WO2020038389A1

WO2020038389A1 - Welding seam negative defect recognition method

Info

Publication number: WO2020038389A1
Application number: PCT/CN2019/101716
Authority: WO
Inventors: 董绍华; 谢书懿; 孙玄; 张河苇; 张来斌
Original assignee: 中国石油大学(北京)
Priority date: 2018-08-23
Filing date: 2019-08-21
Publication date: 2020-02-27
Also published as: CN109115812A

Abstract

A welding seam negative defect recognition method, specifically comprising: acquiring a welding seam negative of a welding seam by means of radiography (S101); acquiring welding seam image texture feature data of the welding seam negative by utilizing an image recognition technique (S102); establishing a recognition model via a machine learning algorithm on the basis of the welding seam image texture feature data and of welding seam negative defect types (S103); acquiring the welding seam image texture feature data of a welding seam negative to be detected (S104); and calculating on the basis of the welding seam image texture feature data of the welding seam negative to be detected and of the recognition model to acquire the defect type of the welding seam negative to be detected (S105). With this, effectively solved is the problem of inaccuracy in the texture precision of welding negative recognition and feature determination, thus greatly increasing the precision of welding negative image edge detection and defect tracking and recognition, and laying a foundation for the development of a welding seam feature recognition technique.

Description

Defect recognition method for welding film

Technical field

The invention relates to the field of safety inspection of oil and gas pipelines, in particular to a method for identifying a defect of a weld negative.

Background technique

As a basic process method of connecting members, welding has played an important role in the production and daily maintenance of China's industrial field. Due to various defects in the welding construction, affected by the stability of various parameters during the welding process, various defects such as slag inclusion, cracks, and pores will inevitably occur in the weld. In order to ensure the quality of the welded components, there are It is necessary to carry out detailed inspection and reasonable evaluation of weld defects. Conventional non-destructive testing methods mainly include ultrasonic testing and X-ray testing. The X-ray inspection method provides reliable guidance for subsequent welding work with its high sensitivity, intuitive and reliable, and good repeatability, so it is widely used in oil and gas pipelines and petrochemical fields. In the past, the specific location and type of defects in the weld were mostly judged manually. However, this method has the disadvantages of high labor intensity, low efficiency, strong subjectivity, etc., and it is easy to be affected by the quality.

X-ray digital real-time imaging detection technology has the characteristics of noisy images, low contrast, blurred weld edges and more textures. It is difficult to meet the required quality standards. In order to improve the quality of inspection objects, it is necessary to carry out subsequent computer image processing. And combined with the computer evaluation form to complete the automatic identification of defects in the weld. However, due to many problems in the actual detection process, such as motion-induced image capture blurring (edges and defects are blurred and rough), noise caused by various aspects is trapped in the sample to be detected, resulting in long recognition process and low accuracy. , Which severely limits the application of image processing technology in this field; in addition, for the same workpiece, the angle of ray irradiation during detection is different, and the quality of the obtained image also varies greatly. In view of the above problems, there are several methods to extract and identify the defects of the solder pads.

1. Research on the welding defect image recognition algorithm based on fuzzy theory proposed by Wang Peng [1]. This method is based on the automatic selection of thresholds and the use of fuzzy sets for gray-scale transformation to enhance image contrast; it uses the maximum inter-class variance algorithm (0TSu) Adaptive threshold to detect and extract the edge of the weld; identify different defects based on the geometric characteristics of the defect. For some small area defects, the solution is only one point after thinning, and it is deleted after denoising. Therefore, for some defects, the method described in this paper has a serious over segmentation phenomenon, and the technical methods need to be improved and improved.

2.Tang Guowei proposed to study weld defect based on fuzzy neural network, use the concept of fuzzy set to describe feature parameters, establish a fuzzy rule base of feature parameters, and construct fuzzy feature parameters as input layers and fuzzy rules as hidden layers. The fuzzy neural network model with layer and defect prediction recognition is classified as output. Analyzing the experimental results, the approximate location of the defect in the digital image and edge detection are successfully located; the scheme has a lower recognition rate for categories with a large feature set crossover. For example, for unwelded and unfused defects, the recognition rates are 92.30% and 88.46%, respectively. The misidentification of weld defects is also related to the accuracy of feature selection and the number of training samples. Welding seam texture feature extraction involves boundary definition and local information refinement. Traditional methods use binary method LTP, LBP, CLBP texture feature description methods, but the accuracy of texture description is affected. It also affects the judgment of defects in the weld.

