CN118015305A - Method and system for identifying same welding ray negative film - Google Patents
Method and system for identifying same welding ray negative film Download PDFInfo
- Publication number
- CN118015305A CN118015305A CN202211352784.5A CN202211352784A CN118015305A CN 118015305 A CN118015305 A CN 118015305A CN 202211352784 A CN202211352784 A CN 202211352784A CN 118015305 A CN118015305 A CN 118015305A
- Authority
- CN
- China
- Prior art keywords
- defect
- film
- model
- images
- image
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000003466 welding Methods 0.000 title claims abstract description 70
- 238000000034 method Methods 0.000 title claims abstract description 52
- 230000007547 defect Effects 0.000 claims abstract description 224
- 238000012549 training Methods 0.000 claims abstract description 41
- 238000013528 artificial neural network Methods 0.000 claims abstract description 33
- 230000008569 process Effects 0.000 claims abstract description 19
- 230000008859 change Effects 0.000 claims abstract description 18
- 238000012216 screening Methods 0.000 claims abstract description 9
- 230000004913 activation Effects 0.000 claims description 57
- 210000002569 neuron Anatomy 0.000 claims description 25
- 238000012163 sequencing technique Methods 0.000 claims description 16
- 238000004364 calculation method Methods 0.000 claims description 5
- 238000007621 cluster analysis Methods 0.000 claims description 4
- 239000000523 sample Substances 0.000 claims 1
- 238000001514 detection method Methods 0.000 abstract description 22
- 230000005855 radiation Effects 0.000 description 8
- 238000010586 diagram Methods 0.000 description 6
- 238000003384 imaging method Methods 0.000 description 5
- 230000008901 benefit Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 3
- 206010070834 Sensitisation Diseases 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000035945 sensitivity Effects 0.000 description 2
- 230000008313 sensitization Effects 0.000 description 2
- 238000012360 testing method Methods 0.000 description 2
- 238000012795 verification Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000007670 refining Methods 0.000 description 1
- 238000006467 substitution reaction Methods 0.000 description 1
- 230000000007 visual effect Effects 0.000 description 1
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for identifying identical plates of a welding ray negative, which comprises the following steps: establishing a historical film image library; training a first preset model by using a historical film image to obtain a first model for identifying defect classification characteristics, collecting dynamic information used for representing the change state of a neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index characteristics indicating defect characteristics on the film image; predicting defect classification characteristics and defect index characteristics of the negative film to be identified by utilizing each model respectively; and screening out images with defects of the same type as the negative film to be identified from an image library, marking the images as first images, comparing the first images with predicted defect index features, and diagnosing whether the images with the same welding features as the negative film to be identified exist in the image library. The invention can identify the weld defects and realize the detection of the weld defects.
Description
Technical Field
The invention belongs to the technical field of welding ray detection of refining devices, and particularly relates to a method and a system for identifying the same piece of a welding ray negative.
Background
Currently, radiation detection is one of the main means of welding nondestructive detection. The radiation detection is a workpiece quality detection method which adopts radiation (X-rays or gamma-rays) to transilluminate a workpiece and judges the quality condition of the inside of the detected workpiece through sensitization on a film, and the blackness of an image on the film and the distribution thereof after imaging reflect the type and the magnitude of defects inside the workpiece. Radiographic imaging has the characteristics of high resolution and high sensitivity, so that the information amount of film images obtained by radiographic detection is large. Meanwhile, the ray detection means is flexible, and the structural shapes of the detectable workpieces are more.
In carrying out the invention, the inventors have found that either an operator error or deliberate falsification may result in a large number of identical films having different weld numbers during the radiation detection process. When the engineering inspection is carried out, if only one welding seam image is counterfeited, whether the image is counterfeited or not can be judged according to whether the lap joint of the current welding seam image and the front and rear adjacent welding seam images is matched or not, and once the whole welding seam is counterfeited by re-shooting, the lap joint of each obtained welding seam image related to the welding seam is perfect, at the moment, the defects existing at the welding seam cannot be accurately detected, and the defects which are not detected possibly cause great economic loss and safety accidents.
