Fuel rod end plug welding seam segmentation method and device based on deep convolutional neural network
By using multi-layer feature pyramids and attention fusion technology from deep convolutional neural networks, the problem of low efficiency in traditional manual inspection has been solved, and high-precision automated inspection of fuel rod end plug welds has been achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CHINA NORTH NUCLEAR FUEL CO LTD
- Filing Date
- 2025-11-25
- Publication Date
- 2026-04-24
AI Technical Summary
Traditional manual visual inspection of fuel rod end plug weld quality is inefficient and highly susceptible to subjective factors, making it difficult to guarantee the accuracy and consistency of inspection results.
A deep convolutional neural network-based approach is adopted to extract multi-layer feature information from weld seam images through multi-layer feature pyramids and attention fusion techniques. The spatial resolution of the image is restored by upsampling and skip connections to generate an accurate pixel-level segmentation mask.
This improved the accuracy and efficiency of weld quality assessment, reduced missed and false detections, and ensured high accuracy and consistency of test results.
Smart Images
Figure CN121921616A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of nuclear power technology, specifically relating to a method and apparatus for segmenting fuel rod end plug welds based on deep convolutional neural networks. Background Technology
[0002] In the field of nuclear engineering, fuel rods are the core components of nuclear reactors, and the quality of their end plug welds directly affects the fuel rod's sealing performance, structural integrity, and the safe operation of the reactor. Traditional methods for inspecting fuel rod end plug weld quality primarily rely on manual visual inspection, which largely depends on the experience of the inspectors. Due to potential issues such as noise interference, uneven grayscale, and complex textures in weld images, manual inspection is inefficient and highly susceptible to subjective factors, making it difficult to guarantee the accuracy and consistency of the results. Especially after prolonged work, fatigue can lead to misjudgments. Therefore, there is an urgent need to improve the accuracy and efficiency of weld quality assessment. Summary of the Invention
[0003] To overcome the problems existing in related technologies, a method and apparatus for segmenting fuel rod end plug welds based on deep convolutional neural networks are provided.
[0004] According to one aspect of the embodiments of this disclosure, a method for segmenting fuel rod end plug welds based on a deep convolutional neural network is provided, the method comprising:
[0005] Step 1: For the target image, extract feature maps with different downsampling ratios;
[0006] Step 2: Obtain a multi-layer feature pyramid based on multiple feature maps; Step 2 includes:
[0007] Step 21: Perform max pooling multiple times on the feature maps of some or all layers to generate pooled feature maps of multiple scales.
[0008] Step 22: Create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size.
[0009] Step 3: Perform attention fusion on adjacent feature layers in the feature pyramid to generate an enhanced feature map;
[0010] Step 4: The high-level semantic information generated by the encoder is fused with the low-level detail features. Through upsampling and skip connection operations, the spatial resolution of the image is gradually restored to generate an accurate pixel-level segmentation mask, ensuring high precision of the segmentation results.
[0011] Step 5: Evaluate the segmentation results and calculate the segmentation accuracy index.
[0012] In one possible implementation, in step 1, a feature extraction network is used to extract feature maps at multiple levels, which correspond to downsampling scales of 2, 4, 8, 16, and 32 times that of the target image, respectively.
[0013] In one possible implementation, in step 21, the kernel size of max pooling is 2, the stride is 2, and the number of pooling operations is 2, generating pooled feature maps with the same size as the original feature map, and 0.5 times and 0.25 times the original feature map size.
[0014] In one possible implementation, in step 22, each layer of the feature pyramid is composed of stacked pooled feature maps of different levels, specifically:
[0015] The first layer consists of the original scale feature maps of the third layer;
[0016] The second layer is formed by superimposing the 0.5x feature map of the third layer with the original scale feature map of the fourth layer;
[0017] The third layer consists of a superposition of the 0.25x feature map from the third layer, a 0.5x feature map from the fourth layer, and the original scale feature map from the fifth layer.
[0018] The fourth layer is formed by superimposing a 0.25x feature map from the fourth layer with a 0.5x feature map from the fifth layer.
[0019] The fifth layer consists of a feature map that is 0.25 times larger than the fifth layer.
[0020] In one possible implementation, in step 3, the features of two adjacent feature layers in the feature pyramid are concatenated and convolved to obtain a weight map; then, feature transformation is performed on any one of the two feature layers to form an attention map, which is used to identify important positions; the weight map and the attention map are fused to generate an enhanced feature map.
