Training data acquisition device and training data acquisition method

Abstract

To easily collect teacher data used for supervised learning of artificial intelligence.SOLUTION: A teacher data collection apparatus for collecting teacher data used for supervised learning of artificial intelligence comprises: a camera device which is configured to generate image data of an imaging range including a learning subject; a display device which is configured to display the image data and generate target range data indicating a range being a portion of the imaging range and including a learning target on the basis of the designation of an operator on the displayed image data; and a recording device which is configured to input the image data generated by the camera device and the target range data and record the image data by applying the target range data to the image data.SELECTED DRAWING: Figure 1

Description

【Technical Field】 , , , 【0004】 , , 【0006】 , , 【0005】 , , , 【0007】 , , 【0001】 The present invention relates to a teacher data collection device and a teacher data collection method. 【Background Art】 【0002】 Patent Document 1 discloses a device for detecting scratches on the surface of steel products and the like. 【Prior Art Documents】 【Patent Documents】 【0003】 【Patent Document 1】 Japanese Unexamined Patent Application Publication No. 2014-112064 【Summary of the Invention】 【Problems to be Solved by the Invention】 【0004】 In relation to detecting scratches and the like on the surface of steel products and the like, a technique of performing a work of determining scratches and the like (hereinafter sometimes referred to as "scratch and the like determination work") by having artificial intelligence (hereinafter sometimes referred to as "AI") learn it has been attracting attention in recent years. 【0005】 By the way, in order to have AI learn the work of determining scratches and the like, it is necessary to collect teacher data used for supervised learning of AI. These teacher data are collected by photographing the range including scratches and the like and recording the photographed image in a recording device. However, collecting such teacher data has been a lot of work and time-consuming. Moreover, this effort has been a double effort of photographing an image and selecting an object from the photographed image. 【0006】 The present invention has been made in view of such problems, and an object thereof is to easily collect teacher data used for supervised learning of artificial intelligence. 【Means for Solving the Problems】 【0007】 The main invention for achieving the above objective is a training data collection device for collecting training data to be used in supervised learning of artificial intelligence, comprising: a camera device configured to generate image data of a shooting range including a learning target; a display device configured to display the image data and to generate target range data indicating a part of the shooting range that includes the learning target, based on the operator's specification on the displayed image data; and a recording device configured to be able to input the image data and the target range data generated by the camera device and to record the image data by adding the target range data to the image data. 【0008】 Other features of the present invention will be made clearer by description in this specification and the accompanying drawings. [Effects of the Invention] 【0009】 According to the present invention, training data used for supervised learning of artificial intelligence can be easily collected. [Brief explanation of the drawing] 【0010】 [Figure 1] This is an explanatory diagram of one embodiment of the training data collection device of the present invention. [Figure 2] This is an explanatory diagram of the hardware configuration of the training data acquisition device of this embodiment. [Figure 3] This is a functional block diagram of the training data acquisition device of this embodiment. [Figure 4] This is a flowchart of one embodiment of the training data collection method of the present invention. [Figure 5] Figure 5A shows the first half of the training data collection method of this embodiment, with Figure 5A showing the determination of the shooting range (S002) and Figure 5B showing the shooting (S003). [Figure 6]The diagram shows the latter half of the training data collection method of this embodiment. Figure 6A shows the generation of target range data (S004), and Figure 6B shows the input of target type data and irradiation angle data (S004). [Figure 7] Figure 7A shows an example of an inspection record, Figure 7B shows an example of an image after information has been added, and Figure 7B shows an example of the added information. [Modes for carrying out the invention] 【0011】 The following information will become clear from the description and drawings described later. 【0012】 A training data collection device for collecting training data to be used in supervised learning of artificial intelligence is revealed, comprising: a camera device configured to generate image data of a shooting range including a learning target; a display device configured to display the image data and to generate target range data indicating a part of the shooting range that includes the learning target, based on the operator's specification on the displayed image data; and a recording device configured to be able to input the image data and the target range data generated by the camera device, and to record the image data by adding the target range data to the image data. 