A method and system for identifying identical frames of weld radiographic film

Abstract

The application discloses a method for identifying same weld X-ray films, which comprises the following steps: establishing a historical film image library; training a first preset model by using the historical film images, obtaining a first model for identifying defect classification features, and collecting dynamic information for representing the change state of a neural network in the training process of the first model, and based on the dynamic information, training a second preset model to obtain a second model for identifying defect index features indicating defect features on a film image; respectively predicting defect classification features and defect index features of a to-be-identified film by using the models; screening images with same type defects from the film image library as first images, and comparing and analyzing the first images with the predicted defect index features to diagnose whether there are images with same weld features as the to-be-identified film in the film image library. The application can identify weld defects and realize the detection of weld defects.

Description

Technical Field

[0001] This invention belongs to the field of welding radiographic testing technology for refining and chemical equipment, and in particular relates to a method and system for identifying identical welding radiographic films. Background Technology

[0002] Currently, radiographic testing is one of the main methods for non-destructive testing of welding. Radiographic testing uses X-rays or gamma rays to illuminate the workpiece, and the internal quality of the workpiece is determined by the light captured on film. The density and distribution of the image on the film reflect the type and magnitude of defects within the workpiece. Radiographic imaging features high resolution and high sensitivity, resulting in a large amount of information in the obtained film images. Furthermore, radiographic testing is flexible and can inspect a wide variety of workpiece structures and shapes.

[0003] In the process of developing this invention, the inventors discovered that operational errors or deliberate falsification can lead to a large number of identical films with different weld numbers appearing during radiographic testing. During project acceptance, if only one weld image is falsified, it can be determined whether the image is falsified by checking whether the overlap between the current weld image and adjacent weld images matches. However, if falsification is achieved by re-photographing the entire weld joint, the resulting weld images related to the joint will overlap perfectly. In this case, defects at the weld joint cannot be accurately detected, and undetected defects may lead to significant economic losses and safety accidents. Summary of the Invention

[0004] To address the aforementioned problems, this invention provides a method for identifying identical welding radiographs, comprising: establishing a historical radiograph image library; training a first preset model using the historical radiograph images to obtain a first model for identifying defect classification features, and collecting dynamic information used to characterize the changing state of the neural network during the training of the first model; based on this, training a second preset model to obtain a second model for identifying defect index features, wherein the defect index features are used to indicate defect features on the radiograph image; using the first model and the second model to predict the defect classification features and defect index features of the radiograph to be identified, respectively; selecting images with the same type of defects as the radiograph to be identified from the image library and recording them as first images, and comparing and analyzing the first images with the predicted defect index features to diagnose whether there are images with the same welding features as the radiograph to be identified in the image library.

[0005] Preferably, the first preset model and the second preset model are constructed using a Transformer neural network framework.

[0006] Preferably, the process of generating the first model includes: marking the area covered by defects on the historical film image and marking the defect area with the corresponding defect type; then, based on the first preset model, using the defect area as the input information of the first preset model and the defect type information as the output information of the first preset model, thereby obtaining the first model by training the first preset model.

[0007] Preferably, the defect index features include: encoding information generated by the encoding structure of the neural network; and activation information generated by neurons in the neural network.

[0008] Preferably, the step of establishing a historical film image library further includes: labeling the historical film images with a first defect index feature value and a second defect index feature value, wherein: cluster analysis is performed on the encoding information corresponding to images with the same type of defect in the image library to obtain clustering degree information for characterizing the similarity of defect features, and the encoding information of the cluster centers in the clustering degree information is sorted from high to low according to the clustering degree, thereby determining the first defect index feature value according to the sorting result; and the activation information corresponding to images with the same type of defect in the image library is analyzed to obtain activation degree data for characterizing the activation state of neurons and activation data for representing the activation features of neurons, and the activation degree data of each layer of neurons corresponding to images with the same type of defect are sorted from high to low, thereby determining the corresponding activation data as the second defect index feature value according to the sorting result.

[0009] Preferably, the method utilizes the activation function corresponding to the neuron to obtain the corresponding activation data.

[0010] Preferably, the step of comparing and analyzing the first image with the predicted defect index features includes: obtaining a first defect index feature value and a second defect index feature value of the film to be identified based on the defect index features of the film to be identified; calculating the similarity between the first defect index feature value of the first image and the first defect index feature value of the film to be identified, and calculating the similarity between the second defect index feature value of the first image and the second defect index feature value of the film to be identified; sorting each of the first images according to the similarity calculation results, thereby determining the same film of the film to be identified in the historical film image library according to the sorting results.