[0021] According to another aspect of the embodiments of this disclosure, a fuel rod end plug weld segmentation device based on a deep convolutional neural network is provided, the device comprising:
[0022] The feature extraction module is used to extract feature maps with different downsampling ratios from the target image;
[0023] The multi-level pyramid module is used to generate multi-level feature pyramids based on multiple feature maps, including:
[0024] The pooling module is used to perform multiple max pooling operations on feature maps of some or all layers to generate pooled feature maps of multiple scales.
[0025] The stacking module is used to create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size.
[0026] The attention fusion module is used to perform attention fusion on adjacent feature layers in the feature pyramid to generate an enhanced feature map.
[0027] The decoder module is used to fuse high-level semantic information generated by the encoder with low-level detail features. Through upsampling and skip connection operations, it gradually restores the spatial resolution of the image and generates an accurate pixel-level segmentation mask to ensure high precision of the segmentation results.
[0028] The evaluation module is used to evaluate the segmentation results and calculate the segmentation accuracy index.
[0029] According to another aspect of the embodiments of this disclosure, a fuel rod end plug weld segmentation device based on a deep convolutional neural network is provided, the device comprising:
[0030] processor;
[0031] Memory used to store processor-executable instructions;
[0032] The processor is configured to execute the above-described method.
[0033] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a processor, implement the above-described method.
[0034] The beneficial effects of this disclosure are as follows: This disclosure provides a method for segmenting fuel rod end plug welds based on deep convolutional neural networks, which extracts and modifies the feature path of the target image from top to bottom, the encoder part extracts deep features through multi-layer convolution, and the decoder part uses skip connections to gradually restore the spatial resolution of the image.
[0035] During the construction of the feature pyramid, multi-level feature fusion and progressive refinement are used to achieve accurate segmentation of fine weld seam structures, reducing missed and false detections. Designed to enable centralized information interaction and fusion of multi-level features, a distribution mechanism allocates feature maps of different levels to the feature pyramid, allowing feature information from different levels, scales, and sub-regions to interact within the pyramid. This strategy ensures the accuracy of high-level semantic localization information, prevents irregular feature dilution, and does not consume additional parameter space.
[0036] This method effectively integrates features from adjacent scales within a pyramid, incorporating high-level semantic localization information into the feature maps of those scales. By learning the similarity between adjacent scales to balance positional information and contextual features, it provides attention and localization information from high-level semantic features to the model's decoder or prediction head. This ensures that the output feature map contains positional and attentional information from adjacent features, thus transmitting high-level semantic positional information to the output. This can effectively improve the accuracy and efficiency of weld quality assessment. Attached Figure Description
[0037] Figure 1 This is a flowchart illustrating a method for segmenting the weld seam of a fuel rod end plug based on a deep convolutional neural network, as shown in an embodiment of this disclosure.
[0038] Figure 2 This is a schematic diagram illustrating a feature extraction architecture according to an embodiment of this disclosure.
[0039] Figure 3 This is a schematic diagram of a multi-layered pyramid structure shown in an embodiment of this disclosure.
[0040] Figure 4 This is a schematic diagram illustrating an attention fusion according to an embodiment of this disclosure.
[0041] Figure 5 This is a block diagram of a fuel rod end plug weld seam segmentation device based on a deep convolutional neural network, as shown in an embodiment of this disclosure. Detailed Implementation
[0042] The present disclosure will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0043] Unless otherwise defined, the technical and scientific terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains; the terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure; the term "comprising" and any variations thereof in this disclosure are intended to cover non-exclusive inclusion. Clearly, the embodiments described in this disclosure are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.
[0044] In this disclosure, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this disclosure. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0045] Figure 1 This is a flowchart illustrating a method for segmenting fuel rod end plug welds based on a deep convolutional neural network, as shown in an embodiment of this disclosure. This method can be executed by a terminal device, which can be a server, desktop computer, laptop computer, etc. This disclosure does not limit the type of terminal device. Figure 1 As shown, the method includes:
[0046] Step 1: For the target image, extract feature maps at multiple downsampling ratios. For example, an extractor (such as ResNet50) can be used as the feature extraction path and modified, including removing the average pooling and fully connected layers in the last part. Divide the ResNet50 into five levels, extracting features based on 2, 4, 8, 16, and 32x downsampling of the target image, resulting in feature maps m1, m2, m3, m4, and m5, respectively.
[0047] Step 2: Using a multi-level pyramid model, multiple feature maps are processed to obtain a multi-layer feature pyramid; this enhances the network's ability to accurately capture defect locations, improves the accuracy of defect localization, and reduces false detections and missed detections.
