Ultrasound image generation system and ultrasound image generation method

Abstract

To provide an ultrasonic image generation technique capable of clearly generating an image indicating an internal state of an object.SOLUTION: An ultrasonic image generation system 1 is provided with one or more computers for generating an image indicating an internal state of an object M from ultrasonic data obtained by an ultrasonic probe 10 for flaw detection of an object K included in the object M to be inspected. The computer is configured to input a plurality of determination data to be determined to a machine learned determination model by using a plurality of learning data and a teacher label 40 in which a specific common region 39 indicating a specific region of the detection object existing in the respective learning data is set as a position of the detection object K, to output a determination value capable of determining whether the detection object K exists for each unit indicating a position in the respective determination data, and to generate an image by using the plurality of determination values.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] Embodiments of the present invention relate to ultrasonic image generation technology. [Background technology]

[0002] Traditionally, ultrasonic testing has been widely used as a means of inspecting structures. Inspectors detect the presence or absence of defects inside the structure from cross-sectional images obtained by ultrasonic testing and measure their location and depth. Methods of displaying the ultrasonic waveform used to determine defects include ultrasonic waveform (A-Scope), cross-sectional image (B-Scope), and top view image (C-Scope).

[0003] A-Scope displays waveforms, allowing for the estimation of the material's thickness and the location of defects (discontinuities) by calculating the position (time) and sound velocity of the waveform echoes. Furthermore, by scanning the ultrasonic probe and acquiring A-Scope data at equal intervals, it is possible to create 2D images from the A-Scope waveform data. For example, scanning the ultrasonic probe one-dimensionally can visualize the cross-sectional structure as a B-Scope. Scanning the ultrasonic probe two-dimensionally can visualize the internal structure of a structure viewed from above. The method of scanning the ultrasonic probe two-dimensionally in a U-shape (or rectangular shape) is called raster scanning.

[0004] B-Scope shows the depth of defects. C-Scope shows the extent and distribution of defects. For example, you can see at a glance how far a crack extends or where blowholes are distributed.

[0005] As shown in Figures 19 to 21, conventional drawing methods reconstruct and draw a single C-Scope image 61 (top view image) from A-Scope (waveform) or multiple B-Scope images 60 (cross-sectional images). Here, a general method for generating a C-Scope image 61 from a B-Scope image 60 is described. First, the array transducer 10 is scanned in the X direction around the object M to be inspected, and multiple B-Scope images 60 of the object M are acquired (Figure 19). For each B-Scope image 60, a predetermined range 62 in the Z direction indicating depth is represented by Z1 to Z2, and a C-Scope image 61 can be obtained (Figure 20). For example, the maximum value of the predetermined range 62 from Z1 to Z2 is plotted to generate a C-Scope image 61. However, if the echo signal from the defect is small compared to the noise intensity (low signal-to-noise ratio), the defect echo 63 and noise 64 cannot be separated, and an unclear C-Scope image 61 is generated. As a result, this negatively affects the evaluation of the defect length and distribution. For example, in the C-Scope image 61, the background B is displayed in black, and the defective echo 63 and noise 64 are displayed as gray to white pixels (Figure 21). In this example, the defective echo 63 is visible in the right region, while the image of the noise 64 is visible in the left region.

[0006] Several methods have been proposed to improve the clarity of the C-Scope. One example involves setting a flaw detection gate and threshold to distinguish between shape echoes, noise, and defect echoes. Then, by generating the C-Scope, the defect echoes are made easier to identify. Another example involves dividing the ultrasonic signal by a reference value to obtain a signal intensity ratio, and if this signal intensity ratio is above a predetermined value, it is reflected in the C-Scope.

[0007] In the first example mentioned above, a threshold is set to distinguish between shape echoes and noise and defect echoes. However, this method may fail to distinguish between very small defects or cracks that are closed, or when the defect echo is weak. Furthermore, the threshold must be determined each time depending on the object being inspected, which is time-consuming and inefficient. Also, in the second example mentioned above, there is a concern that when the defect echo is weak, dividing by the reference value will not result in a high signal intensity ratio. [Prior art documents] [Patent Documents]

[0008] [Patent Document 1] Patent No. 6970425 [Patent Document 2] Patent No. 6026245 [Patent Document 3] Japanese Patent Publication No. 2021-72048 [Overview of the project] [Problems that the invention aims to solve]

[0009] The problem that this invention aims to solve is to provide an ultrasonic image generation technology that can generate clear images showing the internal state of an object. [Means for solving the problem]

[0010] An ultrasonic image generation system according to an embodiment of the present invention comprises one or more computers that generate an image showing the internal state of an object from ultrasonic data obtained by an ultrasonic probe that performs flaw detection on objects contained in an object to be inspected, wherein the computers store a plurality of judgment data which are ultrasonic data arranged along the scanning direction of the ultrasonic probe, and input the plurality of judgment data to be judged into a machine-trained judgment model that uses a plurality of training data which are at least one of the ultrasonic data or data that mimics the same, and a teacher label which sets a specific common region indicating a specific area of ​​the object to be detected that exists in each of the training data as the position of the object to be detected, and outputs a judgment value that can determine whether or not the object to be detected exists for each unit indicating the position in each of the judgment data, multiple The system is configured to extract the position of the detection target from the determined data generated based on the determination value, and to generate the image from a plurality of the determined data arranged along the scanning direction of the ultrasonic probe. The determination data is a cross-sectional image showing a cross-section of the object, the determined data is a determined image of the cross-sectional image, and the image generated using multiple determined data generated based on multiple determination values ​​is a view of the object from a different top direction than the cross-sectional image. ru. [Effects of the Invention]

[0011] Embodiments of the present invention provide an ultrasonic image generation technology that can generate clear images showing the internal state of an object. [Brief explanation of the drawing]

[0012] [Figure 1] A block diagram showing an ultrasound image generation system. [Figure 2] A diagram showing the configuration of an ultrasonic probe. [Figure 3] A top view (plan view) showing an ultrasonic probe being scanned in the X direction. [Figure 4] A cross-sectional view (side view) showing how the ultrasonic probe is scanned in the X direction. [Figure 5] This image shows an ultrasound image acquired by scanning the ultrasound probe in the X direction. [Figure 6] An image diagram showing ultrasound images for learning purposes. [Figure 7]An image diagram showing a teacher label. [Figure 8] An explanatory diagram showing the process of selecting channels from learning ultrasonic images. [Figure 9] A graph showing the received waveform of the echo that detected the detection target. [Figure 10] A graph showing the absolute value of the received waveform of the echo that detected the detection target. [Figure 11] An explanatory diagram showing the pixels of the teacher label. [Figure 12] An image diagram showing an ultrasonic image for determination. [Figure 13] An image diagram showing the determined image. [Figure 14] An explanatory diagram showing the process of generating determined data from an ultrasonic image for determination. [Figure 15] An explanatory diagram showing the process of generating a top surface image from the determined data. [Figure 16] A flowchart showing a method for generating a teacher label. [Figure 17] A flowchart showing a method for generating a determination model. [Figure 18] A flowchart showing a method for generating a top surface image. [Figure 19] An explanatory diagram showing the process of generating a conventional top surface image. [Figure 20] An explanatory diagram showing the process of generating a conventional top surface image. [Figure 21] An explanatory diagram showing the process of generating a top surface image from a conventional ultrasonic image for determination.

