Method, device and system for detecting surface characteristics of fuel assembly

Abstract

The invention relates to a fuel assembly surface feature detection method, device and system, and the method comprises the steps: obtaining surface image data of a to-be-detected fuel assembly at different exposure times, the surface image data comprising a low exposure image, a normal exposure image and an overexposure image; performing feature fusion on the low-exposure image, the normal-exposure image and the overexposure image by using a global context attention mechanism to obtain a target fusion image; performing image segmentation on the target fusion image by using a U-Net algorithm to obtain line laser features; performing three-dimensional reconstruction on the to-be-measured fuel assembly according to the linear laser features; determining a surface feature detection result of the to-be-detected fuel assembly according to a three-dimensional reconstruction result; the method is used for reducing the influence of factors such as underwater interference, insufficient light and fuel rod surface reflection on linear laser feature extraction, so that the surface feature detection efficiency of the fuel assembly can be improved.

Description

Technical Field

[0001] The present invention relates to the field of image processing technology, and in particular to a method, device and system for detecting surface features of a fuel assembly. Background Art

[0002] The fuel assembly consists of a group of fuel rods and other components. It is the core component of the nuclear power plant where nuclear fission reactions occur to generate energy. It works in high temperature, high pressure and high radiation circulating water for a long time. Under the influence of high temperature, high pressure and assembly stress, the fuel assembly may deform as the working time increases, which may seriously affect the normal operation of the nuclear power plant. Therefore, it is necessary to regularly detect the surface characteristics of the fuel assembly.

[0003] At present, when detecting the surface features of fuel assemblies, a line laser generator is usually used to illuminate the surface of the fuel assembly first, and then a camera is used to capture the surface image of the fuel assembly, and the line laser features are extracted from the surface image through an image processing terminal. The fuel assembly is reconstructed in three dimensions through the extracted line laser features, and finally the surface features of the fuel assembly are remotely, safely and accurately determined. Therefore, accurately extracting the line laser features in the image is particularly critical for determining the surface features of the fuel assembly. To accurately extract the line laser features in the image, two key points are usually required: one is that the line laser features in the image must be obvious enough; the other is that the line laser features extracted from the image must be accurate enough.

[0004] However, since the fuel assembly is in an underwater working environment for a long time, it will be affected by factors such as underwater interference, insufficient light, and reflection on the surface of the fuel rod. Although the surface image obtained by using the low exposure mode can reduce the impact of underwater interference and reflection, making the overall color darker, the line laser feature will not be obvious due to the dim image, making it difficult to extract the line laser feature. If the exposure time is increased, although the overall brightness of the image will be improved and the difficulty of extracting the line laser feature will be reduced, it will also lead to an increase in interference noise, causing the line laser to overlap with the high reflection in the area where the fuel rod is highly reflective, making it difficult to extract the line laser feature. Therefore, the surface image collected by the camera will be of poor quality due to factors such as underwater interference, insufficient light, and reflection on the surface of the fuel rod, which will increase the difficulty of extracting the line laser feature, and then reduce the efficiency of surface feature detection of the fuel assembly. Summary of the invention

[0005] The technical problem to be solved by the present invention is to provide a method, device and system for detecting the surface features of a fuel assembly, so as to reduce the influence of factors such as underwater interference, insufficient light and reflection on the surface of the fuel rod on the line laser feature extraction, thereby improving the efficiency of the surface feature detection of the fuel assembly.

[0006] The technical solution adopted by the present invention to solve the technical problem is: to provide a method for detecting the surface characteristics of a fuel assembly, comprising:

[0007] S101, acquiring surface image data of a fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image;

[0008] S102, using a global context attention mechanism to perform feature fusion on the underexposure image, the normal exposure image, and the overexposure image to obtain a target fused image;

[0009] S103, performing image segmentation on the target fusion image using a U-Net algorithm to obtain line laser features;

[0010] S104, performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser feature;

[0011] S105. Determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

[0012] Preferably, the step S102 includes:

[0013] Inputting the underexposed image, the normally exposed image, and the overexposed image into a global context attention module for preliminary feature extraction to obtain initial image features;

[0014] Using a global context attention mechanism to fuse the preliminary image features to obtain fused image features;

[0015] Using the initial image features as shortcut branches, establishing interlayer connections between the under-exposure image, the normal-exposure image, the over-exposure image and the fused image features to construct a residual neural network;

[0016] The output of the residual neural network is determined to be a target fused image.

[0017] Preferably, the using of a global context attention mechanism to perform feature fusion on the preliminary image features to obtain fused image features comprises:

[0018] Dividing the preliminary image features into a first portion of features, a second portion of features, and a third portion of features;

[0019] Inputting the second part of features into the first convolutional layer, and performing nonlinear activation on the output of the first convolutional layer using a Softmax function to obtain a first image feature;

[0020] Multiplying the first image feature and the third part feature element by element to obtain a second image feature;

[0021] Using a Leaky ReLU function to nonlinearly activate the second image feature, and inputting the second image feature after nonlinear activation into the second convolutional layer to obtain a third image feature;

[0022] The third image feature is added to the first part of the features element by element to obtain a fused image feature.