[0048] like Figure 2 and Figure 3 As shown, in order to effectively interact with multi-level information, this disclosure uses a distribution mechanism to distribute feature maps of different levels into a feature pyramid. Step 2 includes:
[0049] Step 21: Take some or all of the feature maps from each layer as input, and generate multiple pooled feature maps of different scales from the feature maps of each layer through max pooling, as shown in the following formula:
[0050] m i1 =m i
[0051] m ij =Maxpool(m i(j-1) )
[0052] MaxPool represents max pooling, m i Let be the input feature map of layer i, j represent the pooling iteration order, N represent the maximum number of pooling iterations, and j∈[0,N]
[0053] Continuing from the previous example, we can select m. i As input, i∈(3,4,5), a MaxPool with a kernel size of 2 and a stride of 2 is used to obtain the pooled feature map m. i1 m i2 m i3 m i1 With m i The same size, mi2 For m i 0.5 times, m i3 For m i This is 0.25 times the size of the previous one, allowing information from different scales to interact fully in subsequent processes. As shown in the following formula:
[0054] m i1 =m i
[0055] m i2 =Maxpool(m i1 )
[0056] m i3 =Maxpool(m i2 )
[0057] Step 22: Create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size.
[0058] Continuing from the previous example, let's create a feature pyramid with 5 levels, denoted as q1, q2, q3, q4, and q5, as shown in the following formula:
[0059] q1=m 31
[0060] q2=m 32 +m 41
[0061] q3=m 33 +m 42 +m 51
[0062] q4=m 43 +m 52
[0063] q5=m 53
[0064] This fusion method enables the interaction of multi-level feature information in the feature pyramid, which can ensure the accuracy of high-level semantic localization information, prevent features from being irregularly diluted, and does not consume additional parameter space.
[0065] Step 3, Attention Fusion: The features of two adjacent feature layers in the feature pyramid are concatenated and convolved to obtain a weight map. Then, feature transformation is performed on any one of the two feature layers to form an attention map, which is used to identify important positions. The weight map and the attention map are fused to generate an enhanced feature map, thereby strengthening useful information and suppressing noise. In this way, the attention mechanism highlights important feature information, suppresses irrelevant noise, promotes the interaction of feature information at deeper levels, improves the model's ability to identify defects in complex backgrounds, and enhances the reliability of the segmentation results.
[0066] For example, such as Figure 4 As shown, q is output by the multi-layer pyramid module. k ∈R C×H×W , First, q k After performing downsampling operation and q k+1 A concat operation is performed, followed by a 1×1 convolutional operation, and then a sigmoid operation to obtain the weight map L. Next, After reshaping, it becomes Q∈R C×(H×W) Q and Q T After multiplication, a softmax operation is performed to obtain attention map A. Attention map A and L... T After performing matrix multiplication, the result is reshaped and upsampled to F∈R C×H×W The final output is q′ i The result of multiplying by the corresponding pixel of F and q′ i+1 The result is obtained by multiplying the product of the corresponding pixel value of F and adding the results together.
[0067] Step 4: The high-level semantic information generated by the encoder is fused with the low-level detail features. Through upsampling and skip connection operations, the spatial resolution of the image is gradually restored to generate an accurate pixel-level segmentation mask, ensuring high precision of the segmentation results.
[0068] Step 5: Evaluate the segmentation results and compare them with the real labels to calculate metrics such as accuracy, recall, and intersection-union ratio to measure the accuracy and reliability of the model segmentation.
[0069] In one possible implementation, a fuel rod end plug weld segmentation device based on a deep convolutional neural network is provided, the device comprising:
[0070] The feature extraction module is used to extract feature maps with different downsampling ratios from the target image;
[0071] The multi-level pyramid module is used to generate multi-level feature pyramids based on multiple feature maps, including:
[0072] The pooling module is used to perform multiple max pooling operations on feature maps of some or all layers to generate pooled feature maps of multiple scales.
[0073] The stacking module is used to create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size.
[0074] The attention fusion module is used to perform attention fusion on adjacent feature layers in the feature pyramid to generate an enhanced feature map.
[0075] The decoder module is used to fuse high-level semantic information generated by the encoder with low-level detail features. Through upsampling and skip connection operations, it gradually restores the spatial resolution of the image and generates an accurate pixel-level segmentation mask to ensure high precision of the segmentation results.
