive filtering in an embodiment of the present application;

[0041] Figure 2 is an algorithm flow chart of an ultrasonic phased array weld defect intelligent recognition method based on adaptive filtering in an embodiment of the present application;

[0042] Figure 3 is an example of a seven-high-gradient and one-low-gradient neighborhood window (the only low-gradient direction pixel is E);

[0043] Figure 4 is an example of a seven-high-gradient and one-low-gradient neighborhood window (the only low-gradient direction pixel is ES);

[0044] Figure 5 is the effect before and after the mixed adaptive filter;

[0045] Figure 6 is the verification image recognition result of the girth weld defect recognition of the long oil and gas pipeline Figure 1 ;

[0046] Figure 7 is the verification image recognition result of the girth weld defect recognition of the long oil and gas pipeline Figure 2 ;

[0047] Figure 8 is the verification image recognition result of the girth weld defect recognition of the long oil and gas pipeline Figure 3 ;

[0048] Figure 9 is the verification image recognition result of the girth weld defect recognition of the long oil and gas pipeline Figure 4 . DETAILED DESCRIPTION

[0049] In order to make the purpose, technical solutions and advantages of the present application more clear, the present application will be further described in detail below in combination with embodiments. Of course, the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.

[0050] Embodiment I:

[0051] Referring to Figures 1-9 An ultrasonic phased array weld defect intelligent identification method based on adaptive filtering, characterized in that it comprises:

[0052] S1, sample and arrange the pipeline weld PAUT detection image; perform ultrasonic phased array detection on the pipeline sample, and perform fan-shaped scanning sampling every 1 mm.

[0053] S2, data set division, the collected image samples are divided into a training set and a test set according to a 7:3 ratio; ensure reasonable distribution of defect types, and unify the picture ratio.

[0054] S3, image preprocessing, pixel classification and filter denoising of the training set image are performed through a hybrid adaptive filter; the quality of the detection image is effectively improved, thereby improving the intelligent identification and classification accuracy of the weld defect.

[0055] S4, sample labeling, analyze the imaging characteristics of the pipeline weld PAUT detection image, label the position, size and category of the abnormal signal (defect) of the preprocessed training set image and the unprocessed test set image respectively, and record the fan-shaped scanning and the information of the corresponding defects.

[0056] S5, train the target recognition network through the preprocessed training set, use the target detection framework based on the YOLOv5 target recognition network model, use the lightest YOLOv5n pre-training weight model to train the network; the output layer uses the loss function CIOU_LOSS and the flexible non-maximum suppression SoftNMS.

[0057] S6, verify the target recognition network, load the unprocessed test set picture into the model, output the recognition test result, and adjust the network parameters according to the result;

[0058] S7, collect new pipeline weld defect fan-shaped scanning image as network input, load the trained YOLOv5 weight model file to recognize the weld defect.

[0059] Among them, the image preprocessing, the pixel classification and filter denoising of the training set image are performed through a hybrid adaptive filter, specifically including:

[0060] S31, pixel classification, according to the gradient between the target pixel and the eight directions of the surrounding pixels and the threshold value, determine the number of large and small gradients of each pixel position, and combine the neighborhood structure characteristics of the pixel, the position of the gradient and the connectivity characteristics of the pixel, propose 8 rules, follow the rules to classify the pixel type of the ultrasonic phased array fan-shaped scanning image, the specific rules are as follows:

[0061] Rule 1: Calculate the gradients in eight directions of the target pixel position p∈P. If the gradients are all higher than the set threshold (hereinafter referred to as high gradient), the target pixel p is a noise pixel.

[0062] Rule 2: Calculate the gradients in eight directions at the target pixel position p∈P. If the gradients are all lower than or equal to the set threshold (hereinafter referred to as low gradient), the target pixel p is a smooth area pixel, and the filter window size is 5×5.

[0063] Rule 3: Calculate the gradient in eight directions of the target pixel position p∈P. If there are seven high gradients and one low gradient, it is necessary to further determine the neighborhood of the pixel in the direction of the low gradient and its connectivity characteristics. As shown in the figure, the center grid is the target pixel p. Assuming that the only low gradient direction pixel appears in the four-connected position, taking pixel E as an example, the pixel position in the neighborhood window of E that is not included in the neighborhood window of p is as follows: Figure 3 As shown. Calculate the gradients of these pixels and the target pixel p respectively, that is,

[0064]

[0065]

[0066]

[0067] if If two of them are less than or equal to the threshold, the target pixel p is an edge or detail pixel, and the filter window size is 3×3. Otherwise, the target pixel p is determined to be a noise pixel.