Summary of the Invention

The purpose of the present invention is to provide a method for identifying defects in a weld film, which overcomes the lack of accuracy in amplitude and direction, and is superior in classification accuracy to the direct method of determining shape features and contours, and the traditional LTP, LBP, and CLBP texture feature methods. system.

In order to achieve the above-mentioned object, the method for identifying a defect of a weld negative provided by the present invention specifically includes: obtaining a weld negative of a weld by radiography; obtaining image texture feature data of a weld negative of a weld negative by using image recognition technology; The welding image texture feature data and the defect type of the weld backsheet are used to establish a recognition model by a machine learning algorithm; the weld image texture feature data of the weld backsheet to be inspected is obtained; and the weld image texture feature is based on the weld backsheet to be inspected The data and the identification model are used to calculate and obtain the defect category of the weld backsheet to be detected.

In the above-mentioned welding film defect recognition method, preferably, the use of image recognition technology to obtain the weld image texture feature data of the welding film includes: using the image recognition technology to obtain the size information, the symbol information, and the central region information of the welding film. , Obtaining the texture feature data of the weld image of the weld negative based on the size information, the symbol information, and the central area information.

In the above-mentioned welding film defect recognition method, preferably, obtaining the size information of the welding film by using the image recognition technology includes: obtaining the size information of the welding film by the following formula;

In the above formula, a and b are custom constants, TH is a predetermined threshold, and p is the number of pixels.

In the above-mentioned method for identifying the defect of a weld negative, preferably, obtaining the symbol information of the weld negative by using image recognition technology includes: obtaining the symbol information of the weld negative by calculating by the following formula;

In the above formula, CLTP_S ^* is the symbol information, TH is a predetermined threshold, and p is the number of pixels.

In the above-mentioned welding film defect recognition method, preferably, obtaining the center region information of the welding film by using image recognition technology includes: obtaining the center region information of the welding film by the following formula;

In the above formula, CLTP_C ^* is the central area information, and TH1 is the average value of the image pixels.

In the above-mentioned welding film defect recognition method, preferably, establishing a recognition model by a machine learning algorithm based on the texture image feature data of the welding seam and the defect category of the welding film further includes: obtaining the detection using a defect edge detection and tracking processing technology. A feature parameter of a predetermined category in the texture feature data of the weld image; a defect feature database is established according to the feature parameter and a defect category of the weld negative; and a recognition model is established by a machine learning algorithm according to the defect feature database.

In the above-mentioned welding film defect recognition method, preferably, the characteristic parameters include: welding plate number, image length pixel, image width pixel, gray difference between the defect and the background, relative position of the defect, gray deviation of the defect itself, Defect aspect ratio, equivalent area, circularity, entropy, correlation, moment of inertia, energy.

In the above-mentioned welding film defect recognition method, preferably, establishing a recognition model by a machine learning algorithm based on the welding image texture feature data and the welding seam film defect type includes: according to the welding seam image texture feature data and the welding seam The defect categories of the negatives are used to build an SVM model using an SVM (Support Vector Machine) classification algorithm.

In the above-mentioned welding film defect recognition method, preferably, the establishing the SVM model by the SVM classification algorithm further includes: obtaining a number of categories of the classifier in the SVM model by a global optimization method; and constructing each category according to a category difference exclusion method. The classifier; the categories of the classifier include: cracks, circular slag inclusions, strip slag inclusions, air holes, not welded through, and not fused.

The present invention also provides an electronic device including a memory, a processor, and a computer program stored on the memory and executable on the processor. When the processor executes the computer program, the method for identifying the above-mentioned weld film defect identification method is implemented. step.

The present invention also provides a computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned method for identifying a defect in a weld film are implemented.

The method and system for identifying the defect of a weld negative film provided by the present invention effectively solve the problems of inaccurate texture accuracy and inaccurate judgment of the characteristics of the weld negative identification, and greatly improve the accuracy of the edge detection and defect tracking identification of the weld negative image. Laying the foundation for development.

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings described herein are used to provide a further understanding of the present invention, and constitute a part of the present application, and do not constitute a limitation on the present invention. In the drawings:

FIG. 1 is a schematic flowchart of a method for identifying a defect of a weld negative film according to an embodiment of the present invention; FIG.