【0013】 Such a training data collection device makes it easy to collect training data used for supervised learning of artificial intelligence. 【0014】 Such a training data collection device, wherein the camera device further comprises a laser pointer, It is desirable that the laser pointer is positioned so that the laser is shone onto a predetermined position within the imaging range. 【0015】 Such a training data collection device allows for capturing images while positioning the learning target at a desired location within the shooting range. 【0016】 Such a teacher data collection device is configured such that the display device can further generate target type data indicating the type of the learning target based on an operator's input, and the recording device is configured to be able to input the target type data generated by the display device and further add the target type data to the image data to record the image data. It is desirable that this be the case. 【0017】 According to such a teacher data collection device, teacher data including target type data can be easily collected. 【0018】 Such a teacher data collection device preferably further includes a mounting portion for mounting the camera device and a head-mounted body configured to be wearable on the head of a human body. 【0019】 According to such a teacher data collection device, it is no longer necessary for an inspector to hold the camera device to perform work, and it becomes possible to perform work while holding other devices in both hands. Also, it is possible to easily photograph the learning target in a photographing direction that is substantially the same as the direction of the inspector's line of sight. 【0020】 Such a teacher data collection device preferably further includes a voice input device, and the camera device and the display device are configured to be operable based on a command input by the operator's voice via the voice input device. 【0021】 According to such a teacher data collection device, both hands are being used for other work, and even when it is not possible to operate by hand, the camera device and the display device can be operated. 【0022】 Such a teacher data collection device is configured such that the camera device has a camera azimuth / tilt angle sensor and can further generate photographing angle data indicating the photographing direction, and the recording device is configured to be able to input the photographing angle data generated by the camera device and further add the photographing angle data to the image data to record the image data. It is desirable that this be the case. 【0023】 Such a training data acquisition device makes it possible to easily collect training data, including shooting angle data. 【0024】 Preferably, such a training data acquisition device further comprises a portable lighting device for illuminating the shooting range, which has an illumination direction and tilt angle sensor and is configured to generate illumination angle data indicating the direction of light illumination when the camera device is shooting, and the recording device is configured to be able to input the illumination angle data generated by the portable lighting device and to record the image data by further adding the illumination angle data to the image data. 【0025】 Such a training data acquisition device makes it possible to easily acquire training data, including irradiation angle data. 【0026】 In such a training data collection device, it is desirable that the camera device further includes a GPS (Global Positioning System) device and is configured to generate location information data indicating the position of the camera device, and that the recording device is configured to be able to input the location information data generated by the camera device and to record the image data by further adding the location information data to the image data. 【0027】 Such a training data collection device makes it possible to easily collect training data, including location information data. 【0028】 In such a training data collection device, it is desirable that the learning target is at least one of scratches, defects, and stains on the surface of steel products. 【0029】 Such a training data collection device makes it possible to easily collect training data regarding defects in steel products. 【0030】 A method for collecting training data to be used in supervised learning of artificial intelligence is revealed, characterized by comprising: generating image data of a shooting range including the learning target using a camera device; generating target range data indicating a part of the shooting range that includes the learning target based on a specification on the image data displayed on a display device; and recording the image data by adding the target range data to the image data. 【0031】 This method of collecting training data makes it easy to collect training data used for supervised learning of artificial intelligence. 