Disclosure of Invention
To solve the above problems, an embodiment of the present invention provides a method for identifying identical sheets of welding ray film, including: establishing a historical film image library; training a first preset model by using a historical film image to obtain a first model for identifying defect classification characteristics, collecting dynamic information used for representing the change state of a neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index characteristics, wherein the defect index characteristics are used for indicating the defect characteristics on the film image; predicting defect classification features and defect index features of the negative film to be identified by using the first model and the second model respectively; and screening out images with defects of the same type as the negative film to be identified from the image library, marking the images as first images, comparing and analyzing the first images with predicted defect index features, and diagnosing whether the images with the same welding features as the negative film to be identified exist in the image library.
Preferably, the first preset model and the second preset model are constructed based on a Transformer neural network framework.
Preferably, in generating the first model, the method includes: marking the area covered by the defect on the historical film image, marking the corresponding defect type for the defect area, and then taking the defect area as input information of the first preset model and taking the defect type information as output information of the first preset model based on the first preset model, so that the first model is obtained through training of the first preset model.
Preferably, the defect index feature includes: coding information generated by the coding structure of the neural network; and activation information generated by neurons in the neural network.
Preferably, in the step of establishing the historical negative image library, further comprising: marking the historical negative image with a first defect index feature value and a second defect index feature value, wherein: performing cluster analysis on coding information corresponding to images with the same type of defects in the image library respectively to obtain clustering degree information for representing similarity of defect characteristics, and sequencing the coding information of a clustering center in the clustering degree information from high to low according to the clustering degree, so as to determine the first defect index characteristic value according to a sequencing result; and analyzing the activation information corresponding to the images with the defects of the same type in the image library to obtain activation degree data used for representing the activation state of the neurons and activation data used for representing the activation characteristics of the neurons, and sequencing the activation degree data of the neurons of each layer corresponding to the images with the defects of the same type from high to low, so that corresponding activation data is determined to be the second defect index characteristic value according to the sequencing result.
Preferably, the method uses the activation function corresponding to the neuron to obtain the corresponding activation data.
Preferably, in the step of comparing the first image with the predicted defect index features, the step of comparing the first image with the predicted defect index features comprises: obtaining a first defect index characteristic value and a second defect index characteristic value of the negative film to be identified according to the defect index characteristics of the negative film to be identified; respectively calculating the similarity between a first defect index characteristic value of the first image and a first defect index characteristic value of the negative film to be identified, and respectively calculating the similarity between a second defect index characteristic value of the first image and a second defect index characteristic value of the negative film to be identified; and sequencing the first images according to the similarity calculation result, so that the same piece of the negative to be identified is determined in the historical negative image library according to the sequencing result.
Preferably, the first defect index characteristic value is 5-10 pieces of coding information with the top cluster degree order; the second defect index characteristic value is activation data of 50-100 neurons with the activation degree ordered at the front; the same film is 5 film images with the top similarity sequence in the history film image library.
Preferably, the similarity is obtained by calculating cosine similarity.
In another aspect, the present invention also provides a system for identifying identical sheets of welding ray film, comprising: the image library establishing module is used for establishing a historical film image library; the model generation module is used for training a first preset model by utilizing a historical film image to obtain a first model for identifying defect classification characteristics, collecting dynamic information used for representing the change state of a neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index characteristics, wherein the defect index characteristics are used for indicating the defect characteristics on the film image; a defect feature recognition module for respectively predicting defect classification features and defect index features of the negative film to be recognized by using the first model and the second model; the same film identification module is used for screening out images with defects of the same type as the negative film to be identified from the image library, marking the images as first images, comparing and analyzing the first images with predicted defect index features, and diagnosing whether the images with the same welding features as the negative film to be identified exist in the image library.