2 is a schematic diagram of a complete local three-valued mode (CLTP) calculation provided by an embodiment of the present invention;

3 is a schematic structural diagram of an SVM multi-class classifier according to an embodiment of the present invention;

4 is a schematic structural diagram of a welding film defect recognition system according to an embodiment of the present invention;

FIG. 5 is a schematic structural diagram of an electronic device according to an embodiment of the present invention.

detailed description

In order to make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the following further describes the present invention in detail with reference to the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and the description thereof are used to explain the present invention, but are not intended to limit the present invention.

In the description of this specification, the description with reference to the terms “an embodiment”, “a specific embodiment”, “for example”, etc. means that specific features, structures, materials, or characteristics described in conjunction with the embodiment or example are included in this application In at least one embodiment or example. In this specification, the schematic expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples. The sequence of steps involved in each embodiment is used to schematically illustrate the implementation of this application, and the sequence of steps therein is not limited, and can be appropriately adjusted as needed.

In recent years, China ’s pipeline construction has developed at a rapid rate of nearly 125,000 kilometers. In 2020, the future plan will reach 169,000 kilometers. The reliability of pipeline construction welding inspection technology directly affects the quality of pipeline construction and the safety of future operations. Imaging technology has been widely used in the pipeline welding inspection process. Ray imaging has laid the foundation for intelligent auxiliary evaluation. Although computer-assisted intelligent evaluation has developed rapidly, a certain number of negative film analysis processing systems have been put into use, but the application of these systems is not yet Mature, most of the current review work requires manual intervention and human-computer interaction. X-ray film defect recognition is through the digital processing of negatives and the high-speed processing capabilities of computers to convert the manual evaluation into image processing technology. The standardized image processing technology for weld negatives includes image preprocessing and segmentation of image weld areas. , Defect feature extraction, defect classification and recognition, and finally display the defect results. Therefore, a complete computer intelligent evaluation system is a very important research branch of radiographic inspection.

In order to effectively identify the defects of each welding seam according to the welding film, an embodiment of the present invention provides a welding film defect recognition method. Please refer to FIG. 1, the method specifically includes: S101 is obtained by radiography technology Weld seam negatives of the weld seam; S102 uses image recognition technology to obtain the seam image texture feature data of the seam negative film; S103 establishes a recognition model through a machine learning algorithm based on the seam image texture feature data and the defect type of the seam negative film; S104 Obtaining the texture feature data of the weld image of the to-be-detected backsheet; S105 calculating and obtaining the defect category of the weld plate of the to-be-tested according to the texture feature data of the weld image of the to-be-detected backsheet and the recognition model. In this embodiment, first, a radiographic image of the weld seam is obtained by using radiographic imaging technology, and then a machine model is trained based on the known defect types of the weld seam and the texture feature data of the weld seam image identified according to image recognition technology. The subsequent machine model obtains an identification model for identifying the type of the weld negative film defect; after obtaining the weld negative film to be inspected, the texture feature data of the weld image of the weld negative film to be inspected can be directly used with the recognition model. Determine the defect type of the weld; compared with the existing amplifier, this method not only has higher recognition accuracy, but also is more efficient.

In the image recognition technology in the above embodiments, there are two types of feature extraction methods: linear projection feature extraction and non-linear feature extraction methods; in linear projection analysis, the most representative are PCA (Principal Component Analysis) and LDA (Fisher Linear discriminant analysis); the extracted image features are mainly represented by texture features and morphological (shape) features. For welds, one is the extraction of geometric features of weld defects, and the other is the texture features of weld defects. Extraction. The extraction of geometric features is the quantification of the size, shape, ellipticity, and aspect ratio of defects. At present, the direct judgment method of shape features is generally used. There are cases where the judgment accuracy is insufficient and the accuracy rate is not high. Welding texture feature extraction is designed. Boundary definition, local information refinement, etc. Among them, texture features are more widely used, and their performance is more stable. Generally, their performance is better than simple shape feature description. Based on the foregoing, in an embodiment of the present invention, the obtaining the weld image texture feature data of the weld negative by using image recognition technology includes: obtaining the size information, symbol information, and center area information of the weld negative by using image recognition technology. The size information, the symbol information, and the central area information are used to obtain the texture feature data of the weld image of the weld negative. In this embodiment, the improved CTLP (Complete Local Local Ternary Patterns) algorithm proposed by the present invention is used to obtain the weld image texture feature data of the weld negative. The extracted weld image texture feature method retains richer textures. Structure information, based on the reconstruction of the central descriptor, including the symbol descriptor and the size descriptor; it is mainly divided into three components: two components contain size information, symbol information, and the other component retains the center area of the weld image Information, to achieve the fusion of texture features and weld shape features, the characterization data is closer to the actual situation.