[0011] Preferably, the first defect index feature value is the top 5 to 10 encoded information values ​​ranked by clustering degree; the second defect index feature value is the activation data of the top 50 to 100 neurons ranked by activation degree; and the identical film refers to the top 5 film images ranked by similarity in the historical film image library.

[0012] Preferably, the similarity is obtained by calculating cosine similarity.

[0013] On the other hand, the present invention also provides a system for identifying identical films on welding radiographs, comprising: an image library establishment module for establishing a historical film image library; a model generation module for training a first preset model using historical film images to obtain a first model for identifying defect classification features, and collecting dynamic information used to characterize the changing state of the neural network during the training of the first model, and based on this, training a second preset model to obtain a second model for identifying defect index features, wherein the defect index features are used to indicate defect features on the film image; a defect feature identification module for predicting the defect classification features and defect index features of the film to be identified using the first model and the second model, respectively; and an identical film identification module for selecting images with the same type of defects as the film to be identified from the image library and recording them as first images, and comparing and analyzing the first images with the predicted defect index features to diagnose whether there are images with the same welding features as the film to be identified in the image library.

[0014] Compared with the prior art, one or more embodiments of the above solutions may have the following advantages or beneficial effects:

[0015] This invention proposes a method for identifying identical weld radiographs. The method trains a pre-established historical film image database to train a pre-defined model, obtaining a recognition model for identifying defect classification features. Simultaneously, dynamic information representing the changing states of the neural network is collected during training. Then, the collected dynamic information is used to train another pre-defined model, obtaining a recognition model for identifying defect index features, where the defect index features indicate defect characteristics on the film image. Finally, the corresponding recognition model is used to identify the defect classification features and defect index features of the film to be identified. Based on the welding features (defect classification features and defect index features) of the current film to be identified, images with the same welding features as the film to be identified are determined from the historical film image database; these are identified as identical films. This invention utilizes a computer and a film scanner to generate weld radiograph images related to the original weld radiograph, achieving digital management of weld radiographs and enabling the identification and detection of weld defects.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the description, claims, and drawings. Attached Figure Description

[0017] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with the embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0018] Figure 1 This is a step diagram of a method for identifying identical welding radiographs according to an embodiment of this application.

[0019] Figure 2 This is a schematic diagram of the neural network framework for a method of identifying identical films on welding radiographs according to an embodiment of this application.

[0020] Figure 3 This is a block diagram of a system for identifying identical welding radiographs according to an embodiment of this application. Detailed Implementation

[0021] The embodiments of the present invention will be described in detail below with reference to the accompanying drawings and examples, so that the process of how the present invention uses technical means to solve technical problems and achieve technical effects can be fully understood and implemented accordingly. It should be noted that, as long as there is no conflict, the various embodiments and features in the various embodiments of the present invention can be combined with each other, and the resulting technical solutions are all within the protection scope of the present invention.

[0022] Furthermore, the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0023] Currently, radiographic testing is one of the main methods for non-destructive testing of welding. Radiographic testing uses X-rays or gamma rays to illuminate the workpiece, and the internal quality of the workpiece is determined by the light captured on film. The density and distribution of the image on the film reflect the type and magnitude of defects within the workpiece. Radiographic imaging features high resolution and high sensitivity, resulting in a large amount of information in the obtained film images. Furthermore, radiographic testing is flexible and can inspect a wide variety of workpiece structures and shapes.

[0024] In the process of developing this invention, the inventors discovered that operational errors or deliberate falsification can lead to a large number of identical films with different weld numbers appearing during radiographic testing. During project acceptance, if only one weld image is falsified, it can be determined whether the image is falsified by checking whether the overlap between the current weld image and adjacent weld images matches. However, if falsification is achieved by re-photographing the entire weld joint, the resulting weld images related to the joint will overlap perfectly. In this case, defects at the weld joint cannot be accurately detected, and undetected defects may lead to significant economic losses and safety accidents.

[0025] Therefore, to address the problem of engineers using newly marked, high-quality historical welds to impersonate poor-quality welds in radiographic testing, this invention provides a method for identifying identical welding radiographic films. This method trains a pre-established historical film image library to train a preset model, obtaining a recognition model for identifying defect classification features. Simultaneously, dynamic information representing the neural network's changing state is collected during training. Then, the collected dynamic information is used to train another preset model, obtaining a recognition model for identifying defect index features, where defect index features indicate defect characteristics on the film image. Finally, the corresponding recognition model is used to identify the defect classification features and defect index features of the film to be identified. Based on the welding features (defect classification features and defect index features) of the current film to be identified, images with the same welding features as the film to be identified are determined from the historical film image library; these are identified as identical films. This invention achieves digital management of welding radiographic films, enabling the identification of weld defects in welding radiographic film images and the determination of whether welding radiographic films are falsified.