[0023] Preferably, step 103 includes:

[0024] Using the encoder of the U-Net algorithm to gradually extract features from the target fused image, and gradually reducing the size of the generated feature map;

[0025] When the size of the feature map is reduced to a first preset threshold, the current feature map is input into the decoder of the U-Net algorithm for deconvolution operation, and the size of the feature map is gradually restored;

[0026] When the size of the feature map is restored to the size of the target fused image, the probability distribution of each pixel in the current feature map is obtained using the Sigmoid function, and the pixel points belonging to the line laser feature are determined according to the second preset threshold.

[0027] Preferably, the encoder using the U-Net algorithm gradually extracts features from the target fused image and gradually reduces the size of the generated feature map, including:

[0028] Inputting the target fused image into the first convolution module in the encoder of the U-Net algorithm for feature extraction to obtain a feature map, wherein the first convolution module includes a third convolution layer, a fourth convolution layer and a Leaky ReLU function;

[0029] Inputting the feature map into a first pooling layer to reduce the size of the feature map;

[0030] When determining whether the size of the feature map is smaller than a first preset threshold;

[0031] If not, the reduced-size feature map is sequentially input into the second convolution module and the second pooling layer in the encoder for secondary size reduction;

[0032] This cycle is repeated until the size of the feature map is reduced to the first preset threshold.

[0033] Preferably, inputting the current feature map into the decoder of the U-Net algorithm to perform a deconvolution operation and gradually restoring the size of the feature map includes:

[0034] Inputting the feature map output by the encoder into the first deconvolution module in the decoder of the U-Net algorithm to increase the size of the feature map;

[0035] Copying and cropping the feature map output by the first convolution module to obtain a first sub-feature map;

[0036] Concatenating the feature map output by the first deconvolution module with the first sub-feature map;

[0037] Determine whether the stitched feature map is smaller than the size of the target fused image;

[0038] If so, the concatenated feature map is input into the second deconvolution module in the decoder, and the feature map output by the second deconvolution module is concatenated with the second sub-feature map corresponding to the second convolution module;

[0039] This cycle is repeated until the spliced ​​feature map is restored to the size of the target fused image.

[0040] The present invention also provides a surface feature detection device for a fuel assembly, comprising:

[0041] An acquisition unit, used to acquire surface image data of the fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image;

[0042] A feature fusion unit, configured to perform feature fusion on the underexposed image, the normally exposed image, and the overexposed image using a global context attention mechanism to obtain a target fused image;

[0043] An image segmentation unit, used for performing image segmentation on the target fusion image by using a U-Net algorithm to obtain line laser features;

[0044] A three-dimensional reconstruction unit, used for performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser feature;

[0045] The determination unit is used to determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

[0046] The present invention also provides an image acquisition system for a fuel assembly, comprising an image acquisition device and the surface feature detection device of the fuel assembly, wherein the image acquisition device is electrically connected to the surface feature detection device of the fuel assembly, and the image acquisition device comprises: a fixing device, a transmission component and an imaging component;

[0047] The transmission assembly is mounted on the fixing device;

[0048] The imaging assembly is movably connected to the transmission assembly, and the transmission assembly is used to drive the imaging assembly to move in a first direction and a second direction respectively, the first direction is a circumferential direction, and the second direction is a linear direction, and the imaging assembly is used to collect a surface image of the fuel assembly to be tested.

[0049] Preferably, the transmission assembly comprises: a first motor, a second motor and a screw rod;

[0050] One end of the screw rod is respectively connected to the first motor and the second motor, and the other end is connected to the imaging assembly. The first motor is used to drive the imaging assembly to move in a first direction through the screw rod, and the second motor is used to drive the imaging assembly to move in a second direction through the screw rod, and the central axis of the first direction is perpendicular to the second direction.

[0051] Preferably, the imaging assembly comprises: a line laser generator, a camera and a pressing plate;

[0052] The pressing sheet comprises a first pressing plate and a second pressing plate, and grooves are respectively provided at both ends of the first pressing plate and the second pressing plate;

[0053] After the first pressing plate and the second pressing plate are fixedly connected, the line laser generator and the camera are respectively clamped in the installation space formed by the grooves at both ends.