In actual work, the extraction of weld texture features involves boundary sharpness and local information refinement. Traditional methods use binary method LTP, LBP, CLBP texture feature description methods, but the accuracy of texture description is affected, so continuous improvement is needed. Algorithm; To this end, in an embodiment of the present invention, a fully local three-valued CLTP mode is proposed, which includes three core operators, called the central descriptor, the symbol descriptor, and the size descriptor, respectively using CLTP_C, CLTP_S, and CLTP_M Represents that after transforming the operators, CLTP_C ^* , CLTP_S ^* , and CLTP_M ^* are finally obtained to represent the central area information, symbol information, and size information, respectively. Their calculation process is as follows:

CLTP_S ^* calculation of symbol information is equivalent to CLTP_M ^* calculation of size information:

among them,

Among them, the meaning of TH is a given fixed threshold, a and b are custom constants, and a = 0.3 and b = 0.7 are preferable.

As for the central area information CLTP_C ^* is equal to:

Among them, TH1 is the average value of image pixels, and then use

The modes are converted into encoded values, CLTP_S ^* CLTP_M ^* and CLTP_C ^* respectively corresponding to those shown in FIG. 2.

In the above embodiment, the CLTP algorithm mainly uses some amplitude information lost in the existing texture extraction algorithm to improve the recognition rate. After obtaining the CLTP feature, in order to achieve the purpose of the algorithm having complete rotation invariance, the present invention constructs a uniform The Fourier feature of the histogram of the pattern; specifically, to construct a fully local three-valued pattern, re-assign the symbol ps according to a given threshold and rewrite it to the following form: the center pixel is still gc, and the neighborhood radius Is R, the pixels in the neighborhood are gp, where the value of P is: p = [0,1 ..., P-1], the number of pixels is P, and the difference between the center pixel and the pixels in the neighborhood is Let dp = gp-gc. Among them, the value of dp consists of two parts, one is the symbol part, and the other is the absolute value of the difference between the values, which are denoted as sp and mp, respectively. Then dp can be expressed by multiplying the two parts, that is, dp = sp.mp , The symbolic part and numerical part can be expressed like this,

In order to construct a completely local three-valued pattern in the above embodiment, the symbol part can be re-assigned according to a given threshold, and finally modified CLTP_C ^* , CLTP_S ^* , CLTP_M ^{* are} used to represent the central area information, symbol information, and size information .

In the figure, a represents a 3 × 3 sample block with a center pixel of 38, and the eight pixels in the neighborhood are [27,72,69,32,25,43,26,88]; abs in b is absolute Value, TH = 24 is the obtained threshold, calculate the local difference, and the obtained operator result is [-11, 34, 31, -6, -13, 5, -12, 50]; in CL, CLTP_S ^* local The three-valued encoding vector of the difference sign is [0, 2, 2, 0, 0, 1, 0, 2]; in d, the CLTP_M ^* operator value is [1,2,2,0,1,0,1 , 2]; TH1 = 47.75 in e is the average value calculated from [27, 72, 69, 32, 25, 43, 26, 88], which indicates that the CLTP_C ^* value of CLTP is [01100101].

Before constructing the recognition model, a film defect library can also be established. In an embodiment of the present invention, establishing a recognition model by a machine learning algorithm based on the texture image data of the weld image and the defect category of the weld film also includes: using defect edges The detection and tracking processing technology obtains feature parameters of a predetermined category in the texture feature data of the weld image; establishes a defect feature database according to the feature parameters and the defect category of the weld negative; and establishes recognition through a machine learning algorithm according to the defect feature database model. The characteristic parameters include: pad number, image length pixel, image width pixel, gray difference between defect and background, relative position of defect, gray deviation of defect itself, defect aspect ratio, equivalent area, circle Degree, entropy, correlation, moment of inertia, energy. In actual work, this embodiment mainly integrates the above shape features and texture features, and the corresponding film blackness is different to form different defect features to construct a defect feature database shown in Table 1 below.