[0068] Assuming that the only low gradient direction pixel appears in the eight-connected position, taking pixel ES as an example, the pixel position in the ES neighborhood window that is not included in the p neighborhood window is as follows: Figure 4 As shown in Figure 2, the gradients of these pixels and the target pixel p are calculated respectively. If there are three low gradients, the target pixel p is an edge or detail pixel, and the filter window size is 3×3. Otherwise, the target pixel p is determined to be a noise pixel.

[0069] Rule 4: Compute the gradient of the target pixel position p e P in eight directions, if there are six high gradients and two low gradients, then compute the gradient of the neighborhood of the pixels in the direction of low gradients using the same method as rule 3. If the number of low gradients is less than four, then the target pixel p is determined as a noise pixel. Otherwise, take the median value of the current 3x3 window to assist the judgment, i.e. MedVal(p) = median{Y(i+k,j+l)}, k, l e {-1,0,1}: when |MedVal(p)-Y(i,j)|<T, the target pixel p is determined as an edge or detail pixel, and the vector filtering window size is 5x5. Otherwise, average the pixel values of the high gradient set H and the low gradient set L of the neighborhood of the pixels in the direction of low gradients to obtain Hmean and Lmean, respectively. If |Hmean-Lmean|<0.5T, then the target pixel p is a noise pixel, otherwise it is an edge or detail pixel, and the filtering window size is 4x4.

[0070] Rule 5: Compute the gradient of the target pixel position p e P in eight directions, if there are five high gradients and three low gradients, or four high gradients and four low gradients, then the target pixel p is an edge or detail pixel, and the filtering window size is 4x4.

[0071] Rule 6: Compute the gradient of the target pixel position p e P in eight directions, if there are three high gradients and five low gradients, then compute the absolute value of the difference between the maximum and minimum values of the high gradient pixel value set H in the direction of high gradients, i.e. AbsolutVal = |max(H)-min(H)|. If AbsolutVal<2T, the target pixel p is determined as an edge or detail pixel, and the filtering window size is 4x4; otherwise, it is a smooth region pixel, and the filtering window size is 3x3.

[0072] Rule 7: Compute the gradient of the target pixel position p e P in eight directions, if there is one high gradient and seven low gradients, then find the positions of the two high gradient pixels in the current neighborhood window, and then determine whether the two pixels are connected (eight-connected). If they are connected, further determine the difference between the two pixels, and if the difference is greater than T, the target pixel p is determined as a smooth region pixel, and the filtering window size is 3x3. Otherwise, compute the gradient of each eight-connected pixel outside the current neighborhood window with respect to the average value of the two pixels using the same method as rule 3. If the number of low gradients is less than three, the target pixel p is determined as a smooth region pixel, and the filtering window size is 3x3; otherwise, it is an edge or detail pixel, and the filtering window size is 3x3. If the two pixels are not connected, the target pixel p is determined as an edge or detail pixel, and the filtering window size is 3x3.

[0073] Rule 8: Calculate the gradient in eight directions of the target pixel position p∈P. If there are two high gradients and six low gradients, the target pixel p is a smooth area pixel and the filter window size is 3×3.

[0074] S32: Based on the pixel classification results, the image is filtered using three different filters. Noise pixels are filtered using a median filter; edge or detail pixels are filtered using a bilateral filter; and pixels in smooth areas are filtered using a weighted mean filter. The filter window sizes for the bilateral and weighted mean filters are adaptively selected based on the window size specified in the pixel type discrimination rule. The median filter has a filter window size of 3×3.

[0075] The YOLOv5 network structure is as follows: Figure 2 As shown in the figure, the output shows that CIOU_LOSS is used as the bounding box loss function. CIOU adds an influencing factor that takes into account the overlap area, center point distance, and aspect ratio of the predicted and GT boxes. SoftNMS is also used: To avoid misjudgment when dense weld defects are framed as a single defect, softNMS is used to provide a slightly lower score when two defect targets are very close together. This allows the low-confidence ones to be gently suppressed by the high-confidence boxes, rather than being directly set to zero.