Embodiments for Carrying Out the Invention

[0013] Hereinafter, embodiments of an ultrasonic image generation system and an ultrasonic image generation method will be described in detail with reference to the drawings.

[0014] In this embodiment, an example is provided of an ultrasonic testing method generally known as phased array ultrasonic testing. Among these, the following ultrasonic imaging methods can be used: linear scan testing, in which an ultrasonic element driven to form an ultrasonic beam in a constant direction is electronically scanned; sector scan testing, in which the angle at which the ultrasonic beam is formed is changed in a fan shape while the ultrasonic element driven to form the ultrasonic beam is fixed or electronically scanned; Total Focusing Method (TFM), in which the beam is focused by comprehensively setting a focus in an arbitrary coordinate region; or aperture synthesis. In addition, an ultrasonic testing method that scans manually or mechanically using a single probe may also be used. The following description will explain an example using a typical linear scan testing method or sector scan testing method.

[0015] Reference numeral 1 in Figure 1 denotes the ultrasonic image generation system of this embodiment. This ultrasonic image generation system 1 evaluates the presence or absence of defects (detection target K) in the object M to be inspected. In this embodiment, ultrasonic data acquired by a predetermined ultrasonic flaw detection method is automatically analyzed using a computer equipped with artificial intelligence (AI) to generate an image that can evaluate the presence or absence of defects in the object M. For example, the ultrasonic image generation system 1 generates a top view image 50 (C-Scope) showing the object M as seen from above, from a cross-sectional image (B-Scope) showing a cross-section of the object M (Figure 15).

[0016] By evaluating the presence or absence of defects inside object M, this can be used for diagnosing the integrity of object M. In other words, the ultrasound image generation system 1 analyzes object M using artificial intelligence obtained through machine learning.

[0017] In this embodiment, the term "ultrasonic data" refers to the data being processed and includes the meaning of images (2D data, pixel data). For example, the ultrasonic data acquired by the ultrasonic flaw detection method, and the ultrasonic data processed during the determination process, are ultrasonic images. Furthermore, the ultrasonic data may be either still images or moving images. In the following explanation, still image ultrasonic data is used as an example to aid understanding.

[0018] The ultrasonic image generation system 1 comprises an array transducer 10, which is an example of an ultrasonic transducer, an ultrasonic flaw detection device 20, an ultrasonic image processing device 30, and an ultrasonic image evaluation device 100.

[0019] The ultrasonic flaw detection device 20, the ultrasonic image processing device 30, and the ultrasonic image evaluation device 100 are all composed of a computer that has hardware resources such as a CPU (Central Processing Unit), ROM (Read Only Memory), RAM (Random Access Memory), and HDD (Hard Disk Drive), and the CPU executes various programs to realize software-based information processing using the hardware resources. Furthermore, the ultrasonic data evaluation method of this embodiment is realized by having the computer execute various programs.

[0020] Each component of the ultrasound image generation system 1 does not necessarily need to be installed on multiple computers. For example, each component of the ultrasound image generation system 1 may be installed on a single computer.

[0021] As shown in Figure 2, the array transducer 10 emits ultrasonic waves U towards the object M to be inspected and detects the reflected waves reflected by the object K to be detected. In the following explanation, the reflected waves may be referred to as echoes. The array transducer 10 comprises a transducer 11 in which a plurality of ultrasonic elements 11A that emit ultrasonic waves U are arranged. These ultrasonic elements 11A are arranged linearly in a direction perpendicular to the scanning direction of the array transducer 10. The ultrasonic image (ultrasonic data) acquired by this array transducer 10 is an image showing a cross-section of the object M. Although the array transducer 10 is shown as an example of an ultrasonic transducer, it can of course also be applied to ultrasonic transducers arranged in two dimensions or to single probes.

[0022] As an example of the object M, a component made of a metallic material is used. Examples of this metallic material include steel materials such as cast steel products or steel plates, or cast products. The detection target K that occurs in or near the weld W (Figure 3) is evaluated by the ultrasonic image generation system 1. Note that materials other than metallic materials, such as reinforced concrete, mortar, or fiber-reinforced plastics, may also be used as the object M. The detection target K that occurs inside or on the surface of such an object M is evaluated by the ultrasonic image generation system 1.

[0023] The detection target K can be arbitrarily defined by the user of the ultrasonic image generation system 1. The detection target K (defect) can be arbitrarily defined as cracks, scratches, cavities, round holes, delamination, thinning, inclusions, etc., occurring inside or on the surface of the object M. The detection target K can be arbitrarily defined as long as ultrasonic waves are reflected at the boundary between the object M and the detection target K.

[0024] The array probe 10 is controlled by the ultrasonic flaw detection device 20 by applying a voltage to it. This array probe 10 exemplifies a phased array ultrasonic flaw detection method in which the ultrasonic element 11A to be driven is electronically scanned while forming an ultrasonic beam U in a constant direction. Note that this embodiment can be applied to other methods. For example, methods such as sector imaging, TFM, or aperture synthesis may be applied. Furthermore, this embodiment may also be applied to an ultrasonic flaw detection method in which a single probe is used as the ultrasonic probe to perform mechanical scanning.

[0025] During inspection of the object M, an acoustic propagation medium 2, referred to as a wedge, is placed between the array transducer 10 and the object M. This acoustic propagation medium 2 is designed to cause ultrasonic waves U to be incident on the object M at a highly directional angle.

[0026] As the acoustic propagation medium 2, an isotropic material is used that allows ultrasonic waves U to propagate and whose acoustic impedance is known. However, if the surface of the object M is flat, the acoustic propagation medium 2 does not need to be used.

[0027] Examples of isotropic materials used as the acoustic propagation medium 2 include acrylic, polyimide, gel, and other polymers. The acoustic propagation medium 2 can be made of a material with an acoustic impedance close to or the same as that of the front plate (not shown) of the ultrasonic element 11A. Alternatively, a material with an acoustic impedance close to or the same as that of the object M can be used. Furthermore, a composite material that changes the acoustic impedance in a stepwise or gradual manner may be used.