Table 1

The defect feature database contains parameters such as grayscale difference, equivalent area, circularity, entropy, correlation, etc., which facilitates the improvement of the recognition model by machine learning algorithms at the later stage. At the same time, the defect feature database further characterizes the characteristics of weld defects The size, symbol information and central area information, the algorithm overcomes the lack of precision in amplitude and direction, and is superior to the simple shape feature contour direct judgment method and the traditional LTP, LBP, CLBP texture feature methods in classification accuracy.

In a preferred embodiment of the present invention, establishing a recognition model by a machine learning algorithm based on the weld image texture feature data and the defect type of the weld backsheet includes: according to the weld image texture feature data and the defect of the weld backsheet The SVM model is established by the SVM classification algorithm. Wherein, the establishment of the SVM model by the SVM classification algorithm further includes: obtaining the number of classes of the classifier in the SVM model through a global optimization method; constructing the classifiers of each class according to a class difference exclusion method; Categories include: cracks, round slag inclusions, strip slag inclusions, pores, not welded, and not fused. In actual work, when constructing a multi-classifier SVM model, that is, an M-SVM classifier, a single class and the remaining classes can be used to determine the criteria for determining this classifier. Repeat the above process for all classes. Find out the value of each judgment function and compare these values. The largest category is the best classification structure of the sample. In this method, it is necessary to construct M classifiers, use the direct global optimization method (M-SVM), construct multiple batches of classifiers, and solve the M classifiers at one time; the specific method can be as follows:

Assume:

Is the sample training set, l is the number of samples, N is the number of samples, and M is the type of the sample. The non-negative relaxation variable ξ _i is introduced. The optimization problem of this method is as follows:

The discriminant function is: ST.y _i (w ^T φ (x _i ) + b) ≥1-ξ _i

Among them, ξ _i ≥0, i = 1,2, ... l, φ (*) is a non-linear mapping function from the input space to the high-dimensional feature space. By solving the optimization problem, the corresponding optimal decision function can be obtained. f (x):

In the above embodiment, each classifier structure should preferentially adopt the method of class difference exclusion, that is, in the overall structure, firstly, the similarity is regarded as a category, and then based on the small changes between the similarities, the model algorithm is used to distinguish. This classification has extremely high efficiency in classification accuracy and balance accuracy. Specifically, this paper constructs six types of defects, such as cracks, circular slag inclusions, strip slag inclusions, pores, unwelded, unfused, and other typical defects. As shown in Figure 3 below.

In order to test the effectiveness of the above-mentioned welding film defect recognition method, using the above model, firstly perform a full local three-valued mode CLTP texture recognition on the welding film, and use defect edge detection and tracking processing technology to calculate various parameters. The calculation parameters include: image length pixels, image width pixels, the gray difference between the defect and the background Δh, the relative position d of the defect, the gray deviation of the defect itself δ, the defect aspect ratio, the equivalent area S / C, and the circularity e. Entropy ENT, correlation COR, moment of inertia CON, energy parameters, etc. All the characteristic parameters are input to the SVM model, and the defects of the SVM welding film are identified, and the category of the defects is finally obtained. After the welded negatives are completely identified by the CLTP texture in the local three-valued mode, the SVM data classification technology based on the defect database is applied. The CTLP mode greatly improves the accuracy of edge detection and defect tracking identification of welded negatives. The SVM defect classification model enables automatic recognition and judgment of image defects. The accuracy has been greatly improved, and has basically reached the level of industrial application. The above method was used to verify that the X-ray film of the weld at the 7.28 accident section of the West-East China Gas Pipeline shows the unfused characteristic at the lower left of position 38, and its computer system judgment is completely consistent with the result of manual evaluation. The above method was used to verify that the second-line welded negative film of Shaanxi-Beijing in China showed the characteristics of incomplete penetration, and the main defects were located between the position points 6-7. This shows that the method has better accuracy and accuracy.

As shown in FIG. 4, an embodiment of the present invention further provides a welding film defect recognition system. The system includes a data acquisition unit, a model construction unit, and an identification unit. The data acquisition unit is used for radiography technology. Obtaining the weld seam film of the weld seam; and obtaining the seam image texture feature data of the weld seam film using image recognition technology; the model construction unit is configured to pass the machine according to the weld image texture feature data and the defect type of the weld seam film A learning algorithm establishes a recognition model; the recognition unit is configured to obtain texture feature data of a weld image of a negative plate of the weld to be tested; and calculate and obtain the texture feature data of the weld image of the negative plate of the weld to be tested and the recognition model Defect category of the negative to be tested.