[0054] The implementation of the present invention has the following beneficial effects: firstly, the surface image data of the fuel assembly to be tested under different exposure times are obtained, and the surface image data contains low-exposure images, normal-exposure images and over-exposure images; then, the low-exposure images, normal-exposure images and over-exposure images are feature-fused using the global context attention mechanism to obtain a target fused image; then, the target fused image is segmented using the U-Net algorithm to obtain line laser features; then, the fuel assembly to be tested is three-dimensionally reconstructed according to the line laser features; finally, the surface feature detection result of the fuel assembly to be tested is determined according to the three-dimensional reconstruction result. In this way, the surface images with different exposure times can be fused using the global context attention mechanism, and because the global context attention mechanism can dynamically adjust the content weights of images with different exposure times, the line laser features of images with different exposure times can be effectively retained in the target fused image, so that the target fused image can have both the anti-interference and anti-reflection characteristics of the low-exposure image and the high-brightness characteristics of the image after the exposure time is increased. Thereby, the influence of underwater interference, insufficient light and reflection on the surface of the fuel rod on the extraction of line laser features can be reduced, and the surface feature detection efficiency of the fuel assembly can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention will be further described below with reference to the accompanying drawings and embodiments, in which:

[0056] Figure 1 It is a schematic diagram of an embodiment of the surface feature detection method of a fuel assembly of the present invention;

[0057] Figure 2 is a schematic diagram of another embodiment of the surface feature detection method of a fuel assembly of the present invention;

[0058] Figure 3 It is a schematic diagram of an embodiment of a surface feature detection device for a fuel assembly of the present invention;

[0059] Figure 4 It is a structural schematic diagram of the image acquisition device of the present invention;

[0060] Figure 5 It is a structural schematic diagram of the transmission assembly of the present invention;

[0061] Figure 6 is an exploded view of the first motor of the present invention;

[0062] Figure 7 It is a schematic diagram of the structure of the acquisition component of the present invention;

[0063] Figure 8 is an exploded view of the line laser generator of the present invention;

[0064] Fig. 9 is an exploded view of the camera of the present invention. DETAILED DESCRIPTION

[0065] In order to have a clearer understanding of the technical features, purposes and effects of the present invention, the specific embodiments of the present invention are now described in detail with reference to the accompanying drawings. Features defined as "first", "second", "third", etc. may explicitly or implicitly include one or more of the features. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0066] In the following description, specific details such as specific system structures, technologies, etc. are provided for the purpose of illustration rather than limitation, so as to provide a thorough understanding of the embodiments of the present invention. However, it should be clear to those skilled in the art that the present invention may be implemented in other embodiments without these specific details. In other cases, detailed descriptions of well-known systems, devices, circuits, and methods are omitted to prevent unnecessary details from obstructing the description of the present invention.

[0067] The present invention provides a method, device and system for detecting the surface characteristics of a fuel assembly, which are used to reduce the influence of factors such as underwater interference, insufficient light and reflection on the surface of the fuel rod on line laser feature extraction, thereby improving the efficiency of surface characteristic detection of the fuel assembly.

[0068] The surface feature detection method of the fuel assembly provided by the present invention is applied to a system, a server or a terminal with logic analysis and processing capabilities for execution and implementation. The embodiment of the present invention is described by taking the system as an example. Figure 1 As shown, one embodiment of the surface feature detection method of the fuel assembly of the present invention comprises:

[0069] S101, obtaining surface image data of a fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image;

[0070] In this embodiment, the surface image data is a multi-frame image obtained by the camera at the same position of the fuel assembly to be tested according to different exposure times. The surface image data includes a low-exposure image with an exposure time less than a preset time range, a normal exposure time with an exposure time within a preset time range, and an overexposure time with an exposure time greater than a preset time range. Among them, the preset time range is the exposure time range commonly used by the camera when collecting images of the fuel assembly to be tested, for example: the preset time range is 0.05s to 0.09s, the exposure time of the low-exposure image is 0.01s, the exposure time of the normal exposure image is 0.06s, and the exposure time of the overexposure image is 0.12s. It can be understood that the longer the exposure time, the more light the camera senses, and the higher the brightness of the collected image; similarly, the shorter the exposure time, the less light the camera senses, and the lower the brightness of the collected image. Therefore, due to the different exposure times, there will be a large difference in brightness between the low-exposure image, the normal exposure image, and the overexposure image, and the line laser features therein will also show significant differences.

[0071] S102, using a global context attention mechanism to perform feature fusion on the underexposed image, the normally exposed image, and the overexposed image to obtain a target fused image;

[0072] In this embodiment, the global context attention mechanism adopts a global context attention module and a residual learning architecture. Among them, the key to the residual learning architecture is that after the input low-exposure image, normal exposure image, and over-exposure image pass through several global context attention modules each time, the image features before being input to the several global context attention modules and the image features after passing through the several global context attention modules are added element by element, thereby achieving effective feature fusion to obtain a target fused image. This process not only enhances the feature details and dynamic range of the target fused image, but also better focuses on the feature area illuminated by the line laser in the target fused image. In addition, the global context attention mechanism can also improve the robustness under underwater interference and different exposure times, so that the fused target fused image can better retain and fuse the line laser features.