One or more embodiments of the above-described solution may have the following advantages or benefits compared to the prior art:
The invention provides a method for identifying identical plates of a welding ray film. According to the method, a preset model is trained through a pre-established historical film image library, an identification model for identifying defect classification features is obtained, and dynamic information used for representing the change state of the neural network in the training process is collected while the preset model is trained. And then training another preset model by utilizing the collected dynamic change information to obtain an identification model for identifying defect index features, wherein the defect index features are used for indicating defect features on the negative film image. Finally, identifying the defect classification features and defect index features of the negative film to be identified by using the corresponding identification model, so that the images with the same welding features as the negative film to be identified are determined from the historical negative film image library based on the welding features (defect classification features and defect index features) of the current negative film to be identified, namely: the same piece. The invention uses the computer and the film scanner to generate the welding ray film image related to the original film of the welding ray film, thereby realizing the digital management of the welding ray film and the identification and detection of the welding seam defect.
Additional features and advantages of the invention will be set forth in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. The objectives and other advantages of the invention will be realized and attained by the structure particularly pointed out in the written description and claims hereof as well as the appended drawings.
Drawings
The accompanying drawings are included to provide a further understanding of the invention, and are incorporated in and constitute a part of this specification, illustrate the invention and together with the embodiments of the invention, serve to explain the invention, without limitation to the invention. In the drawings:
FIG. 1 is a step diagram of a method for identifying identical sheets of welded ray film in accordance with an embodiment of the present application.
FIG. 2 is a schematic diagram of a neural network framework for a method for identifying identical sheets of welded ray films, in accordance with an embodiment of the present application.
FIG. 3 is a block diagram of a system for identifying identical sheets of welding ray film in accordance with an embodiment of the present application.
Detailed Description
The following will describe embodiments of the present invention in detail with reference to the drawings and examples, thereby solving the technical problems by applying technical means to the present invention, and realizing the technical effects can be fully understood and implemented accordingly. It should be noted that, as long as no conflict is formed, each embodiment of the present invention and each feature of each embodiment may be combined with each other, and the formed technical solutions are all within the protection scope of the present invention.
Additionally, the steps illustrated in the flowcharts of the figures may be performed in a computer system such as a set of computer executable instructions, and although a logical order is illustrated in the flowcharts, in some cases the steps illustrated or described may be performed in an order other than that herein.
Currently, radiation detection is one of the main means of welding nondestructive detection. The radiation detection is a workpiece quality detection method which adopts radiation (X-rays or gamma-rays) to transilluminate a workpiece and judges the quality condition of the inside of the detected workpiece through sensitization on a film, and the blackness of an image on the film and the distribution thereof after imaging reflect the type and the magnitude of defects inside the workpiece. Radiographic imaging has the characteristics of high resolution and high sensitivity, so that the information amount of film images obtained by radiographic detection is large. Meanwhile, the ray detection means is flexible, and the structural shapes of the detectable workpieces are more.
In carrying out the invention, the inventors have found that either an operator error or deliberate falsification may result in a large number of identical films having different weld numbers during the radiation detection process. When the engineering inspection is carried out, if only one welding seam image is counterfeited, whether the image is counterfeited or not can be judged according to whether the lap joint of the current welding seam image and the front and rear adjacent welding seam images is matched or not, and once the whole welding seam is counterfeited by re-shooting, the lap joint of each obtained welding seam image related to the welding seam is perfect, at the moment, the defects existing at the welding seam cannot be accurately detected, and the defects which are not detected possibly cause great economic loss and safety accidents.
Therefore, in order to solve the problem that an engineering party re-shoots a history welding line with good quality by using a new mark to impersonate the ray detection and counterfeiting of the welding line with poor quality, the invention provides a method for identifying the same piece of a welding ray negative film. According to the method, a preset model is trained through a pre-established historical film image library, an identification model for identifying defect classification features is obtained, and dynamic information used for representing the change state of the neural network in the training process is collected while the preset model is trained. And then training another preset model by utilizing the collected dynamic change information to obtain an identification model for identifying defect index features, wherein the defect index features are used for indicating defect features on the negative film image. Finally, identifying the defect classification features and defect index features of the negative film to be identified by using the corresponding identification model, so that the images with the same welding features as the negative film to be identified are determined from the historical negative film image library based on the welding features (defect classification features and defect index features) of the current negative film to be identified, namely: the same piece. The invention realizes the digital management of the welding ray film, can identify the weld defect on the welding ray film image, and can judge whether the welding ray film is counterfeited or not.