[0028] Furthermore, damping materials may be placed inside and outside the acoustic propagation medium 2 to prevent multiple reflected waves within the acoustic propagation medium 2 from affecting the flaw detection results. A mountain-shaped wave-canceling structure may also be provided. A multiple reflection reduction mechanism may also be provided.

[0029] In the following explanation, the term "acoustic propagation medium 2" may be omitted when describing the process of injecting ultrasonic waves U from the array transducer 10 into the target object M.

[0030] An acoustic coupling medium (not shown) for propagating ultrasonic waves U is used at the contact points in the path from the array transducer 10 to the target object M. For example, when using the acoustic propagation medium 2, the acoustic coupling medium (not shown) is used at the contact points between the array transducer 10 and the acoustic propagation medium 2, and at the contact points between the acoustic propagation medium 2 and the target object M. When the acoustic propagation medium 2 is not used, the acoustic coupling medium (not shown) is used at the contact points between the array transducer 10 and the target object M. This acoustic coupling medium can be any medium capable of propagating ultrasonic waves U, such as water, glycerin, machine oil, castor oil, acrylic, polystyrene, or gel.

[0031] As shown in Figure 1, the detection signal containing information about the reflected waves detected by the array transducer 10 is input to the ultrasonic flaw detection device 20. The detection signal obtained by the ultrasonic flaw detection device 20 is then processed by the ultrasonic image processing device 30.

[0032] The ultrasonic image processing device 30 acquires ultrasonic images for flaw detection of the target object K contained in the object M using ultrasonic waves U. This ultrasonic image processing device 30 can generate ultrasonic images along the scanning direction of the ultrasonic element 11A based on the reflected waves of ultrasonic waves U.

[0033] Next, a method for scanning the array probe 10 in the X direction will be described with reference to Figures 3 to 5. For example, suppose that a linearly extending weld W is provided on the object M (Figures 3 to 4). The detection target K is assumed to be a defect such as a crack that has occurred along and near this weld W.

[0034] As shown in Figures 3 and 4, the array probe 10 moves along the upper surface of the object M. For example, when the array probe 10 is scanned along the surface of the object M and in the direction in which the weld W extends, in the X direction (linearly), an ultrasonic image (cross-sectional image) showing the cross-section of the object M at each position is obtained. This ultrasonic image is also called a B-Scope. For example, at each position (X n ~X n+4 ) corresponds to each ultrasound image (X n~X n+4 ) is obtained (Fig. 5). These ultrasonic images (X n ~X n+4 ) are the learning ultrasonic image 31 or the determination ultrasonic image 35. This scan is called a lateral scan. This scan is used to estimate the length of the detection target K in the scan direction.

[0035] As shown in Fig. 5, for example, in the first ultrasonic image (X n ), the appearance image (echo) of the detection target K is not displayed. However, in the next ultrasonic image (X n+1 ) and the next ultrasonic image (X n+2 ), the appearance image of the detection target K gradually becomes larger. Furthermore, when approaching the end of the detection target K, in the ultrasonic image (X n+4 ), the display of the appearance image becomes weaker. This is a characteristic intensity change of the detection target K.

[0036] Fig. 6 illustrates an ultrasonic image (ultrasonic data) as a typical linear scan flaw detection image. This ultrasonic image is used as the learning ultrasonic image 31 as the learning data of the present embodiment. In this learning ultrasonic image 31, the background B is displayed in black, the incident range 32 where the ultrasonic wave U is incident on the object M is displayed in gray, the surface 33 of the object M is displayed as a white line, and the detection target K is displayed in white. Regarding the gray scale, an ultrasonic image with the scale reversed, the background being white and the surface 33 of the object M being black, may also be used. Also, although the ultrasonic image can be a color image represented by RGB or the like, for clarity of explanation, it will be described as a gray scale image. In the example of Fig. 6, one detection target K is shown in the learning ultrasonic image 31. The ultrasonic image is also used as the determination ultrasonic image 35 (Fig. 12) as the determination data of the present embodiment.

[0037] As shown in Figure 2, the ultrasonic waves U incident on the object M from the array transducer 10 have their incident and refraction angles determined according to the delay time, and propagate inside the object M. These propagated ultrasonic waves U are reflected or scattered by the detection target K, etc., and reach the array transducer 10 again as reflected waves. In addition to the detection target K, the ultrasonic waves U are also reflected or scattered by the bottom surface or corners of the object M and reach the array transducer 10.

[0038] The reflected waves, which are scattered waves that reach the array transducer 10, are visualized based on the arrival time and information from the ultrasonic element 11A. At this time, if a target object K is present in the object M, the echo (developed / removed) of the target object K appears in the learning ultrasonic image 31.

[0039] Here, the ultrasound image processing device 30 generates a training ultrasound image 31 (Figure 6) or a judgment ultrasound image 35 (Figure 12). The training ultrasound image 31 or the judgment ultrasound image 35 is input from the ultrasound image processing device 30 to the ultrasound image evaluation device 100.

[0040] The learning ultrasound image 31 may also be a computer graphics (CG) image generated using computer graphics (CG) to mimic the ultrasound image obtained by the ultrasound flaw detection device 20. Alternatively, a wide variety of learning ultrasound images 31 may be generated by manually or automatically editing the ultrasound image obtained by the ultrasound flaw detection device 20. As for editing methods, for example, the ultrasound image may be trimmed to a certain extent or divided into several parts. The ultrasound image may also be pre-processed with filters such as image processing filters.

[0041] Furthermore, ultrasonic images obtained by numerical simulations of ultrasonic wave propagation may be used. For example, an ultrasonic image may be generated using a simulation technique that visualizes the propagation of ultrasonic waves U inside an object M, and this image may be used as the training ultrasonic image 31.

[0042] The ultrasound image evaluation device 100 performs machine learning using the training ultrasound image 31, and based on the results of this machine learning, identifies the position of the target K included in the judgment ultrasound image 35 used for judgment. The ultrasound image evaluation device 100 generates and displays a top view image 50 (C-Scope) using the result of identifying the position of the target K.

[0043] Computer-based analysis can utilize analytical techniques based on artificial intelligence learning. For example, trained models generated by machine learning using neural networks, trained models generated by other machine learning methods, deep learning algorithms, and mathematical algorithms such as regression analysis can be used.

[0044] The ultrasound image generation system 1 includes a computer equipped with artificial intelligence for machine learning. For example, the system may consist of one computer equipped with a neural network, or it may consist of multiple computers equipped with neural networks.