The identification and description of the texture features and shape features of pipeline defects, and the accurate judgment of defects, are the problems of automatic computer image recognition and the main problems facing intelligent welding. The method and system for identifying weld film defects in the present invention solve the problem of inaccurate texture accuracy and feature judgment of weld film identification. An improved CTLP algorithm is provided, and the central descriptor is reconstructed, including the symbol descriptor. And size descriptors, which characterize the feature size, symbol information, and central area information of the weld defect; then establish a weld image defect feature library that includes shape features and texture features, image length pixels, image width pixels, defects and background Gray difference △ h, relative position of defect d, gray deviation of defect itself δ, defect aspect ratio, equivalent area S / C, circularity e, entropy ENT, correlation COR, moment of inertia CON, energy, etc. Parameters; at the same time, based on the defect feature database, a support vector machine (SVM) classification model is trained. Based on the defect feature database, the defect shape features are obtained by classification, and cracks, slag inclusions, pores, unwelded, unfused, Defect characteristics such as strip defects. In summary, the present invention uses the fusion processing technology of the improved CTLP algorithm and the SVM classification algorithm to make the weld defect feature description more accurate, and overcomes the shortcomings of the previous algorithms in terms of amplitude, directional accuracy, and easily confused defect features. The contour feature direct judgment method is superior to the shape feature in the accuracy rate and the traditional LTP, LBP, CLBP texture feature description method; the pipe weld seam defect SVM method based on CLTP texture & shape feature has better accuracy, which makes the welding negative image The accuracy of edge detection and defect tracking recognition has been greatly improved, laying a foundation for the development of weld feature recognition technology.

As shown in FIG. 5, the electronic device 600 may further include a communication module 110, an input unit 120, an audio processing unit 130, a display 160, and a power source 170. It is worth noting that the electronic device 600 does not necessarily need to include all the components shown in FIG. 5; in addition, the electronic device 600 may also include components not shown in FIG. 5, and reference may be made to the prior art.

As shown in FIG. 5, the central processing unit 100 is sometimes referred to as a controller or an operation control, and may include a microprocessor or other processor devices and / or logic devices. The central processing unit 100 receives input and controls each of the electronic devices 600. Operation of parts.

The memory 140 may be, for example, one or more of a buffer, a flash memory, a hard drive, a removable medium, a volatile memory, a non-volatile memory, or other suitable devices. The above-mentioned failure-related information can be stored, and a program for executing the related information can also be stored. In addition, the central processing unit 100 may execute the program stored in the memory 140 to implement information storage or processing.

The input unit 120 provides input to the central processing unit 100. The input unit 120 is, for example, a key or a touch input device. The power source 170 is used to provide power to the electronic device 600. The display 160 is used for displaying display objects such as images and characters. The display may be, for example, an LCD display, but is not limited thereto.

The memory 140 may be a solid-state memory, such as a read-only memory (ROM), a random access memory (RAM), a SIM card, and the like. It may also be a memory that holds information even when power is off, can be selectively erased and is provided with more data, and an example of the memory is sometimes called EPROM or the like. The memory 140 may also be some other type of device. The memory 140 includes a buffer memory 141 (sometimes referred to as a buffer). The memory 140 may include an application / function storage section 142 for storing application programs and function programs or a flow for performing operations of the electronic device 600 through the central processing unit 100.

The memory 140 may further include a data storage section 143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage section 144 of the memory 140 may include various drivers for the electronic device for communication functions and / or for performing other functions of the electronic device (such as a messaging application, an address book application, etc.).

The communication module 110 is a transmitter / receiver 110 that transmits and receives signals via the antenna 111. A communication module (transmitter / receiver) 110 is coupled to the central processing unit 100 to provide input signals and receive output signals, which may be the same as the case of a conventional mobile communication terminal.

Based on different communication technologies, multiple communication modules 110 may be provided in the same electronic device, such as a cellular network module, a Bluetooth module, and / or a wireless local area network module. The communication module (transmitter / receiver) 110 is also coupled to the speaker 131 and the microphone 132 via the audio processor 130 to provide audio output via the speaker 131 and to receive audio input from the microphone 132, thereby realizing general telecommunication functions. The audio processor 130 may include any suitable buffers, decoders, amplifiers, and the like. In addition, the audio processor 130 is also coupled to the central processing unit 100, so that the microphone 132 can be used to record on the unit, and the speaker 131 can be used to play the sound stored on the unit.