[0076] Identification and verification operation process is as follows Figure 2 As shown by the arrows: A new fan-shaped scan image of a pipeline weld defect is collected as network input, and the trained YOLOv5 weight model file is loaded; the network first automatically scales the verification image and then feeds it into the Backbone network for feature extraction; the feature map output by the backbone network is then input into the neck network to detect defect objects and their categories, locations, and sizes; finally, the network annotates the defect categories and defect location boxes onto the unscaled verification image and outputs it for storage.

[0077] At this point, the recognition of YOLOv5 ultrasonic phased array defect images has been realized, which has practical application value for the research on ultrasonic phased array defect intelligent detection and automated evaluation to improve the current status of manual defect assessment. Figures 6-9 shown.

[0078] Example 2:

[0079] An embodiment of the present application further provides an electronic device, which may include: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other via the communication bus.

[0080] Memory for storing computer programs;

[0081] The processor is configured to implement the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in the first embodiment when executing the computer program stored in the memory.

[0082] The communication bus mentioned above can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc.

[0083] The communication interface is configured to communicate between the terminal and other devices.

[0084] The memory can include a Random Access Memory (RAM) and can also include a non-volatile memory such as at least one disk memory. Optionally, the memory can also be at least one storage device located away from the processor.

[0085] The processor mentioned above can be a general processor including a Central Processing Unit (CPU), a Network Processor (NP), etc. and can also be a Digital Signal Processing (DSP), an Application Specific Integrated Circuit (ASIC), a Field-Programmable Gate Array (FPGA) or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component.

[0086] Embodiment three:

[0087] In another embodiment provided in the application, a computer readable storage medium is provided, and the computer readable storage medium stores instructions, and when the instructions are executed on a computer, the computer executes the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in the first embodiment.

[0088] Embodiment four:

[0089] In a further embodiment provided in the present application, a computer program product containing instructions is also provided, which, when executed on a computer, causes the computer to perform the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in the above-mentioned embodiment one.

[0090] In the above embodiments, the implementation can be achieved by software, hardware, firmware or any combination thereof, entirely or partially. When implemented by software, the implementation can be in the form of computer program product, entirely or partially. The computer program product contains one or more computer instructions. When the computer program instructions are loaded and executed on the computer, the processes or functions described in the embodiments of the present application are entirely or partially generated. The computer can be a general purpose computer, a special purpose computer, a computer network or other programmable device. The computer instructions can be stored in a computer readable storage medium or transferred from one computer readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server or data center to another website, computer, server or data center through wired (such as coaxial cable, optical fiber, digital subscriber line) or wireless (such as infrared, wireless, microwave, etc.) mode. The computer readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. integrated with one or more available media. The available media can be magnetic media (such as floppy disk, hard disk, magnetic tape), optical media (such as DVD), or semiconductor media (such as solid state disk), etc.

[0091] The above description is only the preferred embodiment of the present application, and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. An ultrasonic phased array weld seam defect intelligent identification method based on adaptive filtering, characterized in that, It comprises: PAUT detection image of sample sorting pipeline weld; Data set division, dividing the collected image samples into a training set and a test set; Image preprocessing, classifying and filtering noise reduction of the training set images through a hybrid adaptive filter; Sample labeling, labeling the abnormal signals of the preprocessed training set images and the unprocessed test set images respectively, and recording the sector scan images and the information of the corresponding defects; Training the target recognition network through the preprocessed training set, using the YOLOv5n pre-trained weight model to train the network; Collecting new pipeline weld defect sector scan image as network input, loading the trained YOLOv5 weight model file to identify the weld defects.

2. The method according to claim 1, wherein, The output layer of the identification network uses the loss function CIOU_LOSS and the flexible non-maximum suppression SoftNMS.

3. The adaptive filter based ultrasonic phased array weld seam defect intelligent identification method according to claim 1, characterized in that, Classify the pixels of the training set images through a hybrid adaptive filter, including: According to the gradient between the target pixel and the pixels in its eight directions and the threshold value, determine the number of large and small gradients at each pixel position, and set rules by combining the neighborhood structure characteristics of the pixel, the position of the gradient, and the connectivity characteristics of the pixel, and classify the ultrasonic phased array sector scan image pixel types according to the rules.

4. The adaptive filter based ultrasonic phased array weld seam defect intelligent identification method according to claim 3, characterized in that, Filtering and noise reduction, including: According to the pixel classification results, use three different filters to filter the image: For noise pixels, use a median filter for filtering; For edge or detail pixels, use a bilateral filter for filtering; For smooth area pixels, use a weighted mean filter for filtering.