[0073] S103, using the U-Net algorithm to segment the target fusion image to obtain line laser features;

[0074] In this embodiment, since the clarity of the line laser feature is higher in the fused target fusion image, and the background area belongs to a weak texture feature image, a shallow U-Net can be used to segment the target fusion image. U-Net is a convolutional neural network for image segmentation. Its working principle is to gradually extract low-level features and high-level features through an encoder, and then gradually restore the spatial resolution through a decoder, and fuse the low-level features and high-level features to extract image segmentation results with contextual information. The structure of U-Net is a U-shaped structure, which can use global information and local information at the same time. By combining global information and local information, the boundaries of the line laser features can be better learned during the recovery process of the decoder, so as to segment the line laser features and background area features in the target fusion image. Among them, the line laser feature is the feature of the surface area on the fuel assembly to be tested irradiated by the line laser, and the background area is other area features that are not associated with the line laser feature.

[0075] S104, performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser characteristics;

[0076] In this embodiment, after the line laser feature is obtained, the three-dimensional model of the fuel assembly to be tested can be remodeled according to the line laser feature to obtain a three-dimensional model of the fuel assembly to be tested with a higher degree of restoration, which is convenient for subsequent surface feature extraction.

[0077] S105. Determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

[0078] Since the fuel assembly to be tested is usually in an underwater radiation environment, it is difficult to go into the water to directly observe the surface features of the fuel assembly to be tested. Therefore, the surface feature detection results of the fuel assembly to be tested can be obtained based on a three-dimensional model based on line laser features to improve the efficiency of surface feature detection and reduce the safety risks of underwater operations. Among them, the surface feature detection results of the fuel assembly to be tested include the length, curvature and twist of the fuel assembly to be tested, which are not specifically limited here. It is understandable that the surface features of the fuel assembly to be tested can be detected during the transportation process after it is out of the stack to improve the detection efficiency, thereby reducing the risk of damage caused by repeated lifting of the fuel assembly to be tested.

[0079] In this embodiment, the surface image data of the fuel assembly to be tested at different exposure times are first obtained, and the surface image data contains low-exposure images, normal-exposure images, and over-exposure images; then, the global context attention mechanism is used to perform feature fusion on the low-exposure images, normal-exposure images, and over-exposure images to obtain a target fused image; then, the U-Net algorithm is used to perform image segmentation on the target fused image to obtain line laser features; then, the fuel assembly to be tested is three-dimensionally reconstructed according to the line laser features; finally, the surface feature detection result of the fuel assembly to be tested is determined according to the three-dimensional reconstruction result. In this way, the global context attention mechanism can be used to fuse the surface images with different exposure times, and because the global context attention mechanism can dynamically adjust the content weights of images with different exposure times, the line laser features of images with different exposure times can be effectively retained in the target fused image, so that the target fused image can have both the anti-interference and anti-reflection characteristics of the low-exposure image and the high-brightness characteristics of the image after the exposure time is increased. Thereby, the influence of underwater interference, insufficient light, and reflection on the surface of the fuel rod on the extraction of line laser features can be reduced, and the surface feature detection efficiency of the fuel assembly can be improved.

[0080] like Figure 2 As shown, another embodiment of the surface feature detection method of the fuel assembly of the present invention comprises:

[0081] S201, obtaining surface image data of a fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image;

[0082] Step S201 in this embodiment is similar to the above Figure 1 Step S101 in the illustrated embodiment is similar and will not be described in detail here.

[0083] S202, inputting the underexposed image, the normally exposed image, and the overexposed image into a global context attention module for preliminary feature extraction to obtain initial image features;

[0084] S203, using a global context attention mechanism to perform feature fusion on the preliminary image features to obtain fused image features;

[0085] S204, using the initial image features as shortcut branches, establishing interlayer connections between the under-exposure image, the normal-exposure image, the over-exposure image and the fused image features, so as to construct a residual neural network;

[0086] S205, determining the output of the residual neural network as a target fused image;

[0087] Optionally, in this embodiment, the global context attention mechanism includes several global context attention modules, and each global context attention module is connected in a stacked structure. After obtaining the low-exposure image, the normal-exposure image, and the over-exposure image, these images are first input into the global context attention module for preliminary feature extraction. Specifically, this part of the input image will first enter the convolutional layer in the global context attention module for feature extraction, and the extracted features are nonlinearly activated by the Leaky ReLU function to obtain the initial image features.

[0088] Then, the initial image feature is used as a shortcut branch to construct a residual block, which contains multiple global context attention modules, where the input image is connected to the first global context attention module, and the initial image feature of the input image is connected to the output of the last global context attention module as a shortcut branch, so that the inter-layer connection between the input image and the output of the last global context attention module can be established. A neural network composed of multiple residual blocks is a residual neural network. Residual neural networks allow some layers of the neural network to skip the connection of the next layer of neurons and realize inter-layer connections to weaken the strong connection between each layer, thereby reducing the degree of degradation of deep neural networks.

[0089] Among them, the output of the last global context attention module mentioned above refers to the fused image features obtained after the preliminary image features of all input images are fused using the global context attention mechanism.