Example 1
According to the technical rules of X-ray flaw detection, each welding ray negative produced by ray detection is provided with an image quality meter image mark, a positioning mark (a center mark and a lap mark) and an identification mark, wherein the identification mark comprises: line number, weld number, welding job number, welding date, etc. In addition, if the welded ray film needs to be reworked, the film is also marked with a reworking mark and a footer R 1、R2……Rn for representing the reworking times. In order to ensure that the obtained welding ray film can cover all the welding lines of the whole welding junction, a plurality of X-ray flaw detection films are usually required to be shot aiming at the same welding junction. In the shooting process, a small amount of overlapping needs to be carried out on the shooting position of the welding line, and the lead lap mark on the negative film is ensured to be positioned at the lap position of the welding line. In the case of weld imaging, the various marks are set at specified positions in a type form in advance, and the marks are imaged together with the weld. Thus, a welded ray film having various marks was obtained.
FIG. 1 is a step diagram of a method for identifying identical sheets of welded ray film in accordance with an embodiment of the present application. The following describes in detail a method for identifying the same sheet as the welding ray negative sheet according to the embodiment of the present application with reference to fig. 1.
As shown in fig. 1, in step S110, a plurality of history welding ray films with various marks are obtained, each history film is used as a raw film by a computer and/or a film scanner to generate a corresponding history film image, and then a history film image library including all history film images is built. Before training the preset model, the embodiment divides the historical negative image into a training set, a test set and a verification set according to a specified proportion. The historical negative film images in the training set are used for training a preset model; the historical film images in the verification set are used for verifying the network performance (such as accuracy rate and recall rate) of the current preset model, and training is stopped when the network performance is verified to be converged (the network performance is not improved any more); the historical negative images in the test set are then used to evaluate the network performance of the finally obtained model. In addition, the embodiment can also extract the corresponding historical film images from the historical film image library by searching the storage path, the size, the storage date, the feature codes and the labels of the historical film images.
In step S120, the first preset model is trained using the historical negative image, so as to obtain a first model for identifying the defect classification feature. Specifically, the historical film image is input into a preset first preset model to obtain defect classification characteristics, and the first model is obtained based on training of the first preset model. The first model can identify defect classification characteristics of the historical negative film image and generate corresponding defect classification characteristic information.
Next, a detailed description will be given of a process of generating the first model in the embodiment of the present application.
Firstly, marking an area covered by a defect on a historical film image, marking a corresponding defect type for the defect area, then taking the defect area as input information of a first preset model based on the first preset model, and taking the defect type information as output information of the first preset model, so that the first model is obtained through training of the first preset model. In the embodiment of the application, the area covered by the welding defect on each historical film image in the historical film image library is determined, so that the defect type marking information used for representing the defect classification characteristic is marked on the corresponding welding defect area on each historical film image. And substituting the welding defect area marked with the defect type marking information into the first preset model based on the preset first preset model, taking the defect classification characteristic information as output information of the first preset model, starting training the first preset model, and obtaining the first model after training is completed.
Next, collecting dynamic information used for representing the change state of the neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index features, wherein the defect index features are used for indicating defect features on the negative image. Specifically, the first preset model is constructed based on a neural network structure. When the first preset model is trained by using the historical film images, the active state of the neural network forming the first preset model changes along with different defect characteristics corresponding to the historical film images, so that the embodiment collects dynamic information representing the change state of the neural network for the whole training process of the first model so as to mark corresponding information representing the change state of the neural network for the defect area of each historical film image. And substituting the defect area marked with the information representing the change state of the neural network into a second preset model by taking the defect area as input information of the second preset model, taking the information representing the change state of the neural network (namely, defect index features for indicating defect features on the historical film image) as output information of the second preset model, starting training the second preset model, and obtaining the second model after training is completed.
In the embodiment of the application, the first preset model and the second preset model are formed by adopting a framework based on a transducer neural network. FIG. 2 is a schematic diagram of a neural network framework for a method for identifying identical sheets of welded ray films, in accordance with an embodiment of the present application. Referring to fig. 2, a detailed description will be given of a construction process of the model by taking a first preset model as an example.