[0026] Example 1

[0027] According to the X-ray flaw detection process specifications, each welding radiograph produced by radiographic inspection has image quality markers, positioning markers (center marker, lap marker), and identification markers. The identification markers include: pipeline number, weld number, welder number, and welding date. Additionally, if a welding radiograph requires re-examination, a re-examination marker will be added, along with footnotes indicating the number of re-examinations: R1, R2…R… n To ensure that the obtained radiographic film covers the entire weld joint, multiple X-ray films are typically taken for the same weld joint. During the imaging process, the imaging positions of the weld joints need to be slightly overlapped, and the lead overlap marks on the film must be located at the weld overlap. When imaging the weld joint, these marks are pre-set in lead type at designated positions so that they are imaged along with the weld joint. This results in a radiographic film with all the necessary marks.

[0028] Figure 1 This is a step diagram illustrating a method for identifying identical welding radiographs according to an embodiment of this application. Refer below for... Figure 1 The method for identifying identical welding radiographs described in the embodiments of the present invention will be described in detail.

[0029] like Figure 1 As shown, in step S110, several historical welding radiographs with various markings are acquired. Using a computer and / or a radiograph scanner, each historical radiograph is used as the original to generate a corresponding historical radiograph image. A historical radiograph image library containing all historical radiograph images is then established. Before training the preset model, this embodiment divides the historical radiograph images into a training set, a test set, and a validation set according to a specified ratio. The historical radiograph images in the training set are used to train the preset model; the historical radiograph images in the validation set are used to verify the network performance of the current preset model (e.g., accuracy and recall), and training stops when the network performance converges (the network performance no longer improves); the historical radiograph images in the test set are used to evaluate the network performance of the finally obtained model. Furthermore, this embodiment can also extract corresponding historical radiograph images from the historical radiograph image library by retrieving the storage path, size, storage date, feature encoding, and tags of the historical radiograph images.

[0030] In step S120, a first preset model is trained using historical film images to obtain a first model for identifying defect classification features. Specifically, historical film images are input into the preset first model to obtain defect classification features, and the first model is obtained based on the training of the first preset model. The first model is capable of identifying defect classification features of historical film images and generating corresponding defect classification feature information.

[0031] Next, the generation process of the first model in the embodiments of this application will be described in detail.

[0032] First, the areas covered by defects on historical film images are marked, and corresponding defect types are labeled for these areas. Then, based on a first preset model, the defect areas are used as input information to the first preset model, and the defect type information is used as output information. A first model is obtained by training the first preset model. In this embodiment, the areas covered by welding defects on each historical film image in the historical film image library are determined, and defect type labeling information, used to characterize defect classification features, is labeled onto the corresponding welding defect areas on each historical film image. Then, based on the preset first preset model, the welding defect areas labeled with defect type information are used as input information to the first preset model, and the defect classification feature information is used as output information. Training the first preset model begins, and the first model is obtained after training.

[0033] Next, dynamic information representing the changing state of the neural network during the training of the first model is collected. Based on this, the second preset model is trained to obtain a second model for identifying defect index features, where the defect index features are used to indicate defect features on the film images. Specifically, the first preset model is based on a neural network structure. When training the first preset model using historical film images, the activity state of the neural network constituting the first preset model will change with the different defect features corresponding to the historical film images. Therefore, this embodiment collects dynamic information representing the changing state of the neural network throughout the training process of the first model, so as to label the defect areas of each historical film image with the corresponding information representing the changing state of the neural network. Then, the defect areas labeled with the information representing the changing state of the neural network are used as the input information of the second preset model, and the information representing the changing state of the neural network (i.e., the defect index features used to indicate defect features on the historical film images) is used as the output information of the second preset model to start training the second preset model. After training is completed, the second model is obtained.

[0034] In the embodiments of this application, the first preset model and the second preset model are constructed using the Transformer neural network framework. Figure 2 This is a schematic diagram of the neural network framework for a method of identifying identical pieces on welding radiographs according to an embodiment of this application. (Refer to...) Figure 2 The construction process of the model will be explained in detail using the first preset model as an example.