[0045] Here, a neural network is a mathematical model that represents the characteristics of brain function through computer simulation. For example, it shows a model in which artificial neurons (nodes) that form a network through synaptic connections change the strength of their synaptic connections through learning and acquire problem-solving abilities. Furthermore, neural networks acquire problem-solving abilities through deep learning.

[0046] Furthermore, a reward function may be set for each information item to be learned, and deep reinforcement learning, in which the information item with the highest value is extracted based on the reward function, may be used in the neural network.

[0047] In this embodiment, supervised machine learning methods are used. For example, semantic segmentation is used. However, it is not limited to this, and multiple types of machine learning methods may be combined, or any single machine learning method may be adopted. For example, supervised learning that handles time series data, such as a recurrent neural network or a recurrent neural network with a long-short-term memory structure, may be used.

[0048] Other machine learning methods besides semantic segmentation include fully convolutional neural networks, R-CNN (Region Convolutional Neural Network), which has a proven track record in image recognition, ViT (Vision Transformer), k-nearest neighbors, logistic regression, support vector machines (SVM), and deep learning.

[0049] Deep learning encompasses various techniques such as autoencoders, RNNs (Recurrent Neural Networks), LSTMs (Long Short-Term Memory), and GANs (Generative Adversarial Networks). These techniques may be applied, or other machine learning methods may be used.

[0050] As shown in Figure 12, the ultrasonic image 35 for determination is displayed with a black background B, the incident range 32 where the ultrasonic waves U are incident on the object M is displayed in gray, the surface 33 of the object M is displayed as a white line, and the detected object K is displayed in white. In the example in Figure 12, two detected objects K are visible in the ultrasonic image 35 for determination. Then, using the determination model, a determined image 36 (Figure 13) is generated from this ultrasonic image 35 as determined data.

[0051] As shown in Figure 13, in the determined image 36, the background B is displayed in black, and pixels are displayed brighter in areas 37 where there is a high probability that the edges of the detected object K are present. In other words, pixels are displayed in a gray to white color. In the example in Figure 13, areas 37 where there is a high probability that two edges are present are captured in the determined image 36.

[0052] Next, the system configuration of the ultrasound imaging evaluation device 100 will be described with reference to the block diagram shown in Figure 1. Furthermore, the ultrasound imaging evaluation device 100 may include components other than those shown in Figure 1, and some of the components shown in Figure 1 may be omitted.

[0053] The ultrasound image evaluation device 100 comprises an input unit 110, a calculation unit 120, a judgment model generation unit 130, a storage unit 140, and an output unit 150.

[0054] The input unit 110 receives an ultrasonic image obtained by incidentating an ultrasonic wave U onto an object M from an external source. In this embodiment, an example is shown in which the ultrasonic image is input from an ultrasonic image processing device 30, but the ultrasonic image may be input from other devices.

[0055] Furthermore, the input unit 110 receives flaw detection position data indicating the position of the array probe 10. This flaw detection position data includes the position (displacement) and rotation angle of the array probe 10 when it scans the object M.

[0056] The flaw detection position data includes at least two coordinate values ​​represented by Cartesian coordinates that indicate the position of the array probe 10. This makes it easier to represent the position of the array probe 10 on the surface of the object M when the object M is flat. The Cartesian coordinates are, for example, a two-dimensional coordinate system along the surface of the flat object M.

[0057] Furthermore, the flaw detection position data may include at least two coordinate values, expressed in polar or cylindrical coordinates, that represent the position of the array probe 10. This makes it easier to represent the position of the array probe 10 on the surface of the object M when the object M is spherical, cylindrical, or cylindrical in shape. Polar coordinates are, for example, a two-dimensional coordinate system along the surface of the object M, which is spherical, cylindrical, or cylindrical in shape. For example, the flaw detection position data may include the radial and angular values ​​expressed in polar coordinates. Alternatively, the flaw detection position data may include the angular position and axial position expressed in cylindrical coordinates.

[0058] In this embodiment, a flat object M is exemplified. The two directions along the surface of this object M are represented by the X and Y axes, and the depth direction of the object M is represented by the Z axis (Figures 3 to 4).

[0059] The ultrasound images input to the input unit 110 consist of two types: a training ultrasound image 31 (Figure 6) used for machine learning and a judgment ultrasound image 35 (Figure 12). This input unit 110 is the ultrasound data acquisition unit in this embodiment.

[0060] In this embodiment, when machine learning is performed, multiple training ultrasound images 31 are input to the input unit 110. Then, a group of teacher labels is generated based on this group of training ultrasound images. Furthermore, a judgment model is generated using the group of training ultrasound images and the group of teacher labels. On the other hand, when the detection target K contained in the judgment ultrasound image 35 is determined using this judgment model, the judgment ultrasound image 35 is input to the input unit 110. Note that the input judgment ultrasound image 35 may be one image or multiple images.

[0061] A machine learning-based decision model comprises an input layer that receives ultrasound data, an output layer that outputs a decision value, and an intermediate layer whose parameters are learned using training data (training data) that includes ultrasound data and training labels. This decision model causes the computer to function by inputting ultrasound data into the input layer, performing calculations in the intermediate layer, and outputting a decision value from the output layer.

[0062] For training data, at least one of the following is used: ultrasonic data obtained using the array transducer 10 or ultrasonic data that mimics it.

[0063] The calculation unit 120 includes a teacher label generation unit 121, a determination unit 122, an extraction unit 123, and an image generation unit 124. These are realized by the CPU executing a program stored in memory or on the HDD.

[0064] The memory unit 140 stores ultrasonic data (images), inspection location data, judgment models, and judged data (judgment values) obtained by ultrasonic testing. This memory unit 140 comprises a learning ultrasonic image storage unit 141, a teacher label storage unit 142, a judgment model storage unit 143, and a judgment ultrasonic image storage unit 144. These are collections of information that are stored in memory, an HDD, or the cloud and organized so that they can be searched or accumulated.

[0065] The output unit 150 outputs predetermined information. This output unit 150 can be any device capable of displaying digital data, such as a PC display, television, projector, or head-mounted display. Alternatively, it may be a device that first converts the signal to analog before displaying the image on its screen, like a cathode ray tube. In other words, the ultrasound image evaluation device 100 includes a device that displays the image outputting the analysis results. Furthermore, the output unit 150 may be a printer that prints predetermined information onto paper. Alternatively, the output unit 150 may write predetermined information to a removable and portable storage medium such as a USB memory stick.

[0066] The output unit 150 may also display the synthesized ultrasonic echo signal, the image result, the coordinates of the array transducer 10 and its relative position to the detection target K, the delay time, the focal depth, and the flaw detection refraction angle, among other flaw detection conditions. Furthermore, the output unit 150 may have a user interface function that generates an alarm by sound or light emission depending on the set conditions, or allows operation input as a touch panel.