Specifically, the first preset model is formed by a visual transducer neural network based framework. Firstly, each historical negative image is split into a plurality of small blocks according to a specified size, each small block is mapped into a one-dimensional vector through linear mapping, then each mapped small block is input into an embedding layer, and the format of the historical negative image is converted into an input format (namely, a vector sequence) of a coding structure in the embedding layer. The vector sequence is then transmitted to the encoding structure to classify defects having the same defect characteristics. And finally, identifying defect classification characteristics by utilizing a feedforward network, and outputting corresponding defect classification characteristic information. The encoding structure of the present embodiment mainly includes: self-care layer and neural network. Since the self-attention mechanism corresponding to the self-attention layer does not contain a positional relationship, each small block needs to be position-coded before the vector sequence related to the history negative image is input into the coding structure.
Further, the defect index features of the present embodiment include: coding information generated by the coding structure of the neural network; and activation information generated by neurons in the neural network. The defect index feature is information of a neural network change state of a first preset model in the process of obtaining the first model, and specifically comprises the following steps: when the vector sequence related to the historical film image enters the coding structure, the coding structure identifies the coding information of the defect feature generated after the defect feature, and when the defect feature is identified, the activation degree feature and the activation feature in the activation information of each layer of neurons in the neural network.
Next, the present embodiment marks the first defect index feature value and the second defect index feature value for the history negative image.
In the step of marking the first defect index feature value, cluster analysis is carried out on coding information corresponding to images with the same type of defects in an image library, clustering degree information used for representing similarity of defect features is obtained, coding information of a clustering center in the clustering degree information is ranked from high to low according to the clustering degree, and therefore the first defect index feature value is determined according to a ranking result. Specifically, the historical film image library comprises a plurality of images with defects of the same type, in other words, the defects of each image with the defects of the same type belong to the same defect type and have different defect characteristics. In the process of training a first preset model, collecting the coding information of defect characteristics generated by a coding structure, extracting the coding information of all defect characteristics of a plurality of historical film images under the current defect classification characteristics according to the defect classification characteristic information identified by a feedforward network, performing cluster analysis to obtain the clustering degree of the similarity of the characteristic defect characteristics corresponding to each defect characteristic under the current defect classification characteristics, and determining the coding information of the defect characteristics in the clustering center. And then, sorting the coding information of the defect features in the clustering center according to the order of the clustering degree from high to low, screening the coding information with the front clustering degree as a first defect index feature value, and marking the first defect index feature value for each historical negative image under the current defect classification feature.
In a specific embodiment of the present application, the first defect index feature value uses 5 to 10 pieces of encoded information that are ranked first by the clustering degree.
In the step of marking the second defect index feature value, activation information corresponding to the images with the same type of defects in the image library is analyzed to obtain activation degree data used for representing the activation state of the neurons and activation data used for representing the activation characteristics of the neurons, and the activation degree data of the neurons of each layer corresponding to the images with the same type of defects are ranked from high to low, so that corresponding activation data is determined to be the second defect index feature value according to the ranking result. Specifically, in the process of training the first preset model, the activation state of each layer of neurons in the neural network in the coding structure under the current defect classification characteristic is analyzed, and activation degree data is calculated according to the activation degree of the neurons in the activation state. And then, sequencing the activation degree data in each layer of the neural network according to the order of the activation degree from high to low, and screening the activation degree data with the activation degree being the front in each layer. And finally, acquiring the activation data of the neurons to which the activation degree data belong according to the screened activation degree data, taking the activation data as activation characteristics of the neurons, and marking the activation data as second defect index characteristic values on each historical negative film image under the current defect classification characteristics.
In one embodiment of the present application, the second defect index feature value uses activation data of the top 50-100 neurons ordered by activation level.
And marking the first defect characteristic value and the second defect characteristic value on the historical negative image by adopting the marking method of the first defect characteristic value and the second defect characteristic value, so that a second preset model is trained by using the historical negative image marked with the first defect characteristic value and the second defect characteristic value, and a second model is obtained.
