A method for detecting coated fuel particles in an image

Abstract

The application discloses a kind of detection methods of coated fuel particles in image, comprising: obtaining the input image of slice containing coated fuel particles;The input image is input into the backbone network combined with feature pyramid structure to extract features, and the first feature map is obtained;The first feature map is input into anchor frame generation network, and the anchor frame is obtained, and the first feature map is corrected by anchor frame, and the second feature map after correction is obtained;The anchor frame and the second feature map are input into suggestion proposal network, and the suggestion region is obtained;The mapping of suggestion region on the second feature map is carried out region of interest alignment operation, and the third feature map of fixed size is obtained;The third feature map is input into full convolution network, and the class and position of coated fuel particles in input image are obtained.

Description

Technical Field

[0001] This invention relates to the field of coated fuel particle detection technology, and in particular to a method, apparatus, computing device, and readable storage medium for detecting coated fuel particles in an image. Background Technology

[0002] Nuclear energy, as the only clean, low-carbon, safe, and efficient baseload energy source, has enormous development potential. The High-Temperature Gas-Cooled Reactor (HTGR) is one of the most advanced reactor types in Generation IV nuclear power systems, possessing strong passive safety characteristics, primarily dependent on the coated fuel particles. Coated fuel particles (also referred to as particles in this invention) are the fuel elements of the HTGR, consisting of a fuel core at its center and four coating layers on the outside. These four coating layers are the main barriers preventing the release of fission products, and their performance directly affects the safe operation of the nuclear reactor; therefore, the performance of the coating layers is crucial to the safety of the HTGR. The coating layer thickness is an important factor affecting particle performance. After particle preparation, the coating layer thickness must be measured, and the measurement results are used to guide adjustments and improvements to the production process.

[0003] Methods for measuring coating thickness can generally be divided into non-destructive and destructive methods. Common non-destructive methods include X-ray radiography and 3D micro-CT. These methods ensure particle integrity but are very time-consuming and expensive. Ceramic slicing is a typical destructive method. This method heavily relies on human intervention and operation but has great development potential. Existing solutions have designed a vision system based on ceramic slicing to achieve automatic measurement of the thickness of each coating layer of particles. Before measuring the thickness of particles in the ceramic slice, they need to be accurately located. Inaccurate particle positioning or detection of particles whose thickness cannot be measured will affect the accuracy and efficiency of coating thickness measurement. Therefore, particle detection and positioning is a very important task.

[0004] Particle detection and localization is essentially target detection, a fundamental and important problem in computer vision. Existing solutions improve Faster-RCNN in automatic particle coating thickness measurement systems to detect particles in images, but there is still considerable room for improvement in their detection accuracy.

[0005] Therefore, the present invention provides a detection scheme for fuel particles coated in images to solve the problems in the prior art. Summary of the Invention

[0006] Therefore, the present invention provides a method, apparatus, computing device, and readable storage medium for detecting fuel particles coated in an image, in order to solve or at least alleviate the problems mentioned above.

[0007] According to a first aspect of the present invention, a method for detecting fuel-coated particles in an image is provided. The method includes: acquiring an input image containing a slice of fuel-coated particles; inputting the input image into a backbone network incorporating a feature pyramid structure for feature extraction to obtain a first feature map; inputting the first feature map into an anchor box generation network to obtain anchor boxes, and correcting the first feature map using the anchor boxes to obtain a corrected second feature map; inputting the anchor boxes and the second feature map into a proposal network to obtain a proposal region; performing a region of interest alignment operation on the mapping of the proposal region onto the second feature map to obtain a third feature map of fixed size; and inputting the third feature map into a fully convolutional network to obtain the category and location of the fuel-coated particles in the input image.

[0008] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, the backbone network includes a classification network. The input image is input into the backbone network with a feature pyramid structure for feature extraction, including: inputting the input network into the classification network, calculating through the convolution kernel in the classification network to obtain the output of multiple stages; performing an upsampling operation on the output of each stage to obtain multiple fourth feature maps; fusing the multiple fourth feature maps with the output of their corresponding stages; and performing a convolution operation on the fused result.

[0009] Optionally, the method for detecting fuel particles coated in an image according to the present invention further includes: downsampling the result obtained from the convolution operation.

[0010] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, the anchor frame generation network includes an anchor frame generation module and a feature adaptation module. The anchor frame generation module includes an anchor frame position prediction branch and an anchor frame shape prediction branch. Inputting a first feature map into the anchor frame generation network includes: the anchor frame generation module determining whether each feature point on the first feature map is the center point of the anchor frame based on the anchor frame position prediction score; the anchor frame generation module determining the width and height corresponding to the center point of the anchor frame based on the anchor frame shape prediction score, and transforming the width and height corresponding to the center point of the anchor frame to obtain the shape information of each anchor frame; the feature adaptation module performing a deformable convolution operation on the shape information of each anchor frame.

[0011] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, the width and height corresponding to the center point of the anchor frame are converted by the following formula:

[0012] w=σ·s·e dw

[0013] h=σ·s·e dh

[0014] Among them, the outputs dw and dh of the anchor frame shape prediction branch can be mapped to w and h, s is the step size, and σ is the hyperparameter.

[0015] Optionally, in the method for detecting fuel-coated particles in an image according to the present invention, the deformable convolution operation is performed by the following formula:

[0016] f′ i =N T (f i ,w i ,h i )

[0017] Among them, f i The feature value representing the position of the i-th anchor frame, (w i ,h i ) represents the width and height of the corresponding anchor frame, N T f′ represents a deformable convolution operation. i This represents the transformed eigenvalues.

[0018] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, performing a region of interest alignment operation on the mapping of the proposed region on the second feature map includes: dividing the proposed region into multiple units on an average basis; dividing each unit into multiple regions on an average basis according to the number of sampling points, and determining the position coordinates of each region; obtaining the pixel value corresponding to the position coordinates of each region by bilinear interpolation according to the position coordinates of each region; and performing a pooling operation on each unit according to the pixel value corresponding to the position coordinates to obtain the pixel value corresponding to each unit, thereby forming a third feature map.

[0019] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, the classification network is a ResNeXt network.

[0020] Optionally, in the method for detecting fuel particles coated in an image according to the present invention, the hyperparameter is 1.6.