[0067] The ultrasound image evaluation device 100 may also have other components. For example, a communication unit (not shown) may be provided. This communication unit communicates with other computers, for example, via a communication line such as the Internet.

[0068] Each component of the ultrasound imaging evaluation device 100 does not necessarily have to be installed on a single computer. For example, a single ultrasound imaging evaluation device 100 may be realized using multiple computers connected to each other via a network.

[0069] The teacher label generation unit 121 generates teacher labels 40 in which a specific common region 39 (Figure 7) indicating a specific area of ​​the target K present in each training ultrasound image 31 is set as the position of the target K. One teacher label 40 is generated corresponding to one training ultrasound image 31. Then, a group of teacher labels corresponding to a group of training ultrasound images is generated.

[0070] As shown in Figure 7, the teacher label 40 is an image (2D data) of the same size as the corresponding training ultrasound image 31. In other words, the teacher label 40 is an image with the same number of pixels as the training ultrasound image 31. In this teacher label 40, the background B is displayed in black, and the specific common region 39 of the detection target K is displayed in white. In other words, the teacher label 40 is a binarized image constructed by setting the pixel values ​​of the specific common region 39 to "1" and the pixel values ​​of all other pixels to "0". The position (coordinates) of the specific common region 39 in this teacher label 40 corresponds to the position of the upper end of the detection target K as seen in the training ultrasound image 31.

[0071] The teacher label 40 is configured with a specific common region 39 corresponding to the area where the echo of the target K is located. The specific common region 39 can be arbitrarily set according to the type of target K, for example, the crack end in the case of a crack, the center in the case of a blowhole. This setting may be done by the user or automatically by the ultrasound image evaluation device 100.

[0072] If the object K being detected is a crack, the end of the echo of the object K is defined as a specific common region 39. Then, the specific common region 39 in the learning ultrasound image 31 corresponds to the position of the crack end. The ultrasound image evaluation device 100 then detects the specific common region 39 from the learning ultrasound image 31. Furthermore, the depth of the crack can be determined by calculating the distance between the specific common region 39 and the echo on the surface or bottom surface of the object M.

[0073] Next, the method for generating the teacher label 40 will be explained using the flowchart in Figure 16. Note that the detection target K will be defined as a crack, and its end will be described as a specific common region 39.

[0074] First, in step S11, the ultrasound image processing device 30 acquires multiple training ultrasound images 31. These training ultrasound images 31 are input to the input unit 110 of the ultrasound image evaluation device 100. The multiple training ultrasound images 31 input to the input unit 110 are stored in the training ultrasound image storage unit 141.

[0075] In the next step, S12, the user, such as an inspector, researcher, or expert, reviews the learning ultrasound images 31. The user examines each learning ultrasound image 31 to determine whether or not cracks are visible.

[0076] In the next step S13, the teacher label generation unit 121 selects a channel for the array transducer 10 if a crack is visible in the training ultrasound image 31. For example, among the multiple (e.g., N) ultrasound elements 11A of the array transducer 10, a received waveform 41 (Figure 9) containing an echo from the edge of the crack is identified. Then, the channel of the array transducer 10 that received this received waveform 41 is selected. Here, the channel number is denoted as m (Figure 8).

[0077] In the next step S14, the teacher label generation unit 121 acquires an echo of the crack edge from the received waveform 41 of channel number m and determines the beam path length Xd at the peak position of this echo. The selection of channel number m may be performed by the teacher label generation unit 121 or by the user.

[0078] For example, as shown in Figure 9, the received waveform 41 for channel number m includes the waveform 42 of the echo from the surface 33 of the object M and the waveform 44 of the echo at the edge of the crack. The amplitude value of the received waveform 41 for channel number m is denoted as Am(X). Furthermore, Figure 10 shows the waveform 45 obtained by obtaining the absolute value of the received waveform 41.

[0079] Here, the teacher label generation unit 121 calculates the peak value of the echo at the edge of the crack, |Am(Xd)|, from |Am(X)|, which is the absolute value of Am(X). Then, the teacher label generation unit 121 can determine the beam path Xd to this position (Figure 8).

[0080] Returning to Figure 16, in the next step S15, the teacher label generation unit 121 determines the coordinates Ex and Ey of the center of the specific common region 39 of the teacher label 40 from the channel number m, beam path length Xd, and refraction angles α and β (Figure 8).

[0081] In the next step S16, the teacher label generation unit 121 sets up a square with side length d centered at coordinates Ex,Ey (Figure 11). Here, a binarized image is generated in which the square is set to "1" and everything else to "0". This binarized image becomes the teacher label 40 (Figure 7).

[0082] In the next step S17, the teacher label generation unit 121 stores the generated teacher labels 40 in the teacher label storage unit 142. Then, the method for generating the teacher labels 40 is completed.

[0083] As shown in Figure 11, the size (maximum length) of the specific common region 39 in this embodiment is represented by d. For example, it is a d × d square. Here, the diagonal of the specific common region 39 is at least one-quarter of the wavelength λ of the echo at the edge of the crack. For example, the following equation holds true.

[0084]

number

[0085] As shown in Figure 9, a length of one-quarter of the wavelength λ corresponds to the length from the rise of the waveform to the peak, and is the minimum length that includes the echo characteristics from the crack edge. If the size (maximum length) of the specific common region 39 is less than one-quarter of the wavelength λ, the specific common region 39 may not contain sufficient echo characteristics, and there is a possibility that the crack edge cannot be correctly detected.

[0086] Even in the case of an echo consisting of several consecutive waves, it is necessary to extract a representative waveform from those waves and determine its wavelength.

[0087] If the specific common region 39 is a circle, the diameter of the circle shall be set to a size of at least one-quarter of the wavelength λ. If the specific common region 39 is an ellipse, the major axis of the ellipse shall be set to at least one-quarter of the wavelength λ. In addition, the specific common region 39 can be set to a polygon or an irregular shape other than a circle. In that case, the maximum length of each shape shall be set to at least one-quarter of the wavelength λ.

[0088] By ensuring that the size d of the specific common area 39 is at least one-quarter of the wavelength of the ultrasonic wave U used by the ultrasonic flaw detection device 20, the characteristics of the target K are sufficiently included in the specific common area 39, enabling correct detection of the target K. Furthermore, the target K can be distinguished from noise.

[0089] As shown in Figure 1, the judgment model generation unit 130 uses the training ultrasound image 31 and the teacher label 40 to generate a judgment model that can calculate a judgment value for each unit indicating the position in the image, which can determine whether or not the target K is present. In other words, the judgment value is generated from the judgment model.

[0090] Here, the unit used to indicate position within an image is, for example, a pixel. The position (coordinates) of a given pixel within an image is managed using conventionally known methods.