[0090] In some embodiments, step S203 uses a global contextual attention mechanism to perform feature fusion on preliminary image features to obtain fused image features, which may include: dividing the preliminary image features into a first part of features, a second part of features, and a third part of features; inputting the second part of features into a first convolutional layer, and using a Softmax function to perform nonlinear activation on the output of the first convolutional layer to obtain a first image feature; multiplying the first image feature and the third part of features element by element to obtain a second image feature; using a Leaky ReLU function to perform nonlinear activation on the second image feature, and inputting the nonlinearly activated second image feature into a second convolutional layer to obtain a third image feature; and adding the third image feature and the first part of features element by element to obtain a fused image feature.

[0091] Specifically, the acquired initial image features can be divided into three parts, namely, the first part of features, the second part of features, and the third part of features according to different exposure times. Among them, the first part of features is used as a shortcut branch to construct a residual block. The second part of features is input into the first convolutional layer for secondary feature extraction, and after nonlinear activation of the Softmax function, the first image feature is obtained. The third part of features is multiplied element by element with the first image feature to obtain the second image feature. After the second image feature is nonlinearly activated by the Leaky ReLU function, the second image feature is input into the second convolutional layer for feature extraction again to obtain the third image feature. Finally, the third image feature is added element by element with the first part of features to obtain a fused image feature.

[0092] S206, using the encoder of the U-Net algorithm to gradually extract features of the target fused image, and gradually reducing the size of the generated feature map;

[0093] S207, when the size of the feature map is reduced to a first preset threshold, the current feature map is input into a decoder of the U-Net algorithm for deconvolution operation, and the size of the feature map is gradually restored;

[0094] S208, when the size of the feature map is restored to the size of the target fused image, the probability distribution of each pixel in the current feature map is obtained by using the Sigmoid function, and the pixel points belonging to the line laser feature are determined according to the second preset threshold;

[0095] Optionally, in this embodiment, the U-Net algorithm includes an encoder and a decoder. Among them, the encoder is used to gradually extract the feature map of the target fusion image, and reduce the spatial resolution during the extraction process to reduce the size of the feature map. It can be understood that the feature map is a data structure commonly used in deep learning. It is an important concept in convolutional neural networks. It is the result of a series of convolution kernels performing convolution operations on the input image. It can be regarded as an abstract representation of the input image, and each pixel in the feature map can be represented as some specific features. The decoder is used to perform a deconvolution operation on the feature map output by the encoder when the size is reduced to a first preset threshold value to gradually restore the size of the feature map. When the size of the feature map is restored to the size of the target fusion image, the Sigmoid function can be used to convert the pixel value of the feature map into a probability value to obtain the probability distribution of each pixel point, and the probability value represents the probability that the corresponding pixel point is a line laser feature. Finally, the probability value of each pixel point is compared with the second preset threshold value, and the pixel point that meets the second preset threshold value is determined to be a line laser feature, and the pixel point that does not meet the second preset threshold value is a background area feature.

[0096] In some embodiments, step S206 uses the encoder of the U-Net algorithm to gradually extract features of the target fused image and gradually reduce the size of the generated feature map, which may include: inputting the target fused image into the first convolution module in the encoder of the U-Net algorithm for feature extraction to obtain a feature map, the first convolution module includes a third convolution layer, a fourth convolution layer and a Leaky ReLU function; inputting the feature map into the first pooling layer to reduce the size of the feature map; determining whether the size of the feature map is less than a first preset threshold; if not, inputting the reduced-size feature map into the second convolution module and the second pooling layer in the encoder in turn for secondary size reduction; and repeating this cycle until the size of the feature map is reduced to the first preset threshold.

[0097] Specifically, the encoder is composed of multiple downsampling modules, each of which contains multiple convolution modules and pooling layers, and each convolution module contains multiple convolution layers and Leaky ReLU functions. The convolution module of each layer is used to gradually extract the feature map of the target fusion image, and the pooling layer of each layer is used to perform spatial dimensionality reduction on the feature map extracted by the convolution module to reduce the size of the feature map. Through this layer-by-layer extraction and layer-by-layer dimensionality reduction method, higher-level features can be extracted to form a smaller feature map until the size of the feature map is reduced to a first preset threshold.

[0098] In some embodiments, step S207, when the size of the feature map is reduced to a first preset threshold, the current feature map is input into the decoder of the U-Net algorithm for deconvolution operation, and the size of the feature map is gradually restored, which may include: inputting the feature map output by the encoder into the first deconvolution module in the decoder of the U-Net algorithm to increase the size of the feature map; copying and cropping the feature map output by the first convolution module to obtain a first sub-feature map; splicing the feature map output by the first deconvolution module with the first sub-feature map; determining whether the spliced ​​feature map is smaller than the size of the target fused image; if so, inputting the spliced ​​feature map into the second deconvolution module in the decoder, and splicing the feature map output by the second deconvolution module with the second sub-feature map corresponding to the second convolution module; and looping in this way until the spliced ​​feature map is restored to the size of the target fused image.