[0076] The evaluation module is used to evaluate the segmentation results and calculate the segmentation accuracy index.
[0077] The description of the above-mentioned apparatus has been elaborated in detail in the description of the above-mentioned method, and will not be repeated here.
[0078] Figure 5 This is a block diagram illustrating a fuel rod end plug weld segmentation device based on a deep convolutional neural network, as shown in an embodiment of this disclosure. For example, device 1900 can be provided as a server. (Refer to...) Figure 5 The apparatus 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.
[0079] Device 1900 may also include a power supply component 1926 configured to perform power management of device 1900, a wired or wireless network interface 1950 configured to connect device 1900 to a network, and an input / output (I / O) interface 1958. Device 1900 can operate on an operating system stored in memory 1932, such as Windows Server™, MacOS X™, Unix™, Linux™, FreeBSD™, or similar.
[0080] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of the device 1900 to perform the above-described method.
[0081] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.
[0082] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0083] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.
[0084] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.
[0085] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0086] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.
[0087] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0088] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0089] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.
Claims
1. A method for segmenting fuel rod end plug welds based on deep convolutional neural networks, characterized in that, The method includes: Step 1: For the target image, extract feature maps with different downsampling ratios; Step 2: Obtain a multi-layer feature pyramid based on multiple feature maps; Step 2 includes: Step 21: Perform max pooling multiple times on the feature maps of some or all layers to generate pooled feature maps of multiple scales. Step 22: Create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size. Step 3: Perform attention fusion on adjacent feature layers in the feature pyramid to generate an enhanced feature map; Step 4: The high-level semantic information generated by the encoder is fused with the low-level detail features. Through upsampling and skip connection operations, the spatial resolution of the image is gradually restored to generate a pixel-level segmentation mask. Step 5: Evaluate the segmentation results and calculate the segmentation accuracy index.
2. The method according to claim 1, characterized in that, In step 1, a feature extraction network is used to extract feature maps at multiple levels, which correspond to downsampling scales of 2, 4, 8, 16 and 32 times that of the target image, respectively.
3. The method according to claim 1, characterized in that, In step 21, the kernel size of max pooling is 2, the step size is 2, and the number of pooling operations is 2, generating pooled feature maps with the same size as the original feature map, and 0.5 times and 0.25 times the original feature map size.
4. The method according to claim 1, characterized in that, In step 22, each layer of the feature pyramid is composed of stacked pooled feature maps of different levels, specifically: The first layer consists of the original scale feature maps of the third layer; The second layer is formed by superimposing the 0.5x feature map of the third layer with the original scale feature map of the fourth layer; The third layer consists of a superposition of the 0.25x feature map from the third layer, a 0.5x feature map from the fourth layer, and the original scale feature map from the fifth layer. The fourth layer is formed by superimposing a 0.25x feature map from the fourth layer with a 0.5x feature map from the fifth layer. The fifth layer consists of a feature map that is 0.25 times larger than the fifth layer.
5. The method according to claim 1, characterized in that, In step 3, the features of two adjacent feature layers in the feature pyramid are concatenated and convolved to obtain a weight map; then, feature transformation is performed on any one of the two feature layers to form an attention map, which is used to identify important positions; the weight map and the attention map are fused to generate an enhanced feature map.
6. A fuel rod end plug weld segmentation device based on a deep convolutional neural network, characterized in that, The device includes: The feature extraction module is used to extract feature maps with different downsampling ratios from the target image; The multi-level pyramid module is used to generate multi-level feature pyramids based on multiple feature maps, including: The pooling module is used to perform multiple max pooling operations on feature maps of some or all layers to generate pooled feature maps of multiple scales. The stacking module is used to create a feature pyramid based on multiple pooled feature maps. Each feature layer of the pyramid is formed by stacking pooled feature maps of the same size. The attention fusion module is used to perform attention fusion on adjacent feature layers in the feature pyramid to generate an enhanced feature map. The decoder module is used to fuse high-level semantic information generated by the encoder with low-level detail features. Through upsampling and skip connection operations, it gradually restores the spatial resolution of the image and generates an accurate pixel-level segmentation mask to ensure high precision of the segmentation results. The evaluation module is used to evaluate the segmentation results and calculate the segmentation accuracy index.
7. A fuel rod end plug weld segmentation device based on a deep convolutional neural network, characterized in that, The device includes: processor; Memory used to store processor-executable instructions; The processor is configured to perform the method according to any one of claims 1 to 5.
8. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method described in any one of claims 1 to 5.