Further, the present embodiment obtains corresponding activation data by using an activation function corresponding to the neuron. Specifically, the activation data reflects the activation characteristics of the neurons, so the present embodiment adopts the output values of the activation functions of the neurons as the corresponding activation data.
After each model is obtained by training, in step S130, the defect classification feature and the defect index feature of the negative film to be identified are respectively predicted by using the first model and the second model, so as to obtain the defect classification feature and the defect index feature of the current negative film to be identified.
Further, in step S140, an image having the same type of defect as the negative film to be identified is screened from the image library and recorded as a first image, and the first image is compared with the predicted defect index feature to diagnose whether an image having the same welding feature as the negative film to be identified exists in the image library. Firstly, screening historical film images which are the same as the defect classification characteristics of the current film to be identified from a historical film image library according to the defect classification characteristics of the current film to be identified, and recording the screened historical film images as first images. And then, analyzing the similarity degree between the defect index features marked on the screened historical film images and the defect index features of the film to be identified currently, sequencing the screened historical film images according to the sequence from high similarity degree to low similarity degree, screening the historical film images with the front similarity degree, and finally judging whether the historical film images with the front similarity degree and the film to be identified have the same welding features or not to obtain corresponding identical films.
When comparing and analyzing the first image and the predicted defect index feature, firstly, obtaining the defect index feature of the current negative to be identified, and obtaining a first defect index feature value and a second defect index feature value of the negative to be identified; then, respectively calculating the similarity between the first defect index characteristic value of each historical film image corresponding to the first image and the first defect index characteristic value of the film to be identified, marking the similarity as the first similarity, and respectively calculating the similarity between the second defect index characteristic value of each historical film image corresponding to the first image and the second defect index characteristic value of the film to be identified, marking the similarity as the second similarity; then, the first similarity and the second similarity are integrated in a summation mode and the like, so that a similarity calculation result is obtained; and finally, sequencing each historical film image corresponding to the first image according to the sequence from high to low of the similarity calculation result, thereby determining the historical film image with the front sequencing as the same film of the film to be identified.
In one embodiment of the application, the same film uses top 5 film images in the historical film image library that are ranked in similarity.
Further, the present embodiment adopts a cosine similarity calculation method to calculate the first similarity and the second similarity.
Example two
Based on the method for identifying the same sheet of the welding ray film according to the first embodiment, the embodiment of the present application further provides a system for identifying the same sheet of the welding ray film (hereinafter referred to as "same sheet identification system"). FIG. 3 is a block diagram of a system for identifying identical sheets of welding ray film in accordance with an embodiment of the present application.
As shown in fig. 3, the identical piece identification system in the embodiment of the present invention includes: an image library creation module 31, a model generation module 32, a defect feature recognition module 33, and a same piece recognition module 34. Specifically, the image library creating module 31 is implemented according to the method described in the above step S110, and is configured to create a history film image library; the model generating module 32 is configured to train the first preset model by using the historical film image to obtain a first model for identifying the defect classification feature, collect dynamic information used for representing the change state of the neural network in the training process of the first model, train the second preset model based on the dynamic information, and obtain a second model for identifying the defect index feature, wherein the defect index feature is used for indicating the defect feature on the film image; the defect feature recognition module 33 is implemented according to the method described in the above step S130, and is configured to predict the defect classification feature and the defect index feature of the negative film to be recognized by using the first model and the second model generated by the model generation module 32, respectively; the same film identifying module 34 is configured to screen out the images having the same type of defects as the film to be identified from the image library and record the images as the first images, compare the first images with the predicted defect index features, and diagnose whether the images having the same welding features as the film to be identified exist in the image library according to the method described in the step S140.
The invention discloses a method for identifying identical plates of a welding ray negative. According to the method, a preset model is trained through a pre-established historical film image library, an identification model for identifying defect classification features is obtained, and dynamic information used for representing the change state of the neural network in the training process is collected while the preset model is trained. And then training another preset model by utilizing the collected dynamic change information to obtain an identification model for identifying the defect index features indicating the defect features on the negative film image. Finally, identifying the defect classification features and defect index features of the negative film to be identified by using the corresponding identification model, so that the images with the same welding features as the negative film to be identified are determined from the historical negative film image library based on the welding features (defect classification features and defect index features) of the current negative film to be identified, namely: the same piece. The invention realizes the digital management of the welding ray film and the identification and detection of the welding seam defect.