[0035] Specifically, the first preset model is constructed using a visual Transformer neural network framework. First, each historical film image is divided into several small blocks of a specified size. Each block is mapped to a one-dimensional vector using a linear mapping. Then, each mapped block is input into an embedding layer, where the format of the historical film image is converted into the input format of the encoding structure (i.e., a vector sequence). Next, the vector sequence is transmitted to the encoding structure to classify defects with the same defect features. Finally, a feedforward network is used to identify defect classification features and output the corresponding defect classification feature information. The encoding structure in this embodiment mainly includes a self-attention layer and a neural network. Since the self-attention mechanism corresponding to the self-attention layer does not include positional relationships, positional encoding of each block is required before inputting the vector sequence related to the historical film image into the encoding structure.

[0036] Furthermore, the defect index features in this embodiment include: the encoding information generated by the encoding structure of the neural network; and the activation information generated by the neurons in the neural network. The defect index features are information on the neural network change state of the first preset model during the process of obtaining the first model, specifically including: the encoding information of the defect features generated by the encoding structure after the vector sequence related to the historical film image enters the encoding structure, and the activation degree features and activation features in the activation information of each layer of neurons in the neural network when identifying the defect features.

[0037] Next, in this embodiment, the first defect index feature value and the second defect index feature value are marked on the historical film image.

[0038] In the step of labeling the first defect index feature value, cluster analysis is performed on the encoded information corresponding to images with the same type of defect in the image library to obtain clustering degree information used to characterize the similarity of defect features. The encoded information of the cluster centers in the clustering degree information is sorted from high to low according to the clustering degree, and the first defect index feature value is determined based on the sorting result. Specifically, the historical film image library includes multiple images with the same type of defect. In other words, the defects in each image with the same type of defect belong to the same defect type, but have different defect features. During the training of the first preset model, the encoded information of the defect features generated by the encoding structure is collected. Based on the defect classification feature information identified by the feedforward network, the encoded information of all defect features of multiple historical film images under the current defect classification feature is extracted and cluster analysis is performed to obtain the clustering degree representing the similarity of defect features corresponding to each defect feature under the current defect classification feature, and the encoded information of the defect features at the cluster center is determined. Next, the encoded information of the defect features at the cluster center is sorted in descending order of clustering degree. The encoded information with the highest clustering degree is selected as the first defect index feature value, and the first defect index feature value is marked for each historical film image under the current defect classification feature.

[0039] In one specific embodiment of this application, the first defect index feature value adopts the top 5 to 10 encoded information in terms of clustering degree.

[0040] In the step of labeling the second defect index feature value, the activation information corresponding to images with the same type of defect in the image library is analyzed to obtain activation degree data to characterize the activation state of neurons and activation data to represent the activation features of neurons. The activation degree data of neurons in each layer corresponding to images with the same type of defect are sorted from high to low, and the corresponding activation data is determined as the second defect index feature value based on the sorting result. Specifically, during the training of the first preset model, the activation state of neurons in each layer of the neural network in the current defect classification feature is analyzed, and activation degree data is calculated based on the activation degree of neurons in the activation state. Then, in each layer of the neural network, the activation degree data is sorted in descending order of activation degree, and the activation degree data with the highest activation degree in each layer is selected. Finally, based on the selected activation degree data, the activation data of the neuron to which the activation degree data belongs is obtained. The activation data is used as the activation feature of the neuron, and the activation data is labeled as the second defect index feature value on each historical film image under the current defect classification feature.

[0041] In one specific embodiment of this application, the second defect index feature value is derived from the activation data of the top 50 to 100 neurons ranked by activation level.

[0042] Using the aforementioned labeling method for the first and second defect features, the first and second defect features are labeled on the historical film images. The second preset model is then trained using the historical film images labeled with the first and second defect features to obtain the second model.

[0043] Furthermore, this embodiment utilizes the activation function corresponding to the neuron to obtain the corresponding activation data. Specifically, the activation data reflects the activation characteristics of the neuron; therefore, this embodiment uses the output value of the neuron's activation function as the corresponding activation data.

[0044] After training and obtaining each model, in step S130, the first model and the second model are used to predict the defect classification features and defect index features of the film to be identified, respectively, so as to obtain the defect classification features and defect index features of the current film to be identified.