5. The ultrasonic phased array weld defect intelligent identification method based on adaptive filtering according to claim 3, the specific rules are as follows: Rule 1: Calculate the gradient of the target pixel position p∈P in eight directions, if the gradient is higher than the set threshold value, the target pixel p is noise pixel, where higher than the set threshold value means high gradient; Rule 2: Calculate the gradient of the target pixel position p∈P in eight directions, if the gradient is lower than or equal to the set threshold value, the target pixel p is a smooth area pixel, and the filter window size is 5x5 at this time, where lower than or equal to the set threshold value means low gradient; Rule 3: Calculate the gradient of the target pixel position p∈P in eight directions, if there are seven high gradients and one low gradient, further distinguish the neighborhood of the low gradient direction pixel and its connectivity characteristics. Rule 4: Calculate the gradient of the target pixel position p∈P in eight directions, if there are six high gradients and two low gradients, use the same method as rule 3 to calculate the gradient of the neighborhood of the low gradient direction pixel; Rule 5: Calculate the gradient of the target pixel position p∈P in eight directions, if there are five high gradients and three low gradients, or four high gradients and four low gradients, the target pixel p is an edge or detail pixel, and the filter window size is 4x4 at this time; Rule 6: Calculate the gradient of the target pixel position p in eight directions, if there are three high gradients and five low gradients, then find the maximum and minimum difference of the high gradient pixel value set H in the direction of the high gradient, that is, AbsolutVal = |max(H)-min(H)|; If AbsolutVal≤2T, determine that the target pixel p is an edge or detail pixel, and the filter window size is 4x4; otherwise, it is a smooth area pixel, and the filter window size is 3x3; Rule 7: Calculate the gradient of the target pixel position p in eight directions, if there is one high gradient and seven low gradients, then find the positions of the two high gradient pixels in the current neighborhood window, and then determine whether the two pixels are connected; If connected, further determine the difference between the two pixels, and if the difference is greater than T, determine that the target pixel p is a smooth area pixel, and the filter window size is 3x3; Otherwise, calculate the gradient of each eight-connected pixel not in the current neighborhood window of the two pixels and the average value of the two pixels in the same way as rule 3, and if the number of low gradients is less than 3, determine that the target pixel p is a smooth area pixel, and the filter window size is 3x3; otherwise, determine that it is an edge or detail pixel, and the filter window size is 3x3; if the two pixels are not connected, determine that the target pixel p is an edge or detail pixel, and the filter window size is 3x3; Rule 8: Calculate the gradient of the target pixel position p in eight directions, if there is one high gradient and seven low gradients, then the target pixel p is a smooth area pixel, and the filter window size is 3x3.

6. The adaptive filter based ultrasonic phased array weld seam defect intelligent identification method according to claim 1, characterized in that, Target recognition network output end explanation: CIOU_LOSS is used as the loss function of the bounding box; CIOU adds an influence factor to consider the overlap area, center point distance and aspect ratio of the predicted box and GT box in the loss; When two defect targets are very close, softNMS is used to provide a slightly lower score to replace the original score, so that the low confidence is softly suppressed by the high confidence box instead of being directly set to zero.

7. The adaptive filter based ultrasonic phased array weld seam defect intelligent identification method according to claim 1, characterized in that, It also includes verifying the target recognition network, specifically: Load the unprocessed test set image into the model and output the recognition test result, and adjust the network parameters according to the result.

8. The adaptive filter based ultrasonic phased array weld seam defect intelligent identification method according to claim 1, characterized in that, Collect new pipe weld defect fan-shaped scan image as network input, load trained YOLOv5 weight model file for weld defect recognition, including: The network first automatically scales the verification image, and then sends it to the Backbone network for feature extraction; Then input the feature map output by the backbone network into the neck network to detect the defect object and its category, position and size information; Finally, the network labels the defect category and defect position box to the un-scaled verification image and outputs and saves it.

9. An electronic device, comprising: It includes: A processor and a memory, the processor is used to execute the adaptive filtering based ultrasonic phased array weld defect intelligent recognition method program stored in the memory, to realize the adaptive filtering based ultrasonic phased array weld defect intelligent recognition method in any one of claims 1-8.

10. A storage medium, characterized by The storage medium stores one or more programs, which can be executed by one or more processors to implement the adaptive filtering based ultrasonic phased array weld defect intelligent identification method in any one of claims 1-8.

Images