【0032】 Preferably, the training data collection method further includes a camera device comprising a laser pointer positioned such that the laser is shone at a predetermined position within the shooting range, and further includes setting the shooting range based on the reflected light of the laser. 【0033】 This method of collecting training data allows for capturing images while positioning the learning target at a desired location within the shooting range. 【0034】 Preferably, such a training data collection method further includes the camera device further comprising a camera orientation and tilt angle sensor, generating shooting angle data indicating the shooting direction using the camera device, and further adding the shooting angle data to the image data and recording the image data. 【0035】 This method of collecting training data makes it easy to collect training data, including shooting angle data. 【0036】 Preferably, such a training data collection method further includes illuminating the shooting range with a portable lighting device equipped with an illumination direction and tilt angle sensor, generating illumination angle data indicating the direction of light irradiation during shooting by the camera device, and further adding the illumination angle data to the image data and recording the image data. 【0037】 This method of collecting training data allows for the easy collection of training data, including irradiation angle data. 【0038】 In such a training data collection method, it is preferable that the learning target is at least one of scratches, defects, and stains on the surface of steel products. 【0039】 This method of collecting training data makes it easy to collect training data on defects in steel products. 【0040】 ===Execution=== <Overview> Figure 1 is an explanatory diagram of one embodiment of the training data acquisition device of the present invention. 【0041】 The training data collection device 10 is a device for collecting training data used in supervised learning of artificial intelligence. Here, "supervised learning" is a machine learning method that enables artificial intelligence to recognize patterns from pre-provided data and make predictions on newly input data. "Training data" is the data necessary for performing supervised learning of artificial intelligence. In other words, training data is the data necessary for artificial intelligence to make predictions, etc. 【0042】 For artificial intelligence (AI) trained using supervised learning models to make highly accurate predictions, it requires training data containing a sufficient amount of information. In other words, the more training data used in supervised learning of AI, the more reliable the AI's predictions become. 【0043】 The training data collected by the training data collection device 10 of this embodiment is data relating to scratches, etc., on steel products. Here, "scratches, etc." refers to at least one of scratches, defects, and stains. Conventionally, the judgment of scratches, etc., was sometimes performed by skilled inspectors through visual inspection. Therefore, training data relating to scratches, etc., on steel products can be collected by having the inspector photograph the area containing the scratches, etc., and recording the captured image in a recording device. However, the work involved in collecting such training data is cumbersome, and furthermore, it is a double process of taking the image and then selecting the target object from the captured image. 【0044】 The training data collection device 10 of this embodiment can easily collect a large amount of data related to scratches and other defects on the surface of steel products. In other words, the training data collection device 10 can easily collect training data used for supervised learning of artificial intelligence. Below, with reference to Figure 1, the configuration of the training data collection device 10 of this embodiment in the scenario of performing scratch and defect judgment work on object 1, which is a scratch or other defect, will be described. Note that "data related to scratches and other defects" is not limited to image data of scratches and other defects, but also includes information related to scratches and other defects (hereinafter sometimes referred to as "additional information"). Additional information includes, for example, the type and size of scratches and other defects. 【0045】 <Structure> The teacher data acquisition device 10 includes a camera device 12, a display device 17, and a recording device 18. Furthermore, the teacher data acquisition device 10 of this embodiment also includes a head-mounted body 19, an audio input device 21, and a portable lighting device 22. 【0046】 The camera device 12 is a device having a camera 13 for photographing object 1. The camera 13 in the camera device 12 is not limited to a camera that takes still images; it may also be a camera that takes video (a video camera). Furthermore, the camera 13 may be a camera capable of securing a wide shooting range, such as a so-called 360-degree camera, a camera equipped with a fisheye lens, or a hemispherical camera. 【0047】 In addition to the camera 13, the camera device 12 includes a laser pointer 14, a camera orientation / tilt angle sensor 15, and a GPS device 16. 