[0045] Further, in step S140, images with the same type of defects as the film to be identified are selected from the image library and recorded as the first image. The first image is then compared and analyzed with the predicted defect index features to diagnose whether there are images in the image library with the same welding features as the film to be identified. First, based on the defect classification features of the current film to be identified, historical film images with the same defect classification features as the current film to be identified are selected from the historical film image library and recorded as the first image. Then, the similarity between the defect index features marked on the selected historical film images and the defect index features of the current film to be identified is analyzed. The selected historical film images are sorted in descending order of similarity. The historical film images with the highest similarity are selected. Finally, by judging whether the historical film images with the highest similarity have the same welding features as the film to be identified, the corresponding identical films are obtained.

[0046] When comparing the first image with the predicted defect index features, the process begins by acquiring the defect index features of the current film to be identified, resulting in the first and second defect index feature values. Then, the similarity between the first defect index feature value of each historical film image corresponding to the first image and the first defect index feature value of the film to be identified is calculated and recorded as the first similarity. The similarity between the second defect index feature value of each historical film image corresponding to the first image and the second defect index feature value of the film to be identified is also calculated and recorded as the second similarity. Next, the first and second similarities are combined using summation and other methods to obtain the overall similarity calculation result. Finally, each historical film image corresponding to the first image is sorted according to the similarity calculation results from highest to lowest, thus identifying the historical film images ranked higher as identical films to the film to be identified.

[0047] In one specific embodiment of this application, the five film images with the highest similarity ranking from the historical film image library are used for the same film.

[0048] Furthermore, this embodiment uses the cosine similarity calculation method to calculate the first similarity and the second similarity.

[0049] Example 2

[0050] Based on the method for identifying identical welding radiographs described in Embodiment 1 above, this invention also provides a system for identifying identical welding radiographs (hereinafter referred to as the "identical radiograph identification system"). Figure 3 This is a block diagram of a system for identifying identical welding radiographs according to an embodiment of this application.

[0051] like Figure 3 As shown, the identical piece recognition system in this embodiment of the invention includes: an image library establishment module 31, a model generation module 32, a defect feature recognition module 33, and an identical piece recognition module 34. Specifically, the image library establishment module 31 is implemented according to the method described in step S110 above, and is configured to establish a historical film image library; the model generation module 32 is implemented according to the method described in step S120 above, and is configured to train a first preset model using historical film images to obtain a first model for identifying defect classification features, and collect dynamic information used to characterize the change state of the neural network during the training process of the first model. Based on this, a second preset model is trained to obtain a second model for identifying defect index features, wherein the defect index features are used to indicate defect features on the film image; the defect feature recognition module 33 is implemented according to the method described in step S130 above, and is configured to use the first model and the second model generated by the model generation module 32 to predict the defect classification features and defect index features of the film to be identified, respectively; the same film recognition module 34 is implemented according to the method described in step S140 above, and is configured to select images with the same type of defects as the film to be identified from the image library and record them as the first image, and compare and analyze the first image with the predicted defect index features to diagnose whether there are images with the same welding features as the film to be identified in the image library.

[0052] This invention discloses a method for identifying identical weld radiographs. The method trains a pre-established historical film image database to train a preset model, obtaining a recognition model for identifying defect classification features. Simultaneously, dynamic information representing the changing states of the neural network is collected during training. Then, the collected dynamic information is used to train another preset model, obtaining a recognition model for identifying defect index features indicating defects on the film image. Finally, the corresponding recognition model is used to identify the defect classification features and defect index features of the film to be identified. Based on the welding features (defect classification features and defect index features) of the current film to be identified, images with the same welding features as the film to be identified are determined from the historical film image database; these are identified as identical films. This invention achieves digital management of weld radiographs and enables the identification and detection of weld defects.

[0053] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0054] Of course, the present invention may have other various embodiments. Without departing from the spirit and essence of the present invention, those skilled in the art can make various corresponding changes and modifications according to the present invention, but these corresponding changes and modifications should all fall within the protection scope of the claims of the present invention.

[0055] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby storing them in a storage device for execution by a computing device, or fabricating them separately as individual integrated circuit modules, or fabricating multiple modules or steps as a single integrated circuit module. Thus, the present invention is not limited to any particular hardware and software combination.

[0056] While the embodiments disclosed in this invention are as described above, the content is merely for the purpose of facilitating understanding of the invention and is not intended to limit the invention. Any person skilled in the art to which this invention pertains may make any modifications and variations in form and detail of the implementation without departing from the spirit and scope disclosed herein; however, the scope of patent protection for this invention shall still be determined by the scope defined in the appended claims.

Claims

1. A method for identifying identical welding radiographs, comprising: Establish a historical film image library; A first preset model is trained using historical film images to obtain a first model for identifying defect classification features. Dynamic information representing the changing states of the neural network during the training process of the first model is collected. Based on this, a second preset model is trained to obtain a second model for identifying defect index features. The defect index features indicate defect features on the film images and include: encoded information generated by the encoding structure of the neural network and activation information generated by neurons in the neural network. In the step of establishing a historical film image library, the historical film images are also labeled with first and second defect index feature values. Cluster analysis is performed on the encoding information corresponding to images with the same type of defects in the image library to obtain clustering degree information to characterize the similarity of defect features. The encoding information of the cluster centers in the clustering degree information is sorted from high to low according to the clustering degree. Based on the sorting result, the first defect index feature value is determined. Activation information corresponding to images with the same type of defects in the image library is analyzed to obtain activation degree data to characterize the activation state of neurons and activation data to represent the activation features of neurons. The activation degree data of neurons in each layer corresponding to images with the same type of defects are sorted from high to low. Based on the sorting result, the corresponding activation data is determined as the second defect index feature value. Using the first model and the second model, the defect classification features and defect index features of the film to be identified are predicted respectively; Images with the same type of defects as the film to be identified are selected from the image library and recorded as the first image. The first image is then compared and analyzed with the predicted defect index features to diagnose whether there are images in the image library with the same welding features as the film to be identified.

2. The method according to claim 1, characterized in that, The first preset model and the second preset model are constructed using the Transformer neural network framework.

3. The method according to claim 1 or 2, characterized in that, The process of generating the first model includes: The areas covered by defects on the historical film images are marked, and the corresponding defect types are labeled for the defect areas. Then, based on the first preset model, the defect areas are used as the input information of the first preset model, and the defect type information is used as the output information of the first preset model, thereby obtaining the first model through training the first preset model.

4. The method according to claim 3, characterized in that, The method utilizes the activation function corresponding to the neuron to obtain the corresponding activation data.

5. The method according to claim 4, characterized in that, The step of comparing and analyzing the first image with the predicted defect index features includes: Based on the defect index features of the film to be identified, the first defect index feature value and the second defect index feature value of the film to be identified are obtained; The similarity between the first defect index feature value of the first image and the first defect index feature value of the film to be identified is calculated respectively; and the similarity between the second defect index feature value of the first image and the second defect index feature value of the film to be identified is calculated respectively. The first images are sorted according to the similarity calculation results, and then, based on the sorting results, the identical images of the negatives to be identified are determined in the historical negative image library.

6. The method according to claim 5, characterized in that, The first defect index feature value is the top 5 to 10 encoded information items ranked by clustering degree; The second defect index feature value is the activation data of the top 50 to 100 neurons ranked by activation level; The identical images refer to the top 5 film images in the historical film image library based on their similarity ranking.

7. The method according to claim 5 or 6, characterized in that, The similarity is obtained by calculating the cosine similarity.

8. A system for identifying identical films on welding radiographic films, the system being used to implement the method as described in any one of claims 1 to 7, the system comprising the following modules: The image library creation module is used to create a historical film image library; The model generation module trains a first preset model using historical film images to obtain a first model for identifying defect classification features. It also collects dynamic information used to characterize the changing states of the neural network during the training of the first model. Based on this information, it trains a second preset model to obtain a second model for identifying defect index features. The defect index feature is used to indicate defect features on the film image. The defect index feature includes: encoded information generated by the encoding structure of the neural network, and activation information generated by neurons in the neural network. In the step of establishing a historical film image library, the historical film images are also labeled with a first defect index feature value and a second defect index feature value, wherein... Cluster analysis is performed on the encoding information corresponding to images with the same type of defects in the image library to obtain clustering degree information to characterize the similarity of defect features. The encoding information of the cluster centers in the clustering degree information is sorted from high to low according to the clustering degree. Based on the sorting result, the first defect index feature value is determined. Activation information corresponding to images with the same type of defects in the image library is analyzed to obtain activation degree data to characterize the activation state of neurons and activation data to represent the activation features of neurons. The activation degree data of neurons in each layer corresponding to images with the same type of defects are sorted from high to low. Based on the sorting result, the corresponding activation data is determined as the second defect index feature value. The defect feature recognition module is used to predict the defect classification features and defect index features of the film to be identified using the first model and the second model, respectively. The same film identification module is used to filter out images with the same type of defects as the film to be identified from the image library and record them as the first image, and compare and analyze the first image with the predicted defect index features to diagnose whether there are images with the same welding features as the film to be identified in the image library.

Images