The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.
Of course, the present invention is capable of other various embodiments and its several details are capable of modification and variation in light of the present invention, as will be apparent to those skilled in the art, without departing from the spirit and scope of the invention as defined in the appended claims.
It will be appreciated by those skilled in the art that the modules or steps of the invention described above may be implemented in a general purpose computing device, they may be centralized on a single computing device, or distributed over a network of computing devices, or they may alternatively be implemented in program code executable by computing devices, such that they may be stored in a memory device and executed by computing devices, or they may be separately fabricated as individual integrated circuit modules, or multiple modules or steps within them may be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any specific combination of hardware and software.
Although the embodiments of the present invention are described above, the embodiments are only used for facilitating understanding of the present invention, and are not intended to limit the present invention. Any person skilled in the art can make any modification and variation in form and detail without departing from the spirit and scope of the present disclosure, but the scope of the present disclosure is still subject to the scope of the appended claims.
Claims (10)
1. A method for identifying identical sheets of welded ray film, comprising:
Establishing a historical film image library;
training a first preset model by using a historical film image to obtain a first model for identifying defect classification characteristics, collecting dynamic information used for representing the change state of a neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index characteristics, wherein the defect index characteristics are used for indicating the defect characteristics on the film image;
Predicting defect classification features and defect index features of the negative film to be identified by using the first model and the second model respectively;
And screening out images with defects of the same type as the negative film to be identified from the image library, marking the images as first images, comparing and analyzing the first images with predicted defect index features, and diagnosing whether the images with the same welding features as the negative film to be identified exist in the image library.
2. The method of claim 1, wherein the first and second pre-set models are constructed using a Transformer neural network based framework.
3. The method according to claim 1 or 2, characterized in that in generating the first model, it comprises:
Marking the area covered by the defect on the historical film image, marking the corresponding defect type for the defect area, and then taking the defect area as input information of the first preset model and taking the defect type information as output information of the first preset model based on the first preset model, so that the first model is obtained through training of the first preset model.
4. A method according to any one of claims 1 to 3, wherein the defect index feature comprises:
coding information generated by the coding structure of the neural network; and
Activation information generated by neurons in the neural network.
5. The method of claim 4, wherein in the step of building a library of historical negative images, further comprising: marking the historical negative image with a first defect index feature value and a second defect index feature value, wherein:
performing cluster analysis on coding information corresponding to images with the same type of defects in the image library respectively to obtain clustering degree information for representing similarity of defect characteristics, and sequencing the coding information of a clustering center in the clustering degree information from high to low according to the clustering degree, so as to determine the first defect index characteristic value according to a sequencing result; and
And respectively analyzing the activation information corresponding to the images with the defects of the same type in the image library to obtain activation degree data used for representing the activation state of the neurons and activation data used for representing the activation characteristics of the neurons, and sequencing the activation degree data of the neurons of each layer corresponding to the images with the defects of the same type from high to low, so that corresponding activation data is determined to be the index characteristic value of the second defect according to the sequencing result.
6. The method of claim 5, wherein the method uses an activation function corresponding to a neuron to obtain the corresponding activation data.
7. The method according to claim 5 or 6, wherein in the step of comparing the first image with the predicted defect index features, comprising:
Obtaining a first defect index characteristic value and a second defect index characteristic value of the negative film to be identified according to the defect index characteristics of the negative film to be identified;
Respectively calculating the similarity between a first defect index characteristic value of the first image and a first defect index characteristic value of the negative film to be identified, and respectively calculating the similarity between a second defect index characteristic value of the first image and a second defect index characteristic value of the negative film to be identified;
And sequencing the first images according to the similarity calculation result, so that the same piece of the negative to be identified is determined in the historical negative image library according to the sequencing result.