[0099] Specifically, the decoder is composed of multiple upsampling modules, and the upsampling operation of the upsampling module can be implemented by a deconvolution module or a transposed convolution module. The upsampling module of each layer can perform a spatial dimension increase on the feature map output by the encoder to restore the size of the feature map. In the process of restoring the size of the feature map, the feature map can be spliced ​​with the feature map output by the corresponding convolution module of the encoder to ensure the retention of spatial information. This splicing process combines local features and contextual information, so that the feature map can better learn the boundaries of the line laser features during the restoration process.

[0100] S209, performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser characteristics;

[0101] S210, determining the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

[0102] Steps S209 to S210 in this embodiment are similar to those in the previous embodiment. Figure 1 Steps S104 to S105 in the illustrated embodiment are similar and will not be described in detail here.

[0103] like Figure 3 As shown, an embodiment of the surface feature detection device of the fuel assembly in the present invention comprises:

[0104] An acquisition unit 301 is used to acquire surface image data of the fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image;

[0105] A feature fusion unit 302 is used to perform feature fusion on the underexposed image, the normal exposed image and the overexposed image using a global context attention mechanism to obtain a target fused image;

[0106] An image segmentation unit 303 is used to perform image segmentation on the target fusion image using a U-Net algorithm to obtain line laser features;

[0107] A three-dimensional reconstruction unit 304 is used to perform three-dimensional reconstruction of the fuel assembly to be tested according to the line laser characteristics;

[0108] The determination unit 305 is used to determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

[0109] In this embodiment, the acquisition unit 301 acquires the surface image data of the fuel assembly to be tested at different exposure times, and the surface image data contains low-exposure images, normal-exposure images, and over-exposure images; the feature fusion unit 302 uses the global context attention mechanism to perform feature fusion on the low-exposure image, the normal-exposure image, and the over-exposure image to obtain a target fusion image; the image segmentation unit 303 uses the U-Net algorithm to perform image segmentation on the target fusion image to obtain line laser features; the three-dimensional reconstruction unit 304 performs three-dimensional reconstruction on the fuel assembly to be tested according to the line laser features; the determination unit 305 determines the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result. In this way, the surface images with different exposure times can be fused using the global context attention mechanism, and because the global context attention mechanism can dynamically adjust the content weights of images with different exposure times, the line laser features of images with different exposure times can be effectively retained in the target fusion image, so that the target fusion image can have both the anti-interference and anti-reflection characteristics of the low-exposure image and the high brightness characteristics of the image after the exposure time is increased. Thereby, the influence of underwater interference, insufficient light, and reflection on the surface of the fuel rod on the extraction of line laser features can be reduced, thereby improving the surface feature detection efficiency of the fuel assembly.

[0110] like Figure 4 As shown, an embodiment of the surface feature measurement system of the fuel assembly of the present invention comprises: an image acquisition device and Figure 3 The surface feature detection device of the fuel assembly in the embodiment shown, the image acquisition device and Figure 3 In the illustrated embodiment, the surface feature detection device of the fuel assembly is electrically connected, and the image acquisition device comprises: a fixing device 41, a transmission assembly 42 and an imaging assembly 43; the transmission assembly 42 is mounted on the fixing device 41; the imaging assembly 43 is movably connected to the transmission assembly 42, and the transmission assembly 42 is used to drive the imaging assembly 43 to move in a first direction and a second direction respectively, the first direction is a circumferential direction, and the second direction is a linear direction, and the imaging assembly 43 is used to acquire the surface image of the fuel assembly to be tested.

[0111] In this embodiment, the fixing device 41 may include a fixing bracket, a cable, a lifting ring and a motor platform, and the materials used may be 304 stainless steel or carbon steel, or aluminum alloy, to improve the corrosion resistance, stability and strength of the overall structure in water. The specific materials used are not limited here. First, the overall structure is locked to the bracket on site by a fixing bracket, and then the cables and lifting rings are used to further improve the overall stability, providing a stable testing environment for the imaging component 43. The motor platform is connected to the fixing bracket to provide a stable working platform for the transmission component 42.

[0112] The transmission assembly 42 is used to drive the imaging assembly 43 to move in two different directions, providing the imaging assembly 43 with two degrees of freedom in different directions, so that the imaging assembly 43 can continuously and repeatedly sample the same position of the fuel assembly to be tested at different exposure times.

[0113] like Figure 5 and Figure 6 As shown, optionally, the transmission assembly 42 includes: a first motor 421, a second motor 422 and a screw rod 423; one end of the screw rod 423 is respectively connected to the first motor 421 and the second motor 422, and the other end is connected to the imaging assembly 43, the first motor 421 is used to drive the imaging assembly 43 to move in a first direction through the screw rod 423, the second motor 422 is used to drive the imaging assembly 43 to move in a second direction through the screw rod 423, and the central axis of the first direction is perpendicular to the second direction.

[0114] Specifically, the first motor 421 is used to drive the screw rod 423 to perform rotational motion in the first direction, so as to drive the imaging assembly 43 fixed on the screw rod 423 to follow the motion. The second motor 422 is used to drive the screw rod 423 to perform linear motion in the second direction, so as to drive the imaging assembly 43 fixed on the screw rod 423 to follow the motion. The rotation center axis of the first direction is perpendicular to the second direction, such as: the first direction is a horizontal rotation direction, and the second direction is a vertical linear direction. The internal structure of the first motor 421 may include: a motor body 4211, a motor housing 4212, a front cover 4213, a rear cover 4214 and a plurality of first sealing rings 4215, and the motor body 4211 is fixed in the internal through hole of the motor housing 4212, for example, the motor body 4211 can be locked in the motor housing 4212 by screws. The front cover 4213 and the rear cover 4214 are respectively fixed at both ends of the motor housing 4212 to form a closed space with the motor housing 4212, and the motor body 4211 is fixed in the motor housing 4212. The connection between the front cover 4213 and the motor housing 4212 and the connection between the rear cover 4214 and the motor housing 4212 are both provided with a plurality of first sealing rings 4215 to improve the connection sealing, and the first sealing ring 4215 can be an O-ring or a square sealing ring, which is not limited here. The internal structure of the second motor 422 is similar to the internal structure of the first motor 421, and will not be repeated here.

[0115] like Figures 7 to 9 As shown, optionally, the imaging component 43 includes: a line laser generator 431, a camera 432 and a pressing plate 433; the pressing plate 433 includes a first pressing plate 4331 and a second pressing plate 4332, and grooves are respectively provided at both ends of the first pressing plate 4331 and the second pressing plate 4332; after the first pressing plate 4331 and the second pressing plate 4332 are fixedly connected, the line laser generator 431 and the camera 432 are respectively clamped in the installation space formed by the grooves at both ends.

[0116] Specifically, the line laser generator 431 is used to emit a line laser to the surface of the fuel assembly to be tested, and the camera 432 is used to collect the surface area image of the fuel assembly to be tested irradiated by the line laser. The first pressing plate and the second pressing plate are connected by bolts or buckles to form a pressing plate 433, and the line laser generator 431 and the camera 432 can be clamped in the installation space inside the pressing plate 433. Among them, the internal structure of the line laser generator 431 may include a first front face 4311, a first sealing kit 4312, a light-transmitting glass 4313, a laser generator 4314, a second sealing ring 4315, and a generator housing 4316; the laser generator 4314 is fixed inside the generator housing 4316; the first front face 4311 and the generator housing 4316 are sealed and connected by the compression of the second sealing ring 4315; the first sealing kit 4312 is nested outside the light-transmitting glass 4313 and assembled with the front face, thereby achieving a sealed connection. The internal structure of the camera 432 may include: a second front face 4321, a third sealing ring 4322, a fill light 4323, a lens 4324, a second sealing sleeve 4325 and a camera housing 4326; the lens 4324 and the fill light 4323 are both fixed in the camera housing 4326, and a second sealing sleeve 4325 is nested on the outside to improve the connection stability and sealing; the third sealing ring 4322 is nested in the front end groove of the camera housing 4326, and the axial and radial compression of the second sealing sleeve 4325 and the third sealing ring 4322 are respectively achieved through the assembly between the second front face 4321 and the camera housing 4326 to improve the overall watertightness of the camera.

[0117] It can be understood that the above embodiments only express the preferred implementation modes of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the patent scope of the present invention. It should be pointed out that, for ordinary technicians in this field, the above technical features can be freely combined without departing from the concept of the present invention, and several deformations and improvements can be made, which all belong to the protection scope of the present invention. Therefore, all equivalent changes and modifications made to the scope of the claims of the present invention should belong to the coverage of the claims of the present invention.

Claims

1. A method for detecting surface characteristics of a fuel assembly, characterized in that: include: S101, acquiring surface image data of a fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image; S102, using a global context attention mechanism to perform feature fusion on the underexposed image, the normally exposed image, and the overexposed image to obtain a target fused image; S103, performing image segmentation on the target fusion image using a U-Net algorithm to obtain line laser features; S104, performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser feature; S105. Determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

2. The surface feature detection method of the fuel assembly according to claim 1, characterized in that: The step S102 includes: Inputting the underexposed image, the normally exposed image, and the overexposed image into a global context attention module for preliminary feature extraction to obtain initial image features; Using a global context attention mechanism to fuse the preliminary image features to obtain fused image features; Using the initial image features as shortcut branches, establishing interlayer connections between the underexposure image, the normal exposure image, the overexposure image and the fused image features to construct a residual neural network; The output of the residual neural network is determined to be a target fused image.

3. The surface feature detection method of the fuel assembly according to claim 2, characterized in that: The global context attention mechanism is used to perform feature fusion on the preliminary image features, Get the fused image features, including: Dividing the preliminary image features into a first portion of features, a second portion of features, and a third portion of features; Inputting the second part of features into the first convolutional layer, and performing nonlinear activation on the output of the first convolutional layer using a Softmax function to obtain a first image feature; Multiplying the first image feature and the third part feature element by element to obtain a second image feature; Using a Leaky ReLU function to nonlinearly activate the second image feature, and inputting the second image feature after nonlinear activation into the second convolutional layer to obtain a third image feature; The third image feature is added to the first part of the features element by element to obtain a fused image feature.

4. The surface feature detection method of the fuel assembly according to claim 1, characterized in that: The step 103 comprises: Using an encoder of a U-Net algorithm to gradually extract features from the target fused image, and gradually reducing the size of the generated feature map; When the size of the feature map is reduced to a first preset threshold, the current feature map is input into the decoder of the U-Net algorithm for deconvolution operation, and the size of the feature map is gradually restored; When the size of the feature map is restored to the size of the target fused image, the probability distribution of each pixel in the current feature map is obtained using the Sigmoid function, and the pixel points belonging to the line laser feature are determined according to the second preset threshold.

5. The method for detecting the surface characteristics of a fuel assembly according to claim 4, characterized in that: The encoder using the U-Net algorithm gradually extracts features from the target fused image and gradually reduces the size of the generated feature map, including: Inputting the target fused image into the first convolution module in the encoder of the U-Net algorithm for feature extraction to obtain a feature map, wherein the first convolution module includes a third convolution layer, a fourth convolution layer and a Leaky ReLU function; Inputting the feature map into a first pooling layer to reduce the size of the feature map; When determining whether the size of the feature map is smaller than a first preset threshold; If not, the reduced-size feature map is sequentially input into the second convolution module and the second pooling layer in the encoder for secondary size reduction; This cycle is repeated until the size of the feature map is reduced to the first preset threshold.

6. The method for detecting the surface characteristics of a fuel assembly according to claim 5, characterized in that: The step of inputting the current feature map into the decoder of the U-Net algorithm to perform a deconvolution operation and gradually restore the size of the feature map includes: Inputting the feature map output by the encoder into the first deconvolution module in the decoder of the U-Net algorithm to increase the size of the feature map; Copying and cropping the feature map output by the first convolution module to obtain a first sub-feature map; Concatenating the feature map output by the first deconvolution module with the first sub-feature map; Determine whether the stitched feature map is smaller than the size of the target fused image; If yes, the concatenated feature map is input into the second deconvolution module in the decoder, and the feature map output by the second deconvolution module is concatenated with the second sub-feature map corresponding to the second convolution module; This cycle is repeated until the spliced ​​feature map is restored to the size of the target fused image.

7. A fuel assembly surface feature detection device, characterized in that: include: An acquisition unit, used to acquire surface image data of the fuel assembly to be tested at different exposure times, wherein the surface image data includes a low-exposure image, a normal-exposure image, and an over-exposure image; A feature fusion unit, configured to perform feature fusion on the underexposed image, the normally exposed image, and the overexposed image using a global context attention mechanism to obtain a target fused image; An image segmentation unit, used for performing image segmentation on the target fusion image by using a U-Net algorithm to obtain line laser features; A three-dimensional reconstruction unit, used for performing three-dimensional reconstruction of the fuel assembly to be tested according to the line laser feature; The determination unit is used to determine the surface feature detection result of the fuel assembly to be tested according to the three-dimensional reconstruction result.

8. A surface feature detection system for a fuel assembly, characterized in that: The device comprises an image acquisition device and the surface feature detection device of the fuel assembly as claimed in claim 7, wherein the image acquisition device is electrically connected to the surface feature detection device of the fuel assembly, and the image acquisition device comprises: a fixing device, a transmission component and an imaging component; The transmission assembly is mounted on the fixing device; The imaging assembly is movably connected to the transmission assembly, and the transmission assembly is used to drive the imaging assembly to move in a first direction and a second direction respectively, the first direction is a circumferential direction, and the second direction is a linear direction, and the imaging assembly is used to collect a surface image of the fuel assembly to be tested.

9. The surface feature detection system of the fuel assembly according to claim 8, characterized in that: The transmission assembly comprises: a first motor, a second motor and a screw rod; One end of the screw rod is respectively connected to the first motor and the second motor, and the other end is connected to the imaging assembly. The first motor is used to drive the imaging assembly to move in a first direction through the screw rod, and the second motor is used to drive the imaging assembly to move in a second direction through the screw rod, and the central axis of the first direction is perpendicular to the second direction.

10. The surface feature detection system of the fuel assembly according to claim 8, characterized in that: The imaging assembly comprises: a line laser generator, a camera and a pressing sheet; The pressing sheet comprises a first pressing plate and a second pressing plate, and grooves are respectively provided at both ends of the first pressing plate and the second pressing plate; After the first pressing plate and the second pressing plate are fixedly connected, the line laser generator and the camera are respectively clamped in the installation space formed by the grooves at both ends.

Images