[0021] According to a second aspect of the present invention, an apparatus for detecting fuel-coated particles in an image is provided, comprising: an image acquisition unit adapted to acquire an input image containing a slice of fuel-coated particles; a feature extraction unit adapted to input the input image into a backbone network incorporating a feature pyramid structure for feature extraction to obtain a first feature map; an anchor frame generation unit adapted to input the first feature map into an anchor frame generation network to obtain anchor frames, and to modify the first feature map using the anchor frames to obtain a modified second feature map; a proposal region generation unit adapted to input the anchor frames and the second feature map into a proposal network to obtain a proposal region; an alignment operation unit adapted to perform a region of interest alignment operation on the mapping of the proposal region onto the second feature map to obtain a third feature map of fixed size; and a particle recognition unit adapted to input the third feature map into a fully convolutional network to obtain the category and location of fuel-coated particles in the input image.

[0022] According to a third aspect of the present invention, a computing device is provided, comprising: at least one processor; and a memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing the methods described above.

[0023] According to a fourth aspect of the present invention, a readable storage medium storing program instructions is provided, which, when read and executed by a computing device, causes the computing device to perform the method described above.

[0024] The technical solution of this invention effectively extracts features of coated fuel particles by incorporating a feature pyramid structure into the backbone network. An anchor frame generation network, capable of generating suitable anchor frames based on a first feature map, improves detection accuracy, and the first feature map is corrected to obtain a second feature map matching the shape of the anchor frame. Furthermore, region-of-interest alignment is performed on the mapping of the proposed region onto the second feature map, further enhancing detection accuracy.

[0025] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it in accordance with the contents of the specification, and in order to make the above and other objects, features and advantages of the present invention more apparent and understandable, specific embodiments of the present invention are described below. Attached Figure Description

[0026] To achieve the foregoing and related objectives, certain illustrative aspects are described herein in conjunction with the following description and accompanying drawings. These aspects indicate various ways in which the principles disclosed herein may be practiced, and all aspects and their equivalents are intended to fall within the scope of the claimed subject matter. The foregoing and other objectives, features, and advantages of this disclosure will become more apparent from the following detailed description, taken in conjunction with the accompanying drawings. Throughout this disclosure, the same reference numerals generally refer to the same parts or elements.

[0027] Figure 1 A block diagram of the physical components (i.e., hardware) of the computing device 100 is shown;

[0028] Figure 2 A schematic diagram of coated fuel particles is shown;

[0029] Figure 3 A schematic diagram of a ceramic slice is shown;

[0030] Figure 4 A schematic diagram is shown for particle types for which the coating thickness cannot be measured;

[0031] Figure 5 A schematic diagram of the components of a microscopic imaging device is shown;

[0032] Figure 6A flowchart of a method 600 for detecting fuel particles coated in an image according to an embodiment of the present invention is shown;

[0033] Figure 7 A schematic diagram of the structure of an improved Faster-RCNN according to an embodiment of the present invention is shown;

[0034] Figure 8 A schematic diagram illustrating the rendering of particles in an input image according to an embodiment of the present invention is shown;

[0035] Figure 9 A schematic diagram of the backbone network incorporating a feature pyramid structure according to an embodiment of the present invention is shown;

[0036] Figure 10 A schematic diagram of the anchor frame generation network according to an embodiment of the present invention is shown;

[0037] Figure 11 A schematic diagram of the process for aligning regions of interest according to an embodiment of the present invention is shown;

[0038] Figure 12 A schematic diagram illustrating the detection effect of the method for detecting fuel-coated particles in images according to the present invention on a test set is shown.

[0039] Figure 13 A magnified view of the detection results is shown;

[0040] Figure 14 The results of different detection methods are shown in magnified views.

[0041] Figure 15 A schematic diagram is shown of suggestion boxes generated by the suggestion proposal network using different anchor box generation methods;

[0042] Figure 16 A schematic diagram of a fuel particle detection device 1600 according to an embodiment of the present invention is shown. Detailed Implementation

[0043] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0044] Figure 1A block diagram of the physical components (i.e., hardware) of a computing device 100 is shown. In a basic configuration, the computing device 100 includes at least one processing unit 102 and system memory 104. According to one aspect, depending on the configuration and type of the computing device, the processing unit 102 may be implemented as a processor. The system memory 104 includes, but is not limited to, volatile memory (e.g., random access memory), non-volatile memory (e.g., read-only memory), flash memory, or any combination of such memory. According to one aspect, the system memory 104 includes an operating system 105 and a program module 106, the program module 106 including a detection device 1600 for fuel particles coated in an image.

[0045] According to one aspect, operating system 105 is, for example, suitable for controlling the operation of computing device 100. Furthermore, examples are practiced in conjunction with graphics libraries, other operating systems, or any other applications, and are not limited to any particular application or system. Figure 1 The basic configuration is illustrated by the components within the dashed lines 108. According to one aspect, the computing device 100 has additional features or functions. For example, according to one aspect, the computing device 100 includes additional data storage devices (removable and / or non-removable), such as disks, optical discs, or magnetic tapes. This additional storage... Figure 1 The image is shown by removable storage device 109 and non-removable storage device 110.

[0046] As stated above, according to one aspect, a program module is stored in system memory 104. According to one aspect, the program module may include one or more applications. The present invention does not limit the type of application; for example, applications may include: email and contact applications, word processing applications, spreadsheet applications, database applications, slideshow applications, drawing or computer-aided applications, web browser applications, etc.

[0047] According to one aspect, examples can be practiced on circuits including discrete electronic components, packaged or integrated electronic chips containing logic gates, circuits utilizing microprocessors, or on a single chip containing electronic components or a microprocessor. For example, it can be practiced via wherein... Figure 1Each or many of the components shown can be implemented as an example by integrating a System-on-a-Chip (SOC) on a single integrated circuit. According to one aspect, such an SOC device may include one or more processing units, graphics units, communication units, system virtualization units, and various application functions, all integrated (or “burned in”) as a single integrated circuit onto a chip substrate. When operating via the SOC, the functions described herein can be operated via dedicated logic integrated on a single integrated circuit (chip) with other components of the computing device 100. Embodiments of the invention can also be implemented using other techniques capable of performing logical operations (e.g., AND, OR, and NOT), including but not limited to mechanical, optical, fluid, and quantum technologies. Additionally, embodiments of the invention can be implemented within a general-purpose computer or in any other circuit or system.