Those skilled in the art should understand that the embodiments of the present invention may be provided as a method, a system, or a computer program product. Therefore, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, and combinations of processes and / or blocks in the flowcharts and / or block diagrams can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing device to produce a machine, so that the instructions generated by the processor of the computer or other programmable data processing device are used to generate instructions Means for implementing the functions specified in one or more flowcharts and / or one or more blocks of the block diagrams.

These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to work in a particular manner such that the instructions stored in the computer-readable memory produce a manufactured article including an instruction device, the instructions The device implements the functions specified in one or more flowcharts and / or one or more blocks of the block diagram.

These computer program instructions can also be loaded on a computer or other programmable data processing device, so that a series of steps can be performed on the computer or other programmable device to produce a computer-implemented process, which can be executed on the computer or other programmable device. The instructions provide steps for implementing the functions specified in one or more flowcharts and / or one or more blocks of the block diagrams.

The specific embodiments described above further describe the objectives, technical solutions, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. The scope of protection, any modification, equivalent replacement, and improvement made within the spirit and principle of the present invention shall be included in the scope of protection of the present invention.

Claims

A method for identifying a defect in a weld negative, wherein the method includes:

Weld seam negatives obtained by radiography;

Use image recognition technology to obtain the weld image texture feature data of the weld negative;

Establishing a recognition model by a machine learning algorithm according to the texture feature data of the weld image and the defect category of the weld negative;

Obtaining the texture feature data of the weld image of the weld backsheet to be detected;

According to the texture feature data of the weld image of the to-be-detected backsheet and the recognition model, a defect category of the to-be-detected back-sheet is calculated and obtained.
The method for identifying defect of a weld negative according to claim 1, characterized in that said obtaining the texture feature data of the weld image of the weld negative by using image recognition technology comprises: obtaining the size information and symbols of the weld negative by using the image recognition technology Information and center area information, and obtain the seam image texture feature data of the weld backsheet according to the size information, the symbol information, and the center area information.
The method for identifying defect of a weld negative according to claim 2, wherein the obtaining the size information of the weld negative by using image recognition technology comprises: obtaining the size information of the weld negative by calculating by the following formula;

In the above formula, a and b are custom constants, TH is a predetermined threshold, and p is the number of pixels.
The method of claim 2, wherein obtaining the symbol information of the weld negative by using image recognition technology comprises: obtaining the symbol information of the weld negative by calculating the following formula;

In the above formula, CLTP_S * is the symbol information, TH is a predetermined threshold, and p is the number of pixels.
The method of claim 2, wherein obtaining the central area information of the weld negative by using image recognition technology comprises: obtaining the central area information of the weld negative by calculating the following formula;

In the above formula, CLTP_C * is the central area information, and TH1 is the average value of the image pixels.
The method of claim 1, wherein establishing a recognition model by a machine learning algorithm according to the texture feature data of the weld image and the defect type of the weld negative further comprises: using defect edge detection and tracking The processing technology obtains a characteristic parameter of a predetermined category in the texture feature data of the weld image; establishes a defect feature database according to the characteristic parameter and the defect category of the weld negative; and establishes a recognition model through a machine learning algorithm according to the defect feature database.
The method of claim 6, wherein the characteristic parameters include: welding plate number, image length pixel, image width pixel, gray difference between the defect and the background, relative position of the defect, defect Self-gray deviation, defect aspect ratio, equivalent area, circularity, entropy, correlation, moment of inertia, energy.
The method for identifying a defect of a weld negative according to claim 1, wherein establishing a recognition model by a machine learning algorithm based on the texture feature data of the weld image and the defect type of the weld negative includes: according to the texture of the weld image The SVM model is established by the SVM classification algorithm based on the feature data and the defect categories of the weld negative.
The method of claim 8, wherein the establishing an SVM model by using an SVM classification algorithm further comprises: obtaining a number of classes of the classifier in the SVM model by using a global optimization method; and according to class differences The exclusion method constructs the classifiers of each class; the classes of the classifier include: cracks, circular slag inclusions, strip slag inclusions, air holes, not welded through, and not fused.
An electronic device includes a memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that when the processor executes the computer program, any one of claims 1 to 9 is implemented The steps of the method for identifying a defect of a weld negative.
A computer-readable storage medium having stored thereon a computer program, characterized in that, when the computer program is executed by a processor, the steps of the method for identifying a defect in a weld film according to any one of claims 1 to 9 are implemented.