[0091] In this embodiment, the unit indicating position in the image is a pixel, but other configurations are also possible. For example, multiple adjacent pixels may be considered as one unit. Alternatively, the image may be divided into multiple grids, and each cell in the grid may be considered as one unit.

[0092] The judgment model generation unit 130 performs supervised learning based on a set of training ultrasound images (a set of ultrasound data) and a corresponding set of teacher labels. By performing machine learning, this judgment model can determine whether a specific common region 39 of the target K is included in the ultrasound image and output the determination value for each pixel.

[0093] In supervised learning, in addition to 2D data (images) used as training data, flaw detection position data indicating the position and angle of the array probe 10 is used as training data. Therefore, it becomes possible to newly learn the defect-specific trend of gradually changing brightness values ​​when the detection target K is continuous. The input and output data used for this learning can be input and output one by one, or multiple data can be input and output together.

[0094] Through this supervised learning, a decision model is generated based on a training ultrasound group tn consisting of n training ultrasound images 31 and a corresponding group of teacher labels ln. This decision model can accept a group of decision ultrasound images Vk consisting of k decision ultrasound images 35 as input data and output a group of decision images Uk consisting of k decision images 36 as output data (Figure 14).

[0095] Next, we will explain the method for generating the judgment model using the flowchart in Figure 17.

[0096] First, in step S21, the judgment model generation unit 130 acquires multiple training ultrasound images 31 (training data) that have been input to the input unit 110 (image input unit 111) and stored in the training ultrasound image storage unit 141.

[0097] In the next step S22, the judgment model generation unit 130 performs supervised learning based on multiple training ultrasound images 31 and the teacher labels 40 corresponding to each training ultrasound image 31.

[0098] In the next step S23, the judgment model generation unit 130 generates a judgment model through supervised learning. By performing supervised learning, a judgment model is generated that has the ability to determine whether a specific common region 39 of the detection target K is included in the image and to output the determination value for each pixel. With this judgment model, when a judgment ultrasound image 35 (Figure 12) is accepted as an input value, the determination value, i.e., the judged image 36 (Figure 13), can be output as an output value.

[0099] Furthermore, for the generation of a judgment model (supervised learning), it is desirable to prepare as many training ultrasound images 31 as possible and use them to generate the judgment model. Of the training ultrasound images 31, it is desirable that at least two images contain the target object K. In addition, the training ultrasound images 31 may include images in which the target object K is not present, but only the echo of the surface 33 of the object M is present.

[0100] In the next step S24, the judgment model generation unit 130 stores the generated judgment model in the judgment model storage unit 143. Then, the method for generating the judgment model ends.

[0101] The determination unit 122 (Figure 1) receives multiple determination data to be determined from a machine learning-based determination model and outputs a determination value that can determine whether or not the target object K exists for each unit indicating the position in each determination data. Here, the determination data is a determination ultrasound image 35. This determination ultrasound image 35 is a cross-sectional image (B-Scope) showing a cross-section of the object M. The determination unit 122 uses the determination model to calculate a determination value for each pixel of the determination ultrasound image 35 indicating whether or not the target object K exists. A determined image 36 is generated from these determination values ​​(Figure 14).

[0102] The extraction unit 123 extracts the position of the target K from the determined image 36, which is determined data generated based on the judgment value.

[0103] The image generation unit 124 generates an image showing the internal state of the object M from the ultrasonic data obtained by the array transducer 10. This image generation unit 124 generates images using multiple judgment values. For example, the image generation unit 124 generates a top view image 50 (C-Scope) showing the state of the object M as seen from above from multiple judged images 36 composed of judgment values ​​(Figure 15).

[0104] Furthermore, the image generated using the judgment value is a view of the object M from a different direction than the judgment ultrasound image 35, which is a cross-sectional image. In other words, the image generated using the judgment value is not only a top view image 50 viewed from a direction perpendicular to the surface of the object M, but may also be an image viewed from an oblique direction to the surface of the object M, or an image viewed from the back of the object M. In this way, the user can understand the internal state of the object M.

[0105] Next, the method for generating the top image 50 will be explained using the flowchart in Figure 18.

[0106] First, in step S31, the ultrasound image processing device 30 acquires multiple ultrasound images 35 (judgment data) for determination. These multiple ultrasound images 35 for determination are input to the input unit 110 of the ultrasound image evaluation device 100. The ultrasound images 35 for determination input to the input unit 110 are stored in the ultrasound image storage unit 144 for determination.

[0107] Here, multiple ultrasound images 35 for judgment are arranged along the scanning direction of the array transducer 10. For example, the first to kth ultrasound images 35 for judgment are arranged to form a set of ultrasound images Vk for judgment (Figure 14).

[0108] In the next step S32, the determination unit 122 inputs multiple ultrasound images 35 stored in the ultrasound image storage unit 144 into a machine learning-based determination model and outputs (generates) multiple determined images 36 (determination values).

[0109] As shown in Figure 13, the determined image 36 is an image (2D data) of the same size as the ultrasound image 35 used for determination. Each pixel of the determined image 36 is represented by a continuous value from "0" to "1" as a determination value. These determination values ​​are continuous values ​​that indicate the degree to which the target K is present.

[0110] In the determined image 36, pixels are displayed brighter in areas 37 where there is a high probability that the edge of the detected target K is located. For example, if the judgment value is "0", the pixel value will be black. If the judgment value is "1", the pixel value will be white. If the judgment value is greater than "0" and less than "1", the pixel value will be gray. In other words, in the determined image 36, the closer the judgment value is to "1", the higher the probability that a specific common area 39 (Figure 7) is located.

[0111] This determination allows for the extraction of only the desired specific common region 39 (Figure 7) from the ultrasound image 35 used for determination. Here, echoes from the ends of cracks included in the ultrasound image are set as the specific common region 39, a teacher label 40 indicating these echoes from the ends of cracks is created, and supervised learning is performed.

[0112] The supervised learning judgment model obtained in this way can output echoes from the crack edges as a specific common region 39, even when a judgment ultrasound image 35 not used in training is input. This output is represented as image data of the judgment value.

[0113] Here, the judgment value will be close to "1" in areas where the echo is likely to be from the edge of a crack. The judgment value will be "0" in areas where no crack exists. Furthermore, although the judgment value is imaged as an echo, it will be close to "0" for echoes from the surface or bottom, echoes from inclusions other than cracks, or other types of echoes.

[0114] As shown in Figure 14, the determination unit 122 generates a set of determined images Uk from the set of ultrasound images Vk for determination. For example, the first to kth determined images 36 are arranged along the scanning direction of the array transducer 10, and the set of determined images Uk is generated. The number of images in the set of ultrasound images Vk for determination and the set of determined images Uk are the same.

[0115] Returning to Figure 18, in the next step S33, the extraction unit 123 extracts the position of the target K from the determined image 36 (Figure 13). The extraction unit 123 performs a binarization process on the determined image 36 to generate a binarized image (not shown) from which a specific common region 39 has been extracted.

[0116] Here, the threshold for the binarization process is set in advance by the user. If the value of each pixel in the determined image 36 exceeds the threshold, it is replaced with the white value "1," and if it falls below the threshold, it is replaced with the black value "0." In this way, an image of the specific common region 39 (Figure 7) can be made to appear in the binarized image. Then, the position and depth of the detection target K can be determined from this binarized image.

[0117] The position of the surface 33 or bottom surface of the object M is assumed to be determined in advance by a predetermined method. For example, the method for calculating the coordinates of the surface 33 or bottom surface of the object M may be to use the maximum value of each echo as the coordinate, or to use the center of echoes that exceed a predetermined threshold.

[0118] In the next step S34, the image generation unit 124 generates the top image 50 (C-Scope).

[0119] For example, as shown in Figure 15, the image generation unit 124 generates a top image 50 from the group of determined images Uk. Here, it is assumed that the first to k determined images 36 are arranged in the X direction, which is the scanning direction of the array probe 10. These k determined images 36 are combined to generate a top image 50 of the object M (Figures 3 to 4) as seen from above (Z direction).

[0120] In the top image 50, the background B is displayed in black, and the detected target K echo 51 (developed) is displayed as gray to white pixels. In this top image 50, the image of the echo 51 appears based on the judgment value. Therefore, no noise image appears, and a clear top image 50 is generated.

[0121] The judgment value is a continuous value indicating the degree to which the detected target K is present, and each judgment value corresponds to the color of the echo 51. In this way, a clear top image 50 can be generated using continuous values ​​indicating the degree of presence. For example, areas where the detected target K is present are displayed in bright white, and areas where the detected target K is not present are displayed in black.

[0122] By generating this top view image 50, the position of the detected object K, the direction in which it extends, and its width become easier to understand. Therefore, it becomes easier for the user to evaluate the state of the detected object K.

[0123] Returning to Figure 18, in the next step S35, the image generation unit 124 stores the generated top image 50 (C-Scope) in the storage unit 140 and outputs (displays) it using the output unit 150. Then, the method for generating the top image 50 is completed.

[0124] In this embodiment, when the top image 50 is generated, the maximum amplitude value in the Z direction is projected onto the position in the Y direction for each determined image 36 (determined value). For example, the top image 50 (C-Scope) can be obtained using the following formula.

[0125]

number

[0126] Here, in an image represented by (y,z) coordinates, a set of k images is represented as the judgment value Uk(y,z). Furthermore, the top image 50 (C-Scope) obtained from the k judgment values ​​Uk(y,z) is represented as C(k,y).

[0127] As shown in Figures 3 and 4, the shorter the acquisition interval for ultrasound images and the greater the number of images taken, the clearer the distribution of cracks inside the object M can be visualized. Figure 15 shows a plot of the top image 50 (C-Scope) obtained here.

[0128] On the other hand, Figures 19 to 21 show how to obtain the general C-Scope image 61 used in conventional inspections. The array probe 10 is scanned in the X direction around the object M to obtain a B-Scope image 60 showing a cross-section of the object M. Then, for each B-Scope image 60, if the range in the Z direction indicating depth is represented as Z1 to Z2, the C-Scope image 61 can be obtained using the following formula.

[0129]

number

[0130] Generally, there is a method of specifying a predetermined range 62 (Figure 20) from Z1 to Z2, called a gate, and taking the maximum value within this predetermined range 62. By setting this gate, it is possible to exclude high-intensity surface echoes or bottom echoes from the predetermined range 62 and take the maximum value. In this way, it is possible to avoid the surface echoes or bottom echoes appearing in the C-Scope image 61.

[0131] However, the C-Scope image 61 obtained using conventional methods plots background noise, such as inclusions or electrical noise inside the object M, as the maximum value. Furthermore, if the intensity of the defect echo itself is low, both the defect echo and the background noise appear in the C-Scope image 61, making differentiation difficult. In other words, with conventional methods, if the echo signal from the defect is small relative to the noise intensity (low signal-to-noise ratio), the defect echo and noise cannot be separated, resulting in the generation of an unclear C-Scope image.

[0132] In contrast, in this embodiment, machine learning is used to extract the echo signal of defects from the judgment ultrasound image 35 (B-Scope), and the top image 50 (C-Scope) is generated using the extraction result (judgment value). As a result, defect echoes and noise are completely separated, and a top image 50 is generated in which only defect echoes are visible. By using this top image 50 for evaluation of non-destructive testing, the location and distribution of defects can be identified more reliably. Therefore, it can be used for diagnosing the soundness of the object M, etc.

[0133] Furthermore, instead of labeling the entire echo of the target K, only the edges of the target K are labeled. This significantly reduces the effort required to create the teacher label 40. Also, because only the edges of the target K are detected, even if the shape of the target K differs in each ultrasound image, the edges (specific common region 39) that are commonly needed for measuring its depth can be reliably detected.

[0134] Furthermore, since the specific common region 39 is the edge of the detection target K, the accuracy of determining the depth at which the detection target K exists can be improved.

[0135] Furthermore, the determined image 36 (Figure 13) is generated based on the determination value calculated from the judgment ultrasound image 35 (Figure 12). The extraction unit 123 extracts the position of the target K from this determined image 36 (Figure 13), thereby extracting the position of the target K from the determined image 36 and detecting the target K present in the judgment ultrasound image 35.

[0136] Furthermore, the extraction unit 123 extracts the location of the detection target K by binarizing the determination value for each unit indicating the position in the image based on a threshold, thereby extracting the portion of the determined image 36 that is highly likely to contain the detection target K.

[0137] Furthermore, by using pixels as the unit in this embodiment, processing can be performed for each pixel, which is the smallest unit of the image, allowing for detailed determination of the ultrasonic image 35 for judgment.

[0138] Furthermore, by calculating the position of the detection target K in the image calculated by the judgment model, a 3D display representing the position of the detection target K becomes possible, making it easier to understand the deterioration status of the object M being inspected. In addition, the evaluation results can be used to propose repairs after inspection.

[0139] In the above-described embodiment, the determination of an arbitrary value (pixel value, judgment value) using a reference value (threshold) may be a determination of "whether the arbitrary value is greater than or equal to the reference value", a determination of "whether the arbitrary value exceeds the reference value", a determination of "whether the arbitrary value is less than or equal to the reference value", or a determination of "whether the arbitrary value is less than or equal to the reference value". Furthermore, the reference value is not fixed but may change. Accordingly, a predetermined range of values ​​may be used instead of a reference value, and a determination may be made as to whether the arbitrary value falls within the predetermined range. In addition, errors occurring in the device may be analyzed in advance, and a predetermined range including the error range centered on the reference value may be used for the determination.

[0140] Although the flowchart of the above-described embodiment illustrates a configuration in which each step is executed in series, the order of each step is not necessarily fixed, and the order of some steps may be reversed. Also, some steps may be executed in parallel with other steps.

[0141] The system of the aforementioned embodiment comprises a control device with highly integrated processors such as an FPGA (Field Programmable Gate Array), GPU (Graphics Processing Unit), CPU, and dedicated chip; storage devices such as ROM and RAM; external storage devices such as HDD and SSD (Solid State Drive); display devices such as a display; input devices such as a mouse and keyboard; and a communication interface. This system can be realized with a hardware configuration using a standard computer.

[0142] Furthermore, the program or learned model to be executed in the system of the aforementioned embodiment is provided pre-installed in ROM or the like. Additionally or alternatively, this program or learned model is provided as an installable or executable file stored on a computer-readable non-temporary storage medium such as a CD-ROM, CD-R, memory card, DVD, or flexible disk (FD).

[0143] Furthermore, the programs or trained models executed by this system may be stored on a computer connected to a network such as the Internet and provided for download via the network. Alternatively, this system can be configured by combining separate modules, each independently performing its respective function, which are interconnected via a network or dedicated line.

[0144] Furthermore, although this embodiment describes the ultrasound image as a grayscale image, other embodiments are also possible. For example, the ultrasound image may be a color image represented by RGB or the like. In that case, for example, pixel values ​​above a threshold may be represented in red, and pixel values ​​below the threshold may be represented in blue, giving the display a distinctive appearance.

[0145] Alternatively, the array transducer 10 may be composed of a matrix array probe, and the ultrasonic data of the object M may be acquired as voxel data. In this case, the specific common area 39 (Figure 7) may be a cube.

[0146] According to the embodiment described above, by generating an image using multiple judgment values, it is possible to generate a clear image showing the internal state of the object M.

[0147] While several embodiments of the present invention have been described, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These embodiments can be carried out in a variety of other forms, and various omissions, substitutions, modifications, and combinations are possible without departing from the spirit of the invention. These embodiments or their variations are included in the scope and spirit of the invention, as well as in the claims and their equivalents. [Explanation of symbols]

[0148] 1…Ultrasonic image generation system, 2…Acoustic propagation medium, 10…Array probe, 11…Probe, 11A…Ultrasonic element, 20…Ultrasonic flaw detection device, 30…Ultrasonic image processing device, 31…Ultrasonic image for learning, 32…Incident range, 33…Surface of the object, 35…Ultrasonic image for judgment, 36…Judged image, 37…Region where an edge is likely to exist, 39…Specific common region, 40…Teacher label, 41…Received waveform, 42…Waveform of echo on the surface of the object, 44…Waveform of echo at the edge of the crack, 45…Absolute value waveform, 50…Top image, 51…Echo, 60…B-S Cope image, 61...C-Scope image, 62...Determined range, 63...Defect echo, 64...Noise, 100...Ultrasonic image evaluation device, 110...Input unit, 120...Calculation unit, 121...Teacher label generation unit, 122...Determination unit, 123...Extraction unit, 124...Image generation unit, 130...Determination model generation unit, 140...Storage unit, 141...Learning ultrasound image storage unit, 142...Teacher label storage unit, 143...Determination model storage unit, 144...Determination ultrasound image storage unit, 150...Output unit, B...Background, K...Detection target, M...Object, R...Reflected wave, U...Ultrasonic, W...Welded area.

Claims

1. The system comprises one or more computers that generate an image showing the internal state of an object from ultrasonic data obtained by an ultrasonic probe used to detect defects contained in the object being inspected. The aforementioned computer, The system stores multiple judgment data, which are ultrasonic data, arranged along the scanning direction of the ultrasonic probe. A machine learning-trained judgment model is input with multiple training data sets, which are at least one of the aforementioned ultrasonic data or data that mimics it, and a teacher label in which a specific common region indicating a particular region of the target to be detected that exists in each of the training data sets is set as the position of the target to be detected. The model is then made to output a judgment value that can determine whether or not the target to be detected exists for each unit indicating the position in each of the judgment data sets. The location of the target to be detected is extracted from the determined data generated based on the multiple determination values. The image is generated from a plurality of the determined data arranged along the scanning direction of the ultrasonic probe. It is configured in such a way, The data used for determination is a cross-sectional image showing a cross-section of the object, The determined data is the determined image of the cross-sectional image, The image generated using multiple determined data based on multiple determination values ​​represents the view of the object from a top direction different from the cross-sectional image. Ultrasound image generation system.

2. The maximum length of the specified common area is one-quarter or more of the wavelength of the ultrasonic frequency used by the ultrasonic probe. The ultrasonic image generation system according to claim 1.

3. The aforementioned specific common region is the end of the object to be detected. The ultrasonic image generation system according to claim 1 or claim 2.

4. The aforementioned ultrasound data is an ultrasound image, The aforementioned unit is a pixel, The aforementioned decision model is based on machine learning performed using segmentation. The ultrasonic image generation system according to claim 1 or claim 2.

5. The aforementioned determination value is a continuous value indicating the degree to which the detected object exists. The ultrasonic image generation system according to claim 1 or claim 2.

6. This method, performed using one or more computers, generates an image showing the internal state of an object from ultrasonic data obtained by an ultrasonic probe used to detect defects contained in the object being inspected. The system stores multiple judgment data, which are ultrasonic data, arranged along the scanning direction of the ultrasonic probe. A machine learning-trained judgment model is input with multiple training data sets, which are at least one of the aforementioned ultrasonic data or data that mimics it, and a teacher label in which a specific common region indicating a particular region of the target to be detected that exists in each of the training data sets is set as the position of the target to be detected. The model is then made to output a judgment value that can determine whether or not the target to be detected exists for each unit indicating the position in each of the judgment data sets. The location of the target to be detected is extracted from the determined data generated based on the multiple determination values. The image is generated from a plurality of the determined data arranged along the scanning direction of the ultrasonic probe. The computer performs the processing, The data used for determination is a cross-sectional image showing a cross-section of the object, The determined data is the determined image of the cross-sectional image, The image generated using multiple determined data based on multiple determination values ​​represents the view of the object from a top direction different from the cross-sectional image. Ultrasound image generation method.

Images