[0048] According to one aspect, computing device 100 may also have one or more input devices 112, such as a keyboard, mouse, pen, voice input device, touch input device, etc. It may also include output devices 114, such as a display, speaker, printer, etc. The foregoing devices are examples and other devices may also be used. Computing device 100 may include one or more communication connections 116 that allow communication with other computing devices 118. Examples of suitable communication connections 116 include, but are not limited to: RF transmitter, receiver and / or transceiver circuitry; Universal Serial Bus (USB), parallel and / or serial ports.

[0049] As used herein, the term computer-readable medium includes computer storage medium. Computer storage medium can include volatile and non-volatile, removable and non-removable media implemented using any method or technology for storing information (e.g., computer-readable instructions, data structures, or program modules). System memory 104, removable storage device 109, and non-removable storage device 110 are examples of computer storage media (i.e., memory storage). Computer storage media can include random access memory (RAM), read-only memory (ROM), electrically erasable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical storage, magnetic tape, magnetic tape, disk storage or other magnetic storage devices, or any other article of manufacture that can be used to store information and is accessible by computer device 100. According to one aspect, any such computer storage medium can be part of computing device 100. Computer storage media does not include carrier waves or other transmitted data signals.

[0050] According to one aspect, a communication medium is implemented by computer-readable instructions, data structures, program modules, or other data in a modulated data signal (e.g., a carrier wave or other transmission mechanism), and includes any information transmission medium. According to one aspect, the term "modulated data signal" describes a signal having one or more sets of characteristics or altered in a manner that encodes information in the signal. By way of example and not limitation, a communication medium includes wired media such as wired networks or direct wired connections, and wireless media such as acoustic, radio frequency (RF), infrared, and other wireless media.

[0051] In some embodiments of the present invention, the computing device 600 includes one or more processors and one or more readable storage media storing program instructions. When the program instructions are configured to be executed by one or more processors, the computing device causes the computing device to perform the image-coated fuel particle detection method of the embodiments of the present invention.

[0052] High-temperature gas-cooled reactors (HTGRs), as one of the most advanced reactor types in fourth-generation nuclear power systems, possess inherent advantages such as safety, high power generation efficiency, and cleanliness. Coated fuel particles are the fuel elements of HTGRs. Figure 2 A schematic diagram of coated fuel particles is shown, such as Figure 2 As shown, the particle's center is a uranium-containing fuel core, surrounded by four cladding layers: a loose pyrolytic carbon layer, an inner dense pyrolytic carbon layer, a silicon carbide layer, and an outer dense pyrolytic carbon layer, from the inside out. The thickness of these cladding layers is a crucial factor affecting particle performance and the safe operation of the high-temperature gas-cooled reactor.

[0053] Figure 3 A schematic diagram of a ceramic slice is shown. Hundreds of particles are embedded in resin to form a ceramic slice, which is then ground and polished to the equatorial plane of the particles, resulting in a product such as... Figure 3 The ceramic slice shown. Figure 4 This diagram illustrates the types of particles whose coating thickness cannot be measured. After grinding and polishing ceramic slices, many particles with unmeasurable coating thickness appear, such as... Figure 4 As shown in the boxes, the four types of particles corresponding to boxes 410-440 are unsuitable for measuring the coating thickness. Specifically, 410 represents particles with core detachment, 420 represents particles with coating detachment, 430 represents particles with severely damaged cores, and 440 represents particles with excessive grinding. If these particles are misdetected during inspection, useless data will be collected subsequently, reducing measurement efficiency. Therefore, this invention designs a method for detecting coated fuel particles in images, incorporating an advanced deep learning model, to achieve high-precision particle detection in images.

[0054] This invention uses a microscopic imaging device to image ceramic slices. Figure 5A schematic diagram of the components of a microscopic imaging apparatus is shown. (For example...) Figure 5 As shown, the microscopic imaging device 510 consists of the following parts: a microscopic camera 511, an eyepiece 512, a light source 513, a 5x objective lens 514, a 10x objective lens 515, a ceramic slide 516, an automatic stage 517, a lifting device 519, a controller 518, and an optical platform 520.

[0055] The microscope camera is used to acquire images of the particles on the surface of ceramic slices. The imaging effects under 5x and 10x objectives are as follows: Figure 5 As shown in 530 and 540. In the imaging experiment, the ceramic slice was moved by controlling the automatic stage, and an automatic focusing device was used to achieve automatic image acquisition.

[0056] Particle images acquired under a 10x objective lens are used to measure the thickness of each coating layer of the particle. However, before acquisition, the particle's coordinates must be accurately obtained from the particle image under a 5x objective lens. If the obtained particle coordinates are off, part of the particle will be outside the field of view when acquiring the image under the 10x objective lens, making it impossible to measure its coating layer thickness. Furthermore, if particles whose coating layer thickness cannot be measured are falsely detected, subsequent acquisition of useless data will reduce efficiency. Therefore, accurate particle localization is a crucial task.

[0057] Figure 6 A flowchart of a method 600 for detecting fuel-coated particles in an image according to an embodiment of the present invention is shown. Method 600 can be executed in a computing device (e.g., the aforementioned computing device 100). Figure 6 As shown, method 600 begins with step 610.

[0058] In step 610, an input image containing a slice of coated fuel particles is obtained.

[0059] In some embodiments, an input image of a slice containing fuel-coated particles is acquired using a microscopic imaging device (such as the aforementioned microscopic imaging device 510). Acquiring the input image using the microscopic imaging device designed in this invention provides high image acquisition efficiency, ensuring accurate particle positioning in the subsequent process. Alternatively, other devices with image magnification and image acquisition capabilities can be used to magnify and acquire the image of the slice containing fuel-coated particles to obtain the input image.

[0060] Subsequently, in step 620, the input image is fed into the backbone network with the feature pyramid structure for feature extraction to obtain the first feature map.

[0061] This invention improves upon Faster-RCNN and includes five modules: backbone network, anchor box generation network, proposal network (RPN), region of interest alignment (ROI align), and fully convolutional network (FCN).

[0062] In some embodiments, the input network is fed into a classification network, and the outputs of multiple stages are calculated by the convolutional kernels in the classification network. The outputs of each stage are upsampled to obtain multiple fourth feature maps. These multiple fourth feature maps are fused with the outputs of their corresponding stages. The fused result is then convolved to obtain a first feature map. Optionally, the result of the convolution operation can also be downsampled to obtain the first feature map.

[0063] Figure 7 A schematic diagram of the structure of an improved Faster-RCNN according to an embodiment of the present invention is shown.

[0064] In some embodiments, the backbone network includes a classification network, specifically a ResNeXt network. The input image is first processed by a ResNeXt backbone network with a feature pyramid structure to extract features, resulting in a first feature map for subsequent anchor box generation. The anchor box generation network uses the first feature map to guide the generation of sparse anchor boxes and modifies the first feature map to match the generated anchor boxes. Subsequently, the anchor boxes and the modified second feature map are input into the RPN. The RPN has two branches: a binary classification network and a regression network. The binary classification network determines the category of the anchor box, whether it belongs to the target or the background. The regression network corrects the position coordinates of the anchor boxes, forming more accurate proposal regions. The mapping of the proposal regions onto the second feature map is then subjected to a ROI alignment operation to obtain a fixed-size third feature map. Finally, these fixed-size third feature maps are input into the FCN for specific classification and position regression, ultimately obtaining the category and precise location of the target (i.e., particles).

[0065] Because particles prepared in different batches may vary in actual production, and the grinding and polishing process of ceramic slices may cause some damage to the particles, thus increasing the difficulty of feature extraction. In addition, the particles in the image are densely packed and there are many particles that extend beyond the image's field of view. Figure 8 A schematic diagram illustrating the rendering of particles in an input image according to an embodiment of the present invention is shown. Figure 8As shown, 801-826 represent particles that extend beyond the image's field of view, indicating that a portion of the labeled particles exceeds the image's field of view. 827-840 represent particles that are entirely within the image's field of view, indicating that the labeled particles are completely within the image. The extent of particles extending beyond the image varies, thus presenting detection challenges related to multiple scales and small particle sizes. Faster-RCNN uses the last layer feature map extracted by the VGG backbone network for subsequent processing, but this last layer feature map often undergoes several downsampling iterations. Small particles inherently possess less pixel information, making them prone to information loss during downsampling, thus affecting their detection performance. Therefore, to avoid this drawback, this invention selects the ResNeXt network as the feature extraction network (classification network) and adds a feature pyramid structure to it to obtain feature maps with strong semantic and spatial information.

[0066] The ResNeXt network borrows the stacking concept and residual structure of the ResNet network, as well as the split-transform-merge strategy of the Inception network. It can increase accuracy while maintaining or reducing network complexity and possessing strong scalability. Without increasing network complexity, the ResNeXt network has better feature representation capabilities and can better extract features from measurable particles. Building on this, this invention introduces a feature pyramid structure to address the issues of incomplete, multi-scale, and small-sized measurable particles. Figure 9 A schematic diagram of a backbone network incorporating a feature pyramid structure according to an embodiment of the present invention is shown. Figure 9 As shown, the backbone network of the feature pyramid structure has three routes: bottom-up, top-down, and lateral connections. The bottom-up route is the feedforward process of the ResNeXt network, where the feature maps are typically reduced in size after being processed by convolutional kernels. However, some feature maps retain their original size; these layers are grouped into a single stage. Here, the outputs of the last layer in each stage of the ResNeXt network are labeled C1, C2, C3, C4, and C5, respectively. The top-down route in the feature pyramid structure uses upsampling. Lateral connections fuse the upsampled feature maps from the feature pyramid structure with the same-sized feature maps generated by the bottom-up ResNeXt network. After fusion, a 3×3 convolutional kernel is used to convolve the fused result to eliminate the aliasing effect of upsampling. The resulting images are labeled P2, P3, P4, and P5, respectively. Additionally, P5 is downsampled by 0.5 to obtain P6, which is then input into the RPN.

[0067] Next, in step 630, the first feature map is input into the anchor box generation network to obtain anchor boxes, and the first feature map is corrected using the anchor boxes to obtain the corrected second feature map.

[0068] The anchor box generation network comprises an anchor box generation module and a feature adaptation module. The anchor box generation module includes an anchor box position prediction branch and an anchor box shape prediction branch. The anchor box generation module uses the anchor box position prediction score to determine whether each feature point on the first feature map is the center point of the anchor box. The anchor box generation module uses the anchor box shape prediction score to determine the width and height corresponding to the anchor box center point, and transforms the width and height corresponding to the anchor box center point to obtain the shape information of each anchor box.

[0069] Faster-RCNN manually pre-sets nine anchor boxes based on experience. This invention, after adding a feature pyramid structure to the backbone network, pre-sets three anchor boxes with different aspect ratios for each of the five feature maps, resulting in a total of 15 pre-set anchor boxes. Using universal anchor boxes for different problems may affect detection accuracy. It also generates a large number of proposed regions, leading to low computational efficiency. This invention introduces a guided anchor box method and designs an anchor box generation network. The anchor box generation network uses semantic features to guide the generation of sparse and arbitrarily shaped anchor boxes, and uses a feature adaptation module to correct the first feature map to better match the anchor box shape. Figure 10 A schematic diagram of the anchor frame generation network according to an embodiment of the present invention is shown. Figure 10 As shown,

[0070] The anchor box generation network can be divided into an anchor box generation module and a feature adaptation module. The anchor box generation module has two branches: the anchor box position prediction branch (NL) and the anchor box shape prediction branch (NS). The anchor box position prediction branch is used to determine whether each feature point on the feature map is the center point of the anchor box. The anchor box shape prediction branch is used to determine the width and height of the corresponding anchor box center point. Since the width and height of the anchor box are variable values ​​and their values ​​are relatively large, it is difficult to predict them directly. Therefore, the following formula is used to transform the width and height corresponding to the anchor box center point:

[0071] w=σ·s·e dw

[0072] h=σ·s·e dh

[0073] In this model, the outputs dw and dh of the anchor frame shape prediction branch can be mapped to w and h, s is the step size, and σ is the hyperparameter. Optionally, the hyperparameter is 1.6.

[0074] Combining position prediction with shape prediction can generate sparse anchor boxes with diverse shapes and positions, but the width and height of these anchor boxes are not guided by the receptive field size of the feature map. Traditional anchor boxes have a consistent shape on feature maps of the same scale, but anchor boxes generated by anchor box generation networks vary considerably and cannot be well matched with the feature map. The feature adaptation module incorporates the shape information of the anchor boxes into the first feature map, and the second feature map after information fusion can adapt to the shape of the anchor box at each position. This module utilizes the idea of ​​deformable convolution to transform the features at each position individually according to the shape. The feature adaptation module performs deformable convolution operations on the shape information of each anchor box, where the deformable convolution operation is performed by the following formula:

[0075] f i ′=N T (f i ,w i ,h i )

[0076] Among them, f i The feature value representing the position of the i-th anchor frame, (w i ,h i ) represents the width and height of the corresponding anchor frame, N T f represents deformable convolution operation. i ′ represents the transformed eigenvalue.

[0077] Subsequently, in step 640, the anchor box and the second feature map are input into the proposal network to obtain the proposal region.

[0078] Subsequently, in step 650, the region of interest alignment operation is performed on the mapping of the proposed region onto the second feature map to obtain a third feature map of fixed size.

[0079] In some embodiments, the proposed region is divided into multiple units on an average basis. Based on the number of sampling points, each unit is further divided into multiple regions on an average basis, and the position coordinates of each region are determined. Based on the position coordinates of each region, the pixel value corresponding to the position coordinates of each region is obtained using bilinear interpolation. Based on the pixel value corresponding to the position coordinates, pooling is performed on each unit to obtain the pixel value corresponding to each unit, forming a third feature map.

[0080] Region of Interest (ROI) pooling pools feature maps corresponding to proposal regions of different sizes into a fixed-size feature map for subsequent classification and bounding box regression operations. The proposal regions are obtained through RPN regression, and their position parameters are typically floating-point numbers. Since the pooled feature map needs to be of a fixed size, ROI pooling involves two quantization operations. First, the position parameters of the proposal region are quantized to integers. Second, the quantized proposal region is divided into n×n units, and the boundaries of each unit are quantized. After these two quantizations, the position of the proposal region will have a certain deviation, which will affect the accuracy of object detection.

[0081] To improve the detection accuracy of measurable particles, this invention replaces the region of interest (ROI) alignment method. The improved approach eliminates the quantization operation and uses bilinear interpolation to obtain floating-point pixel values. Figure 11 A schematic flowchart illustrating region-of-interest alignment according to an embodiment of the present invention is shown. Figure 11 As shown, firstly, keeping the floating-point values ​​of the suggested region location parameters unchanged, the region is divided into n×n units, with the boundaries of each unit being floating-point numbers. Then, based on the number of sampling points s and bilinear interpolation, s location coordinates within each unit are calculated (the coordinates of the center points of each of the s regions after dividing a unit into equal parts). Finally, max or average pooling is performed on each unit.

[0082] Subsequently, in step 660, the third feature map is input into the fully convolutional network to obtain the category and location of the fuel particles coated in the input image.

[0083] This invention validates the proposed method for detecting fuel-coated particles in images by constructing a dataset. First, the construction of the validation dataset and the setting of relevant parameters are explained. The dataset constructed in this paper is used for training and testing. Then, the proposed method is compared with several superior object detection methods, including Mask-RCNN, Cascade-RCNN, Grid-RCNN, RetinaNet, FCOS, Faster-RCNN, and Improved Faster-RCNN. Subsequently, the three improvements proposed in this paper are validated to demonstrate the effectiveness of the proposed method.

[0084] This invention uses a microscopic imaging device to acquire images of 20 different ceramic slices, obtaining a total of 884 images to construct a dataset. The dataset is randomly shuffled into a training set, a validation set, and a test set, with 650, 117, and 117 images respectively. The training set is used to train the model's parameters, and the validation set is used to evaluate the model's generalization ability after each training iteration or hyperparameter adjustment. After model training is complete, this invention uses the test set to evaluate the model's generalization ability.

[0085] To avoid overfitting, this paper augments the training set to ensure data diversity and richness. A random image flipping method is used, meaning there is a 50% probability that an image will be flipped horizontally, vertically, or both.

[0086] After the dataset was constructed, this paper used Labelme software to locate and label all images, which have only one label called "particle". Due to the limited GPU memory and the original image size of 3072×4088, this paper uniformly adjusted the images to 768×1022.

[0087] The main hardware specifications used in this invention are an Intel i5-9400f processor and an Nvidia GeForce GTX2060 super GPU, and the operating system is Ubuntu 20.04LTS 64-bit.

[0088] This invention first trains the backbone network on the ImageNet dataset to initialize network parameters. This method allows the model to learn common features in the data effectively, which, when applied to the dataset used in this invention, enables the model to converge faster and more smoothly. For parameters of layers other than the backbone network, a Gaussian distribution with a mean of 0 and a standard deviation of 0.01 is used for initialization. This invention chooses stochastic gradient descent to train the model and employs a warm-up strategy: an initial learning rate of 0.002, which is then gradually increased to 0.02, and finally decreased using a decay rate of 0.001.

[0089] To verify the model's performance, this paper uses test set images for testing after the model has been trained. Figure 12 A schematic diagram illustrating the detection effect of the method for detecting fuel-coated particles in images according to the present invention on a test set is shown. For example... Figure 12 As shown, all particles in each image were accurately detected, and there were no false positives indicating particles for which the coating thickness could not be measured. Figure 12 In the top row, from left to right, the number of particles in the image increases sequentially, and the method of this invention can accurately detect them all. This shows that regardless of whether the particles in the image are densely distributed, the method proposed in this invention has good detection performance. Figure 12In the two images on the left of the bottom row, neither excessively ground particles nor particles with detached cores were detected. Similarly, in the two images on the right of the bottom row, particles with severely damaged cores were also not detected. This indicates that the model of this invention will not misidentify particles for which the coating thickness cannot be measured. Figure 12 These images demonstrate that the detection results are highly accurate. Below is a magnified view of a localized portion of the detection performance. Figure 13 A magnified view of the detection results is shown. (For example...) Figure 13 As shown, the detection effect of the method proposed in this invention is not only very accurate from a macroscopic perspective, but also, when magnified locally, each frame is exactly tangent to the four boundaries (top, bottom, left, and right) of the measurable particle. This indicates that the method of this invention has very high precision.

[0090] Furthermore, this invention uses accuracy (AR), average recall (AP), and fractional power per second (FPS) as indicators to evaluate the effectiveness of the method. The calculation formulas for AP and AR are shown below.

[0091]

[0092]

[0093]

[0094]

[0095] In this context, TP represents correctly detected bounding boxes (true positives), FP represents bounding boxes that detect background or incorrect categories (false positives), FN represents missed bounding boxes, P (Precision) represents accuracy, and R (Recall) represents recall.

[0096] The proposed method achieves the following image target detection results on the constructed dataset: AP reaches 99.0%, AR reaches 99.7%, and FPS reaches 9.4. The proposed method's efficiency meets the detection requirements, and the high AP and AR values ​​indicate that the method provides accurate and comprehensive detection results.

[0097] The various object detection methods were trained using the dataset constructed in this paper. After convergence, the methods were validated on the test set. The results showed that, from a macroscopic perspective, all object detection methods exhibited good detection performance. Therefore, this invention provides a localized magnification of the detection results. Figure 14 The diagram shows magnified views of the detection results of different detection methods. For example... Figure 14As shown, 1401 corresponds to a magnified view of the detection effect of Faster-RCNN, 1402 corresponds to Mask-RCNN, 1403 corresponds to Cascade-RCNN, 1404 corresponds to Grid-RCNN, 1405 corresponds to RetinaNet, 1406 corresponds to FCOS, 1407 corresponds to Zhang's Improved Faster-RCNN, and 1408 corresponds to the method proposed in this invention.

[0098] By magnifying a portion of the detection results, it was found that in other comparative methods, a portion of the marked frame could not be truly tangent to the particle. The greater the deviation between the detection frame and the particle boundary, the greater the impact on the thickness measurement accuracy. A magnified view of the detection effect of the method proposed in this invention is shown below. Figure 13 and Figure 14 As shown in 1408, it can be observed that regardless of the detection frame, it can be tangent to the particle. Compared with other target detection methods, the method proposed in this invention has a good detection effect on particles.

[0099] In addition, the present invention also used AP and FPS metrics to evaluate the aforementioned method, and the detection performance of different target detection arrays is shown in the table below.

[0100]

[0101] Experimental results show that the proposed method has better detection accuracy than other target detection models, with an average detection accuracy of 99.0%.

[0102] To compare the impact of different backbone networks and feature pyramid structures on detection results, this paper modifies the feature extraction backbone network and adds a feature pyramid structure to the Faster R-CNN model, analyzing its impact on detection results. To ensure fairness, all parameters except the backbone network are kept consistent. The detection results using different backbone networks are shown in the table below.

[0103]

[0104]

[0105] Experimental results show that the model using ResNeXt-50 as the backbone network has higher detection accuracy than models using VGG16, ResNet-50, and Inception V3 as backbone networks, achieving an AP of 95.6%. Adding a feature pyramid structure to these backbone networks all improved detection accuracy to some extent, with the method using a ResNeXt-50 network combined with a feature pyramid structure as the backbone network achieving the highest accuracy of 97.6%. However, adding a feature pyramid structure to the backbone network reduced efficiency to some degree. These results indicate that the ResNeXt-50 network has stronger feature representation capabilities compared to VGG16, ResNet-50, and Inception V3 networks, and the feature pyramid structure can better handle the problem of multi-scale and small-sized particles in images.

[0106] To address the issues of mismatched preset anchor frame sizes and difficulty in parameter adjustment, thereby improving detection accuracy, this invention designs an anchor frame generation network to generate anchor frames. The detection results using the original anchor frame generation method and the AGN method are shown in the table below. Experimental results demonstrate that the anchor frame generation method using the anchor frame generation network improves the model's detection accuracy to a certain extent. The average detection accuracy is improved by 0.7%, while efficiency only decreases slightly.

[0107]

[0108] To verify that RPN using AGN can generate more suitable suggestion boxes, this invention visualizes the suggestion boxes generated in the RPN stage. Figure 15 This diagram illustrates suggestion boxes generated by the suggestion proposal network using different anchor box generation methods. For example... Figure 15 As shown, 1510 and 1520 represent the input image, 1530 and 1540 represent proposals obtained using the original anchor box generation method, and 1550 and 1560 represent proposals obtained using AGN. Experimental results show that proposals obtained using the original anchor box generation method are distributed throughout the entire image, while proposals obtained using AGN are densely distributed around the target object and sparsely distributed in other non-target areas. The method proposed in this invention can effectively improve the quality of proposals, thereby improving detection accuracy.

[0109] To further improve particle detection accuracy, this paper replaces the ROI pooling operation with the ROI align operation. The detection results using ROI pooling and ROI align are shown in the table below. Experimental results show that the detection accuracy using the ROI align method is higher than that of ROI pooling, with an average improvement of 0.7%, but the FPS value remains unchanged. Although bilinear interpolation is introduced in the ROI align operation, its computational cost is negligible compared to the entire calculation process.

[0110]

[0111] This invention proposes a high-precision particle detection method based on an improved Faster-RCNN. Particles from different batches exhibit inherent differences and may be damaged during ceramic slice preparation, resulting in densely packed, small-sized particles with inconsistent scales in the images. Therefore, this invention employs ResNeXt-50 as the backbone network and adds a feature pyramid structure to it to obtain feature maps with high semantic and spatial information. In the anchor box generation stage, this paper designs an anchor box generation network that adaptively generates suitable anchor boxes based on the feature maps. By increasing computational cost slightly, the detection accuracy is improved. Furthermore, this paper replaces ROI pooling with ROI alignment to further enhance detection accuracy. Experimental results show that the proposed method achieves an average detection accuracy of 99.0% for fuel-coated particles, outperforming superior object detection methods such as Mask-RCNN, Cascade-RCNN, Grid-RCNN, RetinaNet, and FCOS, as well as the original Faster-RCNN and Zhang's Improved Faster-RCNN. Overall, the detection method for coated fuel particles proposed in this invention can achieve good detection results and provides effective technical support for the automatic measurement of the thickness of each coating layer of particles.

[0112] The present invention also provides a device for detecting fuel particles coated in an image. Figure 16 A schematic diagram of a detection device 1600 for detecting fuel particles coated in an image according to an embodiment of the present invention is shown. Figure 16 As shown, the detection device 1600 for coated fuel particles includes an image acquisition unit 1610, a feature extraction unit 1620, an anchor frame generation unit 1630, a suggested region generation unit 1640, an alignment operation unit 1650, and a particle recognition unit 1660.

[0113] The image acquisition unit 1610 is adapted to acquire an input image containing a slice of fuel-coated particles.

[0114] The feature extraction unit 1620 is adapted to input the input image into the backbone network with the feature pyramid structure for feature extraction to obtain the first feature map.

[0115] Anchor frame generation unit 1630 is adapted to input the first feature map into the anchor frame generation network to obtain anchor frames, and to modify the first feature map through the anchor frames to obtain the modified second feature map.

[0116] The proposed region generation unit 1640 is adapted to input anchor boxes and second feature maps into the proposal network to obtain proposed regions.

[0117] Alignment operation unit 1650 is adapted to perform region of interest alignment operation on the mapping of the proposed region on the second feature map to obtain a third feature map of fixed size.

[0118] The particle recognition unit 1660 is adapted to input the third feature map into a fully convolutional network to obtain the category and location of fuel particles coated in the input image.

[0119] It should be noted that the working principle and process of the device 1600 provided by the present invention are similar to those of the method 600 described above. For relevant details, please refer to the description of the method 600 described above, which will not be repeated here.

[0120] The technical solution of this invention effectively extracts features of coated fuel particles by incorporating a feature pyramid structure into the backbone network. An anchor frame generation network, capable of generating suitable anchor frames based on a first feature map, improves detection accuracy, and the first feature map is corrected to obtain a second feature map matching the shape of the anchor frame. Furthermore, region-of-interest alignment is performed on the mapping of the proposed region onto the second feature map, further enhancing detection accuracy.

[0121] The various techniques described herein can be implemented in combination with hardware or software, or a combination thereof. Thus, the methods and apparatus of the present invention, or certain aspects or portions thereof, can take the form of program code (i.e., instructions) embedded in a tangible medium, such as a removable hard disk, USB flash drive, floppy disk, CD-ROM, or any other machine-readable storage medium, wherein when the program is loaded into and executed by a machine such as a computer, the machine becomes an apparatus for practicing the present invention.

[0122] When the program code is executed on a programmable computer, the mobile terminal generally includes a processor, a processor-readable storage medium (including volatile and non-volatile memory and / or storage elements), at least one input device, and at least one output device. The memory is configured to store program code; the processor is configured to execute the method for detecting fuel-coated particles in an image according to instructions in the program code stored in the memory.

[0123] By way of example, and not limitation, readable media include readable storage media and communication media. Readable storage media stores information such as computer-readable instructions, data structures, program modules, or other data. Communication media generally embodies computer-readable instructions, data structures, program modules, or other data in the form of modulated data signals such as carrier waves or other transmission mechanisms, and includes any information delivery medium. Any combination of the above is also included within the scope of readable media.

[0124] In the specification provided herein, the algorithms and displays are not inherently related to any particular computer, virtual system, or other device. Various general-purpose systems can also be used with the examples of this invention. The required structure for constructing such systems is apparent from the above description. Furthermore, this invention is not directed to any particular programming language. It should be understood that the contents of the invention described herein can be implemented using various programming languages, and the above description of specific languages ​​is for the purpose of disclosing the best mode of implementation of the invention.

[0125] Numerous specific details are set forth in the specification provided herein. However, it will be understood that embodiments of the invention may be practiced without these specific details. In some instances, well-known methods, structures, and techniques have not been shown in detail so as not to obscure the understanding of this specification.

[0126] Similarly, it should be understood that, in order to streamline this disclosure and aid in understanding one or more of the various aspects of the invention, in the above description of exemplary embodiments of the invention, various features of the invention are sometimes grouped together in a single embodiment, figure, or description thereof.

[0127] Those skilled in the art will understand that modules, units, or components of the devices disclosed in the examples herein can be arranged in the devices described in this embodiment, or alternatively, can be located in one or more devices different from the devices in this example. The modules in the foregoing examples can be combined into a single module or, in addition, can be divided into multiple sub-modules.

[0128] Those skilled in the art will understand that modules in the device of the embodiments can be adaptively changed and placed in one or more devices different from that embodiment. Modules, units, or components in the embodiments can be combined into a single module, unit, or component, and further, they can be divided into multiple sub-modules, sub-units, or sub-components. Except where at least some of such features and / or processes or units are mutually exclusive, any combination can be used to combine all features disclosed herein and all processes or units of any method or device so disclosed. Unless expressly stated otherwise, each feature disclosed herein may be replaced by an alternative feature that serves the same, equivalent, or similar purpose.

[0129] Furthermore, those skilled in the art will understand that although some embodiments described herein include certain features included in other embodiments but not others, combinations of features from different embodiments are meant to be within the scope of the invention and form different embodiments.

[0130] Furthermore, some of the embodiments described herein are methods or combinations of method elements that can be implemented by a processor of a computer system or by other means of performing the functions. Therefore, a processor having the necessary instructions for implementing the methods or method elements forms means for implementing the methods or method elements. Furthermore, the elements described herein in the apparatus embodiments are examples of means for implementing the functions performed by elements for the purposes of carrying out the invention.

[0131] As used herein, unless otherwise specified, the use of ordinal numbers such as “first,” “second,” “third,” etc., to describe ordinary objects merely indicates different instances of similar objects and is not intended to imply that the objects being described must have a given order in time, space, ordering, or any other manner.

[0132] Although the invention has been described with respect to a limited number of embodiments, those skilled in the art will understand from the foregoing description that other embodiments are conceivable within the scope of the invention described herein. Furthermore, it should be noted that the language used in this specification has been chosen primarily for readability and edibility purposes, and not for the purpose of explaining or limiting the subject matter of the invention.

Claims

1. A method for detecting the thickness of a coating layer on fuel particles in an image, the method comprising: Obtain an input image containing a slice of the coated fuel particles, the input image also including coated fuel particles whose coating thickness cannot be measured; The input image is input into the classification network of the trained model, and after being processed by the convolution kernels in the classification network, the output of multiple stages is obtained, including C2, C3, C4 and C5. The output of each stage is upsampled to obtain multiple fourth feature maps; The multiple fourth feature maps are fused with the outputs of their corresponding stages to obtain M2, M3, M4 and M5; The fused results are then subjected to convolution to obtain multiple first feature maps: P2, P3, P4, and P5. Multiple first feature maps are input into the anchor box generation network of the trained model to obtain anchor boxes, and the first feature maps are corrected using the anchor boxes to obtain corrected second feature maps. The anchor frame and the second feature map are input into the suggestion proposal network to obtain the suggestion region. The suggestion proposal network includes a binary classification network and a regression network. The binary classification network is used to determine the category to which the anchor frame belongs. The category includes target or background. The regression network corrects the position coordinates of the anchor frame to form the suggestion region. The proposed region is mapped onto the second feature map and an interest region alignment operation is performed to obtain a third feature map of fixed size; The third feature map is input into a fully convolutional network to obtain the category and location of the fuel particles coated in the input image. The fully convolutional network includes a classification branch and a regression branch. The output of the fully convolutional network will not misidentify particles whose coating thickness cannot be measured, and the bounding box is tangent to the four boundaries of the measurable particles. The coating thickness of the fuel particles is determined based on the type and location of the fuel particles in the input image. The anchor box generation network includes an anchor box generation module and a feature adaptation module. The anchor box generation module includes an anchor box position prediction branch and an anchor box shape prediction branch. Inputting the first feature map into the anchor box generation network includes: the anchor box generation module determining whether each feature point on the first feature map is an anchor box center point based on the anchor box position prediction score; the anchor box generation module determining the width and height corresponding to the anchor box center point based on the anchor box shape prediction score, and transforming the width and height corresponding to the anchor box center point to obtain the shape information of each anchor box; and the feature adaptation module performing a deformable convolution operation on the shape information of each anchor box. The deformable convolution operation is performed by the following formula: ， in, The feature value representing the position of the i-th anchor box. This indicates the width and height of the corresponding anchor frame. This represents a deformable convolution operation. Represents the transformed eigenvalues; The process of aligning the proposed region with the region of interest on the second feature map includes: The proposed region is divided into multiple units on an equal basis, and the boundaries of each unit include floating-point numbers; Based on the number of sampling points, each of the multiple units is divided into multiple regions on an average basis, and the position coordinates of each region are determined. Based on the position coordinates of each region, the pixel value corresponding to the position coordinates of each region is obtained by bilinear interpolation. Based on the pixel values ​​corresponding to the position coordinates, a pooling operation is performed on each unit to obtain the pixel values ​​corresponding to each unit, forming a third feature map.

2. The method according to claim 1, further comprising: The result obtained from the convolution operation is downsampled.

3. The method according to claim 1 or 2, wherein, The width and height corresponding to the center point of the anchor frame are converted using the following formula: ， In this branch, the outputs dw and dh of the anchor frame shape prediction branch can be mapped to w and h, respectively, where s is the step size. This is a hyperparameter.

4. The method according to claim 3, wherein, The classification network is the ResNeXt network.

5. The method according to claim 3, wherein, The hyperparameter is 1.

6.

6. A device for detecting the thickness of a coating layer on fuel particles in an image, comprising: The image acquisition unit is adapted to acquire an input image containing a slice of the coated fuel particles, the input image also including coated fuel particles whose coating thickness cannot be measured; The feature extraction unit is adapted to input the input image into the classification network of the trained model, and calculate the output of multiple stages through the convolution kernels in the classification network, including C2, C3, C4 and C5; The output of each stage is upsampled to obtain multiple fourth feature maps; the multiple fourth feature maps are fused with the output of their corresponding stages to obtain M2, M3, M4 and M5; The fused results are then subjected to convolution to obtain multiple first feature maps: P2, P3, P4, and P5. An anchor box generation unit is adapted to input multiple first feature maps into the anchor box generation network of the trained model to obtain anchor boxes, and to modify the first feature maps using the anchor boxes to obtain modified second feature maps. The suggestion region generation unit is adapted to input the anchor box and the second feature map into the suggestion proposal network to obtain a suggestion region. The suggestion proposal network includes a binary classification network and a regression network. The binary classification network is used to determine the category to which the anchor box belongs. The category includes target or background. The regression network corrects the position coordinates of the anchor box to form a suggestion region. An alignment operation unit, adapted to perform a region of interest alignment operation on the mapping of the proposed region onto the second feature map to obtain a third feature map of fixed size, includes: dividing the proposed region into multiple units on an average basis, each unit's boundary including a floating-point number; dividing each of the multiple units into multiple regions on an average basis according to the number of sampling points, and determining the position coordinates of each region; obtaining the pixel value corresponding to the position coordinates of each region using bilinear interpolation based on the position coordinates of each region; and performing a pooling operation on each unit based on the pixel value corresponding to the position coordinates to obtain the pixel value corresponding to each unit, thus forming the third feature map; The particle recognition unit is adapted to input the third feature map into a fully convolutional network to obtain the category and location of the fuel particles coated in the input image. The fully convolutional network includes a classification branch and a regression branch. The output of the fully convolutional network will not misidentify particles whose coating thickness cannot be measured, and the bounding box is tangent to the four boundaries of the measurable particles. The anchor box generation network includes an anchor box generation module and a feature adaptation module. The anchor box generation module includes an anchor box position prediction branch and an anchor box shape prediction branch. Inputting the first feature map into the anchor box generation network includes: the anchor box generation module determining whether each feature point on the first feature map is an anchor box center point based on the anchor box position prediction score; the anchor box generation module determining the width and height corresponding to the anchor box center point based on the anchor box shape prediction score, and transforming the width and height corresponding to the anchor box center point to obtain the shape information of each anchor box; and the feature adaptation module performing a deformable convolution operation on the shape information of each anchor box. The deformable convolution operation is performed by the following formula: ， in, The feature value representing the position of the i-th anchor box. This indicates the width and height of the corresponding anchor frame. This represents a deformable convolution operation. Represents the transformed eigenvalues; The device for detecting the coating thickness of fuel particles in the image is also adapted to determine the coating thickness of the fuel particles based on the type and location of the fuel particles in the input image.

7. A computing device, comprising: At least one processor; as well as A memory storing program instructions, wherein the program instructions are configured to be executed by the at least one processor, the program instructions including instructions for performing the method as claimed in any one of claims 1 to 5.

8. A readable storage medium storing program instructions that, when read and executed by a computing device, cause the computing device to perform the method as described in any one of claims 1 to 5.

Images