【0048】 The laser pointer 14 is a device that uses a laser to point to object 1. In this embodiment, the laser pointer 14 is positioned adjacent to the camera 13, as shown in Figure 1. The location pointed to by the laser pointer 14 is set to be the center of the camera's shooting range 2. This makes it easy to position object 1 at the center of the camera's shooting range 2 by having the camera 13 capture an image of it with the laser pointer 14 pointing to it. 【0049】 However, the laser pointer 14 does not necessarily have to be positioned at the center of the camera 13's shooting range 2; it may be positioned at any location, such as the upper right corner of the camera 13's shooting range 2. In other words, by positioning the laser pointer 14 so that its laser beam shines at a predetermined location within the shooting range 2, it is possible to position the object 1 at a desired location within the shooting range 2 while taking a photograph. 【0050】 The camera orientation / tilt angle sensor 15 is a device that measures the orientation (i.e., direction in the horizontal plane) and tilt angle (i.e., angle relative to the horizontal plane) of the camera 13's shooting direction. In this embodiment, the camera orientation / tilt angle sensor 15 is a gyro sensor. By measuring the shooting direction of the camera 13 using the camera orientation / tilt angle sensor 15, the camera device 12 can acquire information on the orientation and tilt angle of the camera 13's shooting direction. 【0051】 The GPS device 16 is a device that measures the position of the camera 13 using GPS (Global Positioning System). By measuring the position of the camera 13 with the GPS device 16, the camera device 12 can acquire the position information of the camera 13. 【0052】 The display device 17 is a device that displays images captured by the camera 13. In this embodiment, the display device 17 is, for example, a liquid crystal display (LCD), a plasma display panel (PDP), or an organic electro-luminescence diode (OLED). However, the display device 17 may also be a projector that displays images by projecting them onto a screen or the wall of a structure. 【0053】 As shown in Figure 1, the display device 17 is attached to the portable lighting device 22, but the location where the display device 17 is attached is not limited to this. The display device 17 may be attached to the recording device 18, or to the head-mounted device 19, for example. When the display device 17 is attached to the recording device 18, it may function as, for example, a laptop computer. When the display device 17 is attached to the head-mounted device 19, it may function as, for example, wearable glasses. 【0054】 The recording device 18 is a main memory device such as RAM (Random Access Memory) or an auxiliary memory device such as a hard disk drive or SSD (Solid State Drive). The recording device 18 records programs executed by the training data acquisition device 10 and data generated by the camera device 12, etc. Note that the recording device 18 is not limited to being installed near the inspector or carried by the inspector as shown in Figure 1, but may also be installed in a server that is connected to it via a communication device (not shown). 【0055】 The head-mounted device 19 is a device for attaching the camera device 12 to the head of a person. The head-mounted device 19 is configured to be wearable on the head of an inspector performing a scratch and other damage assessment. In this embodiment, the head-mounted device 19 is a helmet. However, the head-mounted device 19 is not limited to a helmet, and may be, for example, a cap, visor, band, etc. 【0056】 The head-mounted device 19 has a mounting section 20 for attaching the camera device 12. The inspector attaches the head-mounted device 19, to which the camera device 12 (camera 13 and laser pointer 14) is attached, to their head via the mounting section 20. This eliminates the need for the inspector to hold the camera device 12 while working, allowing them to work while holding other devices, such as a portable lighting device 22, in both hands. Furthermore, the inspector's line of sight will be approximately aligned with the shooting direction of the camera 13 and the laser beam direction of the laser pointer 14. That is, the laser pointer 14 can be used to point to the target 1 in approximately the same direction as the inspector's line of sight. Also, the camera 13 can easily photograph the target 1 in approximately the same direction as the inspector's line of sight. However, devices other than the camera device 12, such as a portable lighting device 22, may also be attached to the head-mounted device 19 via the mounting section 20. 【0057】 The voice input device 21 is a device for inputting voice commands related to operating each device of the training data acquisition device 10. In this embodiment, the voice input device 21 is a microphone. The camera device 12 and the display device 17 are configured to operate based on commands input by the inspector's voice via the voice input device 21. 【0058】 In this embodiment, for example, the inspector can activate the shutter of the camera 13 of the camera device 12 by inputting "shutter" into the voice input device 21. In other words, even if the inspector is using both hands for another task and cannot operate the camera device 12 and the display device 17 by hand, they can still operate the camera device 12 and the display device 17. However, the voice input device 21 may also be a device for inputting voice commands to operate devices other than the camera device 12 and the display device 17. For example, the voice data of the voice input by the inspector may be recorded in the recording device 18 via the voice input device 21. 【0059】 The portable lighting device 22 is a device having a portable light 23 that emits a light 4 to illuminate the shooting range 2. The object 1 can be made easier to identify by illuminating it from a direction different from the inspector's line of sight. Therefore, by using the portable light 23, the work of determining scratches, etc., can be made easier. In this embodiment, the portable light 23 is a flashlight. 【0060】 The portable lighting device 22 includes a portable light 23 and a lighting direction / tilt angle sensor 24. The lighting direction / tilt angle sensor 24 is a device that measures the direction (i.e., direction in the horizontal plane) and tilt angle (i.e., angle relative to the horizontal plane) of the illumination direction of the portable light 23. In this embodiment, the lighting direction / tilt angle sensor 24 is a gyro sensor. By measuring the illumination direction of the portable light 23 using the lighting direction / tilt angle sensor 24, the portable lighting device 22 can acquire information on the direction and tilt angle of the illumination direction of the portable light 23. 【0061】 <System Configuration> The training data acquisition device 10 in this embodiment is a system that includes a computer, such as a personal computer or a server. Therefore, the system configuration of the training data acquisition device 10 will be described below with reference to Figures 2 and 3. 【0062】 Figure 2 is an explanatory diagram of the hardware configuration of the training data acquisition device 10 in this embodiment. The training data acquisition device 10 may consist of one computer or multiple computers. 【0063】 The training data acquisition device 10 (computer) consists of a CPU 11 and various devices that are connected to the CPU 11 in a manner that enables communication. Data can be communicated between the CPU 11 and the various devices via wired or wireless means (for example, Wi-Fi® or Bluetooth®). 【0064】 The CPU 11 is a processing unit that controls the training data acquisition device 10. 【0065】 The various devices are a camera device 12, a display device 17, a recording device 18, an audio input device 21, and a portable lighting device 22, all attached to the head-mounted body 19 described above. However, the various devices may also include other devices, such as a communication device for communicating with devices other than the training data acquisition device 10, and input devices (e.g., a keyboard or mouse). 【0066】 Figure 3 is a functional block diagram of the training data acquisition device 10 of this embodiment. 【0067】 The training data acquisition device 10 has a processing unit 30 that receives and processes data from the camera device 12, the voice input device 21, and the portable lighting device 22. The processing unit 30 has a control unit 40 and a data recording unit 50. 【0068】 The control unit 40 controls the various processes that the training data acquisition device 10 should perform. The control unit 40 is realized by the CPU 11 executing a control program recorded in the recording device 18 to perform various controls. The control unit 40 has an image acquisition unit 41, an additional information acquisition unit 42, and an additional information display unit 43. Each function will be described later. 【0069】 The data recording unit 50 is a recording unit for recording predetermined data. The various types of data recorded in the data recording unit 50 are stored in the recording device 18. Details of the various types of data will be described later. 【0070】 <Flowchart for collecting training data> Figure 4 is a flowchart of one embodiment of the training data collection method of the present invention. Figure 5 shows the first half of the training data collection method of this embodiment. Figure 5A shows the determination of the shooting range (S002), and Figure 5B shows the shooting (S003). Figure 6 shows the second half of the training data collection method of this embodiment. Figure 6A shows the generation of target range data (S004), and Figure 6B shows the input of target type data and irradiation angle data (S004). Figure 7 shows an example of an inspection record, Figure 7A shows an example of an image after information has been added, and Figure 7B shows an example of added information. 【0071】 Next, referring again to Figures 1 to 3 mentioned above, and also referring to Figures 4 to 7, the method for collecting training data in this embodiment will be explained. Note that in each step of Figure 4, each process performed by the training data collection device 10 is executed by the CPU 11 reading and executing the program recorded in the recording device 18. 【0072】 First, as shown in Figure 5A, the inspector puts on a head-mounted device 19 to which a camera device 12 is attached, and an audio input device 21, and checks the surface of the object to be inspected 100 (S001 in Figure 4). In other words, the inspector determines whether or not there is a target 1 (scratch, etc.) on the surface of the object to be inspected 100. 【0073】 If object 1 is present on the surface of the object to be inspected 100, the inspector then determines the shooting range 2 of the camera 13 (S002 in Figure 4). In this embodiment, the inspector determines the shooting range 2 so that object 1 is positioned in the center of the shooting range 2, as shown in Figure 5B, while checking with the viewfinder of the camera 13. However, the shooting range 2 of the camera 13 may also be displayed on the display device 17, in which case the inspector determines the shooting range 2 while checking with the display device 17. 【0074】 As described above, since the camera device 12 (camera 13 and laser pointer 14) is attached to the head-mounted device 19 worn on the inspector's head, the inspector's line of sight, the shooting direction of the camera 13, and the laser beam direction of the laser pointer 14 will be approximately the same. Therefore, the laser pointer 14 can be used to point to the target 1 in approximately the same direction as the inspector's line of sight, making it easy to determine the shooting range 2. 【0075】 After determining the shooting range 2, the inspector takes a photograph within shooting range 2 (S003 in Figure 4). In this embodiment, as shown in Figure 5B, the inspector inputs "shutter" into the voice input device 21, which activates the shutter of camera 13 of camera device 12. This makes it possible to take an image under the same conditions as when the inspector was observing object 1, even if both hands are being used for other tasks and the inspector cannot operate the camera by hand. However, the shutter switch of camera 13 may also be operated manually. 【0076】 Furthermore, by taking a photograph, the camera device 12 generates image data of the shooting range 2, which includes the object 1. After the camera device 12 takes the photograph, the image data is recorded in the recording device 18. That is, the image acquisition unit 41 shown in Figure 3 saves the image data stored in the camera device 12 as the original image data to the data recording unit 50. The original image data may contain multiple still images. 【0077】 Next, the inspector inputs additional information (S004 in Figure 4). In this embodiment, the target range data for target 1 can be generated as additional information on the image displayed on the display device 17 based on the inspector's specifications. Here, "target range" refers to a part of the shooting range 2 that includes target 1. Specifically, as shown in Figure 6A, the target range 5 is generated on the display device 17 as the range surrounding target 1. 【0078】 In this embodiment, as shown in Figure 6A, the inspector can input, for example, "15mm" into the voice input device 21 to display a circle with a diameter of 15mm as the target area 5 at the center of the shooting area 2. The size of the frame can be adjusted by the value input into the voice input device 21. By making the target area 5 surround the target 1, the target area data can be used as information about the size of the target 1. The shape of the target area 5 is not limited to a circle, but can be any shape such as a quadrilateral. In this way, the target area data for the target 1 can be generated on the image displayed on the display device 17 based on the inspector's specifications. 【0079】 After the target range data is generated, the target range data is recorded in the recording device 18. Specifically, the additional information acquisition unit 42 shown in Figure 3 acquires the target range data and inputs it to the data recording unit 50. It also adds the target range data to the original image data and saves it in the data recording unit 50. Then, the additional information display unit 43 displays the image with the target range data added to the original image data on the display device 17. 【0080】 Furthermore, in this embodiment, as additional information, target type data for target 1 can also be generated on the image displayed on the display device 17 based on the inspector's specifications. Here, "target type" refers to the type of target 1, such as information like "scratch," "defect," "stain," or "other." However, the target type may also be information such as "normal response" or "emergency response." 【0081】 In this embodiment, as shown in Figure 6B, the inspector can display the target type data by inputting, for example, "scratch" into the voice input device 21. In this way, the target type data for target 1 can be generated on the image displayed on the display device 17 based on the inspector's specification. 【0082】 After the target type data is generated, the target type data is recorded in the recording device 18. Specifically, the additional information acquisition unit 42 shown in Figure 3 acquires the target type data and inputs it to the data recording unit 50. It also adds the target type data to the original image data and saves it in the data recording unit 50. Then, the additional information display unit 43 displays the image with the target type data added to the original image data on the display device 17. 【0083】 Furthermore, the additional information is not limited to target range data or target type data. The training data collection device 10 of this embodiment can also generate additional information such as shooting angle data, illumination angle data, and position information data, as will be described later. 【0084】 As described above, the camera device 12 has a camera orientation and tilt angle sensor 15. By measuring the shooting direction of the camera 13 using the camera orientation and tilt angle sensor 15, the camera device 12 can further generate shooting angle data indicating the shooting direction. 【0085】 After the shooting angle data is generated, the shooting angle data is recorded in the recording device 18. That is, the additional information acquisition unit 42 shown in Figure 3 acquires the shooting angle data and inputs it to the data recording unit 50. The shooting angle data is also added to the original image data and saved in the data recording unit 50. Although not shown, the additional information display unit 43 may also display the image with the shooting angle data added to the original image data on the display device 17. 【0086】 Furthermore, as described above, the portable lighting device 22 has a lighting direction and tilt angle sensor 24. By measuring the illumination direction of the portable lighting device 23 using the lighting direction and tilt angle sensor 24, the portable lighting device 22 can further generate illumination angle data indicating the direction of light illumination when the camera device 12 is taking pictures. 【0087】 After the irradiation angle data is generated, the irradiation angle data is recorded in the recording device 18. That is, the additional information acquisition unit 42 shown in Figure 3 acquires the irradiation angle data and inputs it to the data recording unit 50. It also adds the irradiation angle data to the original image data and saves it in the data recording unit 50. Although not shown, the additional information display unit 43 may also display the image with the irradiation angle data added to the original image data on the display device 17. 【0088】 Furthermore, as mentioned above, the camera device 12 has a GPS device 16. By measuring the position of the camera 13 using the GPS device 16, the camera device 12 can further generate positional information data indicating the position of the camera device 12. 【0089】 After the location information data is generated, the location information data is recorded in the recording device 18. That is, the additional information acquisition unit 42 shown in Figure 3 acquires the location information data and inputs it to the data recording unit 50. It also adds the location information data to the original image data and saves it in the data recording unit 50. Although not shown, the additional information display unit 43 may also display the image with the location information data added to the original image data on the display device 17. 【0090】 As a result, information necessary for the scratch and other defect detection process can be added to the image of target 1, such as the conditions under which the image was taken and the angle of illumination. 【0091】 Finally, the inspector generates the inspection record (S005 in Figure 4). In the inspection record, as shown in Figure 7A, the image with additional information added to the original image can be displayed on the display device 17. In addition, in the inspection record, as shown in Figure 7B, the additional information can also be saved linked to the original image. 【0092】 Therefore, inspectors can easily take pictures of any defects they find, and they can also easily attach information necessary for identifying defects to these images. In other words, it becomes easy to collect training data to be used in supervised learning for artificial intelligence. 【0093】 ===Other=== Although the present invention has been described above using embodiments, the present invention is not limited to these embodiments. The scope of the present invention is determined based on the appended claims, and within that scope, all configurations that omit, modify, or improve upon some of the components shown in the embodiments are included in the present invention. 【0094】 For example, the above embodiment shows a preferred configuration including a head-mounted body 19, an audio input device 21, and a portable lighting device 22. However, the training data acquisition device of the present invention does not necessarily have to include these, or it may include one or two of them. 【0095】 Furthermore, although the learning target in the above embodiment was scratches on the surface of steel products, the present invention is not limited thereto. The present invention can also be suitably applied to a training data collection device and training data collection method for collecting training data with impurities and foreign substances in steel raw materials as the learning target, for example, with the aim of obtaining a trained model for detecting copper materials and hazardous materials when receiving iron scrap into a raw material yard. [Explanation of Symbols] 【0096】 1. Target 2. Shooting range 3. Laser light 4 Lighting 5. Scope 10 Training Data Acquisition Device 11 CPU 12 Camera equipment 13 Cameras 14. Laser Pointer 15. Camera orientation and tilt angle sensor 16 GPS device 17 Display device 18 Recording device 19 Head-mounted body 20 Mounting part 21. Voice input device 22 Portable lighting devices 23 Portable lighting 24. Lighting Direction and Tilt Angle Sensor 30 Processing Unit 40 Control Unit 41 Image acquisition unit 42 Additional Information Acquisition Unit 43 Additional Information Display Unit 50 Data recording unit 100 Items to be inspected

Claims

[Claim 1] A training data collection device for collecting training data used in supervised learning of artificial intelligence, A camera device configured to generate image data of a shooting range that includes a learning target which is at least one of scratches, defects, and stains on the surface of a steel product, A display device that displays the aforementioned image data and is configured to generate target range data indicating a part of the shooting range that includes the learning target, based on the operator's specification on the displayed image data, A recording device configured to be able to input the image data and target range data generated by the camera device, and to record the image data by adding the target range data to the image data, A portable lighting device for illuminating the shooting range is provided, which has an illumination direction and tilt angle sensor and is configured to generate illumination angle data indicating the direction of light illumination when the camera device is taking pictures. A training data acquisition device characterized in that the recording device is configured to be able to input the irradiation angle data generated by the portable lighting device, and to record the image data by further adding the irradiation angle data to the image data. [Claim 2] The camera device further includes a laser pointer, The training data acquisition device according to claim 1, characterized in that the laser pointer is arranged so that the laser is shone at a predetermined position within the shooting range. [Claim 3] The display device is configured to further generate target type data indicating the type of the learning target based on the operator's input. The training data collection device according to claim 1 or 2, characterized in that the recording device is configured to be able to input the target type data generated by the display device, and to be able to record the image data by further adding the target type data to the image data. [Claim 4] The training data acquisition device according to any one of claims 1 to 3, further comprising a head mount having a mounting portion for attaching the aforementioned camera device and configured to be attachable to the head of a human body. [Claim 5] It is further equipped with an audio input device, The training data acquisition device according to claim 4, characterized in that the camera device and the display device are configured to operate based on commands input by the operator's voice via the voice input device. [Claim 6] The camera device is configured to have a camera orientation and tilt angle sensor and to further generate shooting angle data indicating the shooting direction. The training data acquisition device according to any one of claims 1 to 5, characterized in that the recording device is configured to be able to input the shooting angle data generated by the camera device and to record the image data by further adding the shooting angle data to the image data. [Claim 7] The camera device is further equipped with a GPS (Global Positioning System) device and is configured to generate positional information data indicating the position of the camera device. The training data collection device according to any one of claims 1 to 5, characterized in that the recording device is configured to be able to input the location information data generated by the camera device and to record the image data by further adding the location information data to the image data. [Claim 8] A method for collecting training data to be used in supervised learning of artificial intelligence, A camera device and a portable lighting device having an illumination direction and tilt angle sensor, for illuminating a shooting range that includes a learning target which is at least one of scratches, defects, and stains on the surface of a steel product, are used to generate image data of the shooting range and illumination angle data indicating the direction of light illumination when the camera device is shooting. Based on the designation on the image data displayed on the display device, target range data is generated that indicates a part of the shooting range that includes the learning target, A method for collecting training data, characterized by recording the image data by adding the target range data and the irradiation angle data to the image data. [Claim 9] The camera device further comprises a laser pointer positioned so as to irradiate a laser at a predetermined position within the shooting range. The training data acquisition method according to claim 8, further comprising setting the shooting range based on the reflected light of the laser. [Claim 10] The aforementioned camera device further includes a camera orientation and tilt angle sensor, The camera device generates shooting angle data indicating the shooting direction. The training data acquisition method according to claim 8 or 9, further comprising adding the shooting angle data to the image data and recording the image data.

Images