8. The method of claim 7, wherein the step of determining the position of the probe is performed,
The first defect index characteristic value is 5-10 pieces of coding information with the top cluster degree order;
the second defect index characteristic value is activation data of 50-100 neurons with the activation degree ordered at the front;
The same film is 5 film images with the top similarity sequence in the history film image library.
9. The method according to claim 7 or 8, characterized in that the similarity is obtained by calculating cosine similarity.
10. A system for identifying identical sheets of welding ray film, the system comprising the following modules:
the image library establishing module is used for establishing a historical film image library;
the model generation module is used for training a first preset model by utilizing a historical film image to obtain a first model for identifying defect classification characteristics, collecting dynamic information used for representing the change state of a neural network in the training process of the first model, and training a second preset model based on the dynamic information to obtain a second model for identifying defect index characteristics, wherein the defect index characteristics are used for indicating the defect characteristics on the film image;
a defect feature recognition module for respectively predicting defect classification features and defect index features of the negative film to be recognized by using the first model and the second model;
The same film identification module is used for screening out images with defects of the same type as the negative film to be identified from the image library, marking the images as first images, comparing and analyzing the first images with predicted defect index features, and diagnosing whether the images with the same welding features as the negative film to be identified exist in the image library.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211352784.5A CN118015305A (en) | 2022-10-31 | 2022-10-31 | Method and system for identifying same welding ray negative film |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211352784.5A CN118015305A (en) | 2022-10-31 | 2022-10-31 | Method and system for identifying same welding ray negative film |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN118015305A true CN118015305A (en) | 2024-05-10 |
Family
ID=90943421
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211352784.5A Pending CN118015305A (en) | 2022-10-31 | 2022-10-31 | Method and system for identifying same welding ray negative film |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN118015305A (en) |
-
2022
- 2022-10-31 CN CN202211352784.5A patent/CN118015305A/en active Pending
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109086756A (en) | A kind of text detection analysis method, device and equipment based on deep neural network | |
| JP2019114116A (en) | Data generation device, data generation method, and data generation program | |
| CN109344905A (en) | A kind of transmission facility automatic fault recognition methods based on integrated study | |
| JPH07230552A (en) | Method and equipment for image quality analysis | |
| CN109658387A (en) | The detection method of the pantograph carbon slide defect of power train | |
| US11715198B2 (en) | Medical use artificial neural network-based medical image analysis apparatus and method for evaluating analysis results of medical use artificial neural network | |
| CN108711148B (en) | Tire defect intelligent detection method based on deep learning | |
| CN110309768B (en) | Method and equipment for detecting staff at vehicle inspection station | |
| KR101294218B1 (en) | System and method for evaluating quality of polarized film | |
| US20080275747A1 (en) | Incident/accident report analysis apparatus and method | |
| CN105913059A (en) | Vehicle VIN code automatic identifying system and control method therefor | |
| US11464466B2 (en) | Methods and systems for periodontal disease screening | |
| CN111353611B (en) | Nuclear power station in-service inspection large repair inspection report automatic generation system and method | |
| CN107958204A (en) | Reference report recognition methods, device, computer equipment and storage medium | |
| CN111353304B (en) | Crowdsourcing test report aggregation and summarization method | |
| CN117372751A (en) | Pipeline defect detection method, device and storage medium | |
| CN118015305A (en) | Method and system for identifying same welding ray negative film | |
| Zhang et al. | Automatic forgery detection for x-ray non-destructive testing of welding | |
| CN116863122A (en) | Ammeter meter reading processing method, device, cloud end, system and medium | |
| KR102337692B1 (en) | Method of detecting error-suspicious bounding box and computer apparatus conducting thereof | |
| CN115423421A (en) | Method and device for automatically auditing process of inspection report, electronic equipment and medium | |
| CN113327023A (en) | Traversal test method and device, electronic equipment and computer readable storage medium | |
| CN117994191A (en) | Method for identifying defects of welding ray film | |
| KR20240064956A (en) | Pipe wall-thinning condition diagonosis system using ai vision technology and diagonosis method using this | |
| CN117576007A (en) | Method, device, equipment and storage medium for evaluating negative film image |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination |