Foreign object detection device for clothing and accessories, and foreign object detection system

The foreign object detection system improves accuracy by using dual-direction X-ray imaging and AI model updates to address the limitations of visual inspection in existing systems, effectively reducing errors in detecting foreign objects in clothing and miscellaneous goods.

WO2024242205A1PCT designated stage expired Publication Date: 2024-11-28KUWAHARA CO LTD

Patent Information

Application Number
PCT/JP2024/023547
Authority / Receiving Office
WO · WO
Patent Type
Applications
Current Assignee / Owner
Priority Date
2023-05-24
Filing Date
2024-06-28
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Existing foreign object detection systems for clothing and miscellaneous goods have low accuracy due to reliance on visual inspection by operators, leading to potential overlooking or overdetection of foreign objects, and existing solutions like abnormality inspection systems using learning models do not consistently improve detection accuracy.

Method used

A foreign object detection system that uses an X-ray source to irradiate objects from two directions, with upstream and downstream sensors capturing imaging data, and an AI inspection model file generated based on standard sample image data, which is updated using data from overlooked or overdetected objects to improve detection accuracy and prevent future errors.

Benefits of technology

The system significantly reduces the likelihood of overlooking or overdetecting foreign objects by comparing images from two directions and continuously learning from errors, thereby enhancing the accuracy of foreign object detection and facilitating easier learning and updates.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure JP2024023547_28112024_PF_FP_ABST
    Figure JP2024023547_28112024_PF_FP_ABST
Patent Text Reader

Abstract

A foreign object detection unit (14) utilizes an AI inspection model file (16) stored in a file storage unit (25) to detect foreign objects on the basis of first imaging data from a first sensor (6b, 6c) or second imaging data from a second sensor (6d, 6e). The present invention includes an input means (9) operated by a worker when the foreign object detection unit (14) overlooks a foreign object or overdetects a foreign object, and foreign object information memory (17) that stores information concerning foreign objects that are overlooked or overdetected on the basis of signals from the input means (9). The AI inspection model file (16) is an initial model file generated on the basis of standard foreign object sample image data, or a basic model file obtained by updating the initial model file on the basis of the information stored by the foreign object information memory (17), the first imaging data, and the second imaging data.
Need to check novelty before this filing date? Find Prior Art

Description

Foreign object detection device and foreign object detection system for clothing and miscellaneous goods

[0001] The present disclosure relates to a foreign object detection device that detects whether a foreign object is contained in clothing or miscellaneous goods, a foreign object information acquisition method used in the foreign object detection device, and a foreign object detection system that detects foreign objects.

[0002] Clothing includes, for example, accessories such as shoes and bags, as well as clothing. Miscellaneous items include small items used in daily life. Foreign object detection devices that detect foreign objects contained in inspected items such as clothing and miscellaneous items have been known for some time. The foreign object detection device includes, for example, a transport path for transporting the inspected items and an X-ray inspection device. The X-ray inspection device includes an X-ray source that irradiates the inspected items on the transport path with X-rays from above, and a sensor that receives the X-rays from above. The foreign object detection device includes a monitor that displays an image based on the imaging data from the sensor. An operator visually checks the image displayed on the monitor to determine whether or not a foreign object is present in the inspected items. In this way, foreign objects contained in the inspected items are detected.

[0003] However, relying on the visual inspection of an operator to detect foreign objects displayed on a monitor can result in low accuracy in detecting the presence or absence of foreign objects. For example, an operator may mistakenly determine that a foreign object is not present in an inspected object when it is, or may mistakenly determine that a foreign object is present in an inspected object when it is not. To address this issue, anomaly inspection systems using learning models have been developed, as disclosed in Japanese Patent No. 7034529. For example, the accuracy of foreign object detection can be improved by using a mask image that colors the areas with foreign objects black and the rest white, or by using a background image without foreign objects.

[0004] The object of the present disclosure is to provide a foreign object detection device for clothing and miscellaneous goods that can detect foreign objects with high accuracy or that can facilitate learning to improve detection accuracy, a foreign object information acquisition method, and a foreign object detection system.

[0005] According to another feature of the present disclosure, a foreign object detection device for detecting foreign objects contained in clothing and miscellaneous goods includes a conveying path for conveying the clothing and miscellaneous goods. An X-ray source irradiates X-rays onto the clothing and miscellaneous goods on the conveying path from a first direction and a second direction. An upstream sensor is provided below the conveying path and receives X-rays from the first direction. A downstream sensor is provided downstream of the upstream sensor and below the conveying path and receives X-rays from the second direction. A file storage unit stores an AI inspection model file. A foreign object detection unit uses the AI ​​inspection model file to detect foreign objects based on first imaging data from the upstream sensor or second imaging data from the downstream sensor. A first monitor displays a first image based on the first imaging data. A first result monitor is provided alongside the first monitor and displays a first foreign object indication related to the location of a foreign object detected in the first imaging data, superimposed on the first image. A second monitor is provided in a location visible from a location visible to the first monitor and displays a second image based on the second imaging data. A second result monitor is arranged next to the second monitor and displays a second foreign object display related to the location of a foreign object detected in the second imaging data superimposed on the second image. When the foreign object detection unit overlooks or overdetects a foreign object, the operator operates the input means. Based on a signal from the input means, the foreign object information memory stores information on the overlooked foreign object or the overdetected foreign object. The AI ​​inspection model file is an initial model file generated based on standard foreign object sample image data, or a basic model file obtained by updating the initial model file based on the information stored in the foreign object information memory and the first imaging data and second imaging data. The first monitor and the first result monitor are arranged next to each other on the upstream side of the transport of the main body that houses the X-ray source. The second monitor and the second result monitor are arranged next to each other on the downstream side of the transport of the main body.

[0006] Another feature of the present disclosure relates to a foreign object detection device for clothing and miscellaneous goods that detects foreign objects contained in clothing and miscellaneous goods. The foreign object detection device has a conveying path for conveying clothing and miscellaneous goods. An X-ray source irradiates X-rays from a first direction and a second direction onto clothing and miscellaneous goods on the conveying path. A first sensor receives the X-rays from the first direction. A second sensor receives the X-rays from the second direction. A file storage unit stores an AI inspection model file. A foreign object detection unit uses the AI ​​inspection model file to detect a foreign object based on first imaging data from the first sensor or second imaging data from the second sensor. A first monitor displays a first image based on the first imaging data. A first result monitor is arranged alongside the first monitor and displays a first foreign object indication related to the location of a foreign object detected in the first imaging data, superimposed on the first image. A second monitor is arranged in a location visible from a location where the first monitor is visible, and displays a second image based on the second imaging data. A second result monitor is provided in parallel with the second monitor and displays a second foreign object display related to the location of a foreign object detected in the second imaging data superimposed on the second image. The input means is operated by an operator when the foreign object detection unit overlooks or overdetects a foreign object. The foreign object information memory stores information about the overlooked foreign object or the overdetected foreign object based on a signal from the input means. The AI ​​inspection model file is an initial model file generated based on the standard foreign object sample image data, or a basic model file obtained by updating the initial model file based on the information stored in the foreign object information memory and the first and second imaging data.

[0007] Therefore, the presence or absence of foreign objects is detected based on X-rays irradiated from two directions onto clothing and miscellaneous goods. Therefore, compared to detecting the presence or absence of foreign objects based on X-rays irradiated from one direction, the overlooking and overdetection of foreign objects can be reduced. Furthermore, because the first image and the first image with the first foreign object display superimposed thereon are displayed side by side, these two first images can be compared. Therefore, by comparing the first monitor and the first result monitor, the operator can easily determine whether the foreign object detection unit has overlooked or overdetected a foreign object. The same applies to the second image. Furthermore, the foreign object detection unit includes an initial model file or a basic model file that updates the initial model file. The basic model file is trained (updated) based on the first imaging data or second imaging data in which a foreign object was overlooked or overdetected, thereby reducing the overlooking and overdetection of the same foreign object in future. Therefore, the accuracy of foreign object detection is improved. Training is also simplified.

[0008] According to another feature of the present disclosure, the AI ​​inspection model file is generated by an external device using a model file creation program. Therefore, an AI inspection model file with high detection accuracy can be generated. Furthermore, because the AI ​​inspection model file is generated by an external device, the involvement of a user of the foreign object detection device in generating the AI ​​inspection model file can be reduced.

[0009] According to another feature of the present disclosure, the first foreign object display or the second foreign object display is a frame that surrounds the foreign object and is displayed in a flashing manner, thereby preventing an operator monitoring the first result monitor or the second result monitor from overlooking the foreign object.

[0010] According to another feature of the present disclosure, the foreign object detection unit detects a foreign object based on the first imaging data or the second imaging data by dividing the first imaging data or the second imaging data into regions to obtain an image represented by pixels. The foreign object detection unit then determines that a foreign object is present when the number of consecutive pixels detected as foreign objects in the image exceeds a threshold. This allows the foreign object detection unit to easily detect the presence or absence of a foreign object.

[0011] Another feature of the present disclosure relates to a method for acquiring information on a foreign substance sample used in a foreign substance detection device for clothing and miscellaneous goods as foreign substance information. In the foreign substance information acquisition method, the foreign substance sample is inserted into a polyhedral sample holder made of resin. Next, the sample holder is placed with its first surface facing down on a resin tray or a sample tray made of the same material as the tray, which is placed on a transport path and on which clothing and miscellaneous goods are placed. The sample holder is then irradiated with X-rays. Next, the sample holder is placed on the tray or sample tray with its second surface facing down. The sample holder is then irradiated with X-rays. Next, the sample holder is placed on the tray or sample tray with its third surface facing down. The sample holder is then irradiated with X-rays. Next, information on the foreign substance sample in at least three directions is acquired as foreign substance information based on a signal from a sensor that receives the X-rays.

[0012] Therefore, by utilizing at least three faces of the polyhedron, the foreign substance information stored in the foreign substance information memory is stored from viewpoints from at least three directions. Therefore, the foreign substance information to be stored in the foreign substance information memory can be obtained quickly and easily. This improves the accuracy of foreign substance detection based on the foreign substance information.

[0013] According to another feature of the present disclosure, a foreign object detection system including a foreign object detection device for clothing and miscellaneous goods includes a server to which an AI inspection model file is uploaded from an external device, and a writing unit downloads the AI ​​inspection model file from the server and writes the AI ​​inspection model file into a file storage unit of the foreign object detection device for clothing and miscellaneous goods.

[0014] Therefore, when the AI ​​inspection model file stored in the server is updated, the AI ​​inspection model file stored in the file storage unit of the foreign matter detection device for clothing and miscellaneous goods can also be updated. Therefore, the AI ​​inspection model file stored in the foreign matter detection device for clothing and miscellaneous goods can be updated via the server without having to go to the installation location of the foreign matter detection device for clothing and miscellaneous goods. Therefore, updating of the AI ​​inspection model file can be easily performed.

[0015] According to another feature of the present disclosure, the foreign object detection system includes an X-ray inspection device that irradiates clothing and miscellaneous goods with X-rays to obtain imaging data. The foreign object detection means detects foreign objects from the imaging data using an AI inspection model file. The video monitor displays an image based on the imaging data. The result monitor displays a foreign object indication related to the location of a foreign object detected from the imaging data, superimposed on the image. The input means is operated by an operator when the foreign object detection unit overlooks or overdetects a foreign object. The foreign object information memory stores information on the overlooked foreign object or overdetected foreign object based on a signal from the input means. The AI ​​inspection model file is an initial model file generated by an external device, or a basic model file obtained by updating the initial model file by an external device. The basic model file is generated by updating the initial model file based on the information stored in the foreign object information memory and the imaging data.

[0016] For this reason, the foreign object detection unit is provided with an initial model file or a basic model file that is an updated version of the initial model file. The basic model file is learned (updated) based on imaging data in which a foreign object has been overlooked or overdetected, thereby preventing the same foreign object from being overlooked or overdetected in the future. This improves the accuracy of detecting the presence or absence of foreign objects. Learning is also simplified.

[0017] According to another feature of the present disclosure, the X-ray inspection device includes an X-ray source that irradiates X-rays from a first direction and a second direction onto the transported clothing and miscellaneous goods. An upstream sensor receives X-rays from the first direction. A downstream sensor is provided downstream of the upstream sensor and receives X-rays from the second direction. The imaging data includes first imaging data obtained based on X-rays from the upstream sensor and second imaging data obtained based on X-rays from the downstream sensor. The image monitor includes a first image monitor that displays an image based on the first imaging data and a second image monitor that displays an image based on the second imaging data. The result monitor includes a first result monitor that displays a first foreign object indication related to the location of a foreign object detected in the first imaging data superimposed on the first image, and a second result monitor that displays a second foreign object indication related to the location of a foreign object detected in the second imaging data superimposed on the second image. The first image monitor and the first result monitor are arranged side by side upstream of the X-ray inspection device. The second image monitor and the second result monitor are arranged side by side downstream of the X-ray inspection device.

[0018] 1 is a schematic diagram of a foreign object detection system. FIG. 2 is a front view of a foreign object detection device for clothing and miscellaneous goods. FIG. 3 is a block diagram of a foreign object detection system including a foreign object detection device for clothing and miscellaneous goods. FIG. 4 is a perspective view of a sample holder and a tray. FIG. 5 is an X-ray image of a foreign object sample using a sample holder with the first surface facing up. FIG. 6 is an X-ray image of a foreign object sample using a sample holder with the second surface facing up. FIG. 7 is an X-ray image of a foreign object sample using a sample holder with the third surface facing up. FIG. 8 is an X-ray image of a foreign object sample using a sample holder with the fourth surface facing up. FIG. 9 is an X-ray image of a foreign object sample using a sample holder with the fifth surface facing up. FIG. 10 is an X-ray image of a foreign object sample using a sample holder with the sixth surface facing up. FIG. 11 is a flow chart explaining the procedure for foreign object detection. FIG. 12 is an explanatory diagram of an image displayed on a first result monitor during foreign object detection.

[0019] Preferred embodiments of the present disclosure will be described in detail below with reference to the drawings. The same reference numerals in the description refer to the same elements having the same functions, although duplicate descriptions will not be given. One embodiment of the present disclosure will be described with reference to Figures 1 to 17. In the following description, up / down and left / right directions refer to the up / down and left / right directions shown in each figure.

[0020] Foreign Object Detection System 1 As shown in FIGS. 1 and 2 , the foreign object detection system 1 detects whether foreign objects are present in clothing and miscellaneous goods to be inspected. The clothing to be inspected includes, for example, clothing such as a shirt 90, shoes, a bag, and gloves. The miscellaneous goods include everyday items such as stationery. The foreign object detection system 1 includes a foreign object detection device 2 and a server 30. The foreign object detection device 2 includes an X-ray inspection device 2a, a first workstation 11 adjacent to the X-ray inspection device 2a that detects foreign objects based on data from the X-ray inspection device 2a, and a display / input unit 9 that also serves as a video monitor / input means. The first workstation 11 is connected to the server 30 via a communication line 33. The foreign object detection system 1 also includes a second workstation 40 that uses a predetermined tool to learn whether a foreign object is present. The second workstation 40 is connected to the server 30 via a third communication line 33 when necessary, such as for uploading.

[0021] 1 and 2, the X-ray inspection apparatus 2a includes a lower housing 3 having a horizontally elongated rectangular parallelepiped shape, and an upper housing 3d that is provided on the lower housing 3 and protrudes upward. An inlet 3a for receiving an object to be inspected is provided on the right side of the lower housing 3. An outlet 3b for sending out an object to be inspected is provided on the left side of the lower housing 3. The inlet 3a and the outlet 3b are each covered with a lead curtain (not shown).

[0022] As shown in Figures 2 and 3, a conveying path 5 is provided inside the lower housing 3, spanning the entrance 3a and the exit 3b. The conveying path 5 includes a conveying belt 5a, a conveying belt drive unit 5b, and a conveying belt control unit 5c. A first pin conveyor 5d is provided on the upstream side (the entrance 3a side) of the conveying path 5 so as to be at the same height as the conveying belt 5a. A second pin conveyor 5e is provided on the downstream side (the exit 3b side) of the conveying path 5 so as to be at the same height as the conveying belt 5a. This allows the tray 91 carrying the shirts 90 to be smoothly conveyed in the order of the first pin conveyor 5d, the conveying belt 5a, and the second pin conveyor 5e. The conveying belt control unit 5c receives commands from the control unit 7 and controls the operation of the conveying belt drive unit 5b. The conveying belt drive unit 5b drives the conveying belt 5a.

[0023] As shown in Fig. 2, the X-ray inspection apparatus 2a houses an irradiation unit 6. The irradiation unit 6 includes an X-ray source 6a that irradiates X-rays, a first light-emitting unit 6b, a first light-receiving unit 6c, a second light-emitting unit 6d, a second light-receiving unit 6e, and an irradiation control unit 6f. The irradiation control unit 6f shown in Fig. 3 receives commands from the control unit 7, including detailed settings such as the X-ray irradiation time for irradiating X-rays and the X-ray intensity, and controls the X-rays irradiated from the X-ray source 6a.

[0024] As shown in Figure 2, the X-ray source 6a emits X-rays in a first direction a, which is vertically downward, and in a second direction b, which is inclined at a predetermined angle θ (5 to 30°, for example, 20°) relative to the vertically downward. The X-ray source 6a may include separate X-ray sources for the first direction a and the second direction b, or may include only one source. Therefore, the shirt 90 is irradiated with X-rays from two directions (first direction a and second direction b). The first light-emitting unit 6b receives the X-rays in the first direction a and emits light.

[0025] The second light emitter 6d receives and emits X-rays in the second direction b. The light emitted by the first light emitter 6b and the second light emitter 6d is received by the light receivers 6c and 6e shown in FIG. 3. The light receivers 6c and 6e transmit signals (first imaging data, second imaging data) according to the amount of light received to the control unit 7. The first light emitter 6b and the first light receiver 6c correspond to a first sensor. The second light emitter 6d and the second light receiver 6e correspond to a second sensor.

[0026] 3, the X-ray inspection apparatus 2a is provided with a control unit 7, a storage unit 8, a control panel 4, and an alarm device 19. The X-ray inspection apparatus 2a is centrally managed by the control unit 7, and each component is executed based on instructions from the control unit 7. The control unit 7 has one or more arithmetic processing units such as a CPU, an MPU, a GPU, etc., and performs various information processing and control processing by reading and executing programs stored in the storage unit 8.

[0027] The memory unit 8 shown in Fig. 3 stores various information processing and control processing programs executed by the control unit 7. The control unit 7 is electrically connected to the irradiation unit 6, the transport path 5, the display / input unit 9, etc. The control panel 4 is provided on the upper surface of the upper housing 3d. The control panel 4 is equipped with an ammeter 4b, a voltmeter 4c, and various switches 4d. An operator 80 can visually check the ammeter 4b and the voltmeter 4c and can operate the various switches 4d.

[0028] As shown in FIG. 1 , the alarm device 19 is provided on the top surface of the upper housing 3 d. The alarm device 19 is, for example, a light that emits a predetermined light when it detects an X-ray leak, the door of the X-ray inspection device 2 a being left open, the cover falling off, or the lead curtains at the entrance 3 a or exit 3 b being missing. Alternatively, the alarm device 19 is a speaker that emits a predetermined voice or warning sound. Because the alarm device 19 is located in front of the operator 80, it is possible to prevent the operator 80 from overlooking or missing the light, voice, or warning sound emitted by the alarm device 19. The control unit 7 is electrically connected to the transport path 5, the irradiation unit 6, the memory unit 8, and the control panel 4.

[0029] As shown in Figure 3, the display and input unit 9 includes multiple monitors, for example, four monitors. For example, it includes a first monitor 9a, a first result monitor 9b, a second monitor 9c, and a second result monitor 9d. Each of these monitors 9a to 9d is a touch-type monitor that accepts touch operations and has a display function and an input function (a function for accepting touches on the screen). The first monitor 9a and the second monitor 9c, and the first result monitor 9b and the second result monitor 9d are electrically connected to the first workstation 11.

[0030] As shown in FIG. 1, the first monitor 9a is attached to the lower right side of the upper housing 3d via an arm-shaped attachment member 9e. The first result monitor 9b is attached to the upper right side of the upper housing 3d via an arm-shaped attachment member 9e. Therefore, the first monitor 9a and the first result monitor 9b are arranged side by side one above the other. The second monitor 9c is attached to the lower left side of the upper housing 3d via an arm-shaped attachment member 9e. The second result monitor 9d is attached to the upper left side of the upper housing 3d via an arm-shaped attachment member 9e. Therefore, the second monitor 9c and the second result monitor 9d are arranged side by side one above the other. Furthermore, each of the monitors 9a to 9d has the same screen size.

[0031] Therefore, the worker 80 can compare the images displayed on the first monitor 9a and the first result monitor 9b without moving his or her eyes significantly. Therefore, the images displayed on both monitors 9a and 9b can be easily compared. The same applies to the second monitor 9c and the second result monitor 9d. The first monitor 9a and the second monitor 9c display X-ray images based on signals from the first light receiving unit 6c and the second light receiving unit 6e. The first result monitor 9b and the second result monitor 9d display X-ray images of the shirt 90 generated by the image generation unit 13 of the control unit 12 of the first workstation 11. The worker 80 can view all four monitors 9a to 9d simultaneously.

[0032] The first monitor 9a and the first result monitor 9b do not have to be configured as two independent monitors, but may be a single monitor with the screen split into two, either vertically or horizontally. The same applies to the second monitor 9c and the second result monitor 9d. Furthermore, the first monitor 9a to the second result monitor 9d do not have to be configured as four independent monitors, but may be a single monitor with the screen split into four, vertically or horizontally.

[0033] [First Workstation 11] The first workstation 11 shown in Fig. 1 analyzes image data (first imaging data, second imaging data) from the X-ray inspection apparatus 2a. The first workstation 11 includes an analysis device main body 11a. Data is input to the analysis device main body 11a via the input functions of the first monitor 9a and the second monitor 9c. As shown in Fig. 3, the first workstation 11 includes a control unit 12, a foreign substance information memory 17, a storage unit 21, and a transceiver unit 23 (receiving means). The first workstation 11 is centrally managed by the control unit 12, and each component operates based on commands from the control unit 12.

[0034] 3 has one or more arithmetic processing units such as a CPU, an MPU, a GPU, etc., and performs various information processing and control processing by reading and executing programs stored in a storage unit 21. The control unit 12 also includes an image generation unit 13, a foreign object detection unit 14 (foreign object detection means), a foreign object display providing unit 20, an AI operation result data generating and providing means 22, and a writing means 24. The image generation unit 13 receives signals (second imaging data) from the first light receiving unit 6c and the second light receiving unit 6e of the X-ray inspection device 2a and generates an X-ray irradiation image.

[0035] The foreign object detection unit 14 shown in FIG. 3 includes a calculation unit 15 and a file storage unit 25. The calculation unit 15 has one or more processing units such as a CPU, MPU, or GPU, and reads and executes an AI inspection model file 16. The file storage unit 25 stores the AI ​​inspection model file 16. The AI ​​inspection model file 16 is an AI brain file used to detect foreign objects, and is generated by additional learning based on an initial model file generated based on standard foreign object sample image data and a group of image files selected by an operator operating the second workstation 40 from the first and second imaging data. The AI ​​inspection model file 16 is generated by the second workstation 40 using a model file creation program. The AI ​​inspection model file 16 does not include image data such as the first imaging data and the second imaging data.

[0036] The foreign object detection unit 14 detects foreign objects based on the first imaging data from the first light receiving unit 6c or the second imaging data from the second light receiving unit 6e and the AI ​​inspection model file 16 in the file storage unit 25. The foreign object information memory 17 stores the first imaging data or the second imaging data in which the foreign object detection unit 14 overlooked or overdetected a foreign object. The first imaging data or the second imaging data includes image data and AI operation result data. The AI ​​operation result data includes information such as the image file name, the AI ​​inspection model file name, the threshold value, the inspection label, the number of pixel blocks, the circumscribing rectangle of the detected block, the inspection type (0: apparel, 1: shoes), the inspection result (0: OK, 1: NG), the presence or absence of a foreign object (0: no foreign object, 1: foreign object present), the inspection speed (msec), and the inspection exclusion area (offset values ​​from the left edge, top, right edge, and bottom).

[0037] 3 , the foreign object indicator adding unit 20 adds a first foreign object indicator 92 and a second foreign object indicator (not shown) in the form of a rectangular frame that surrounds the foreign object in the first image data or the second image data in which a foreign object has been detected by the foreign object detection unit 14, and preferably flashes. These foreign object indicators 92 are each given a predetermined color (red, yellow, etc.). This allows the operator 80 to easily recognize the foreign object in the image data.

[0038] The memory unit 21 shown in FIG. 3 stores various information processing and control processing programs executed by the control unit 12. The transmission / reception unit 23 is electrically connected to the control unit 12 and the server 30 via the Internet (through the third communication line 33) and transmits and receives signals therebetween. Note that the transmission / reception unit 23 may transmit and receive signals wirelessly, not necessarily via a wired connection. If the output of the foreign object detection unit 14 indicates an oversight or overdetection of a foreign object, the user operates the display / input unit 9. Based on the signal from the display / input unit 9, the AI ​​operation result data generating and providing means 22 generates AI operation result data (presence / absence of a foreign object (0: no foreign object, 1: foreign object present), etc.) for distinguishing between the image data (first imaging data, second imaging data).

[0039] 3 writes (stores) the AI ​​inspection model file 16 in the file storage unit 25 based on a signal from the server 30 via the transmission / reception unit 23. The control unit 12 is electrically connected to the storage unit 21, the foreign substance information memory 17, and the transmission / reception unit 23. The first workstation 11 is electrically connected to the X-ray inspection device 2a and the display / input unit 9 via a first communication line 31 and a second communication line 32. The second communication line 32 is, for example, four display cables and a USB cable.

[0040] 1 , the server 30 is set up, for example, in a location (e.g., on the cloud) different from the X-ray inspection apparatus 2a via an internet line (third communication line 33). The server 30 includes only a server main body 30a. The server 30 is an information processing apparatus capable of various information processing and sending and receiving information, such as a general-purpose computer. The server 30 can save the AI ​​inspection model file 16 of the foreign object detection unit 14 stored in the first workstation 11.

[0041] The server 30 has the following three functions (first to third functions). The first function is a function for authenticating the operation of the first workstation 11. In other words, the server 30 is provided with an authentication function so that only predetermined users authorized by the administrator of the foreign object detection system 1 can use the first workstation 11.

[0042] The second function is a function for uploading the AI ​​inspection model file 16 generated by the second workstation 40 and downloading the uploaded AI inspection model file 16 to the first workstation 11 via the Internet (third communication line 33). In other words, this is a function for remotely installing the AI ​​inspection model file 16 to the first workstation 11. This allows the AI ​​inspection model file 16 to be easily updated (upgraded).

[0043] The third function allows sharing of AI operation result data from the first workstation 11. The AI ​​operation result data is a file containing information such as the image file name, AI inspection model file name, threshold, inspection label, number of block pixels, circumscribing rectangle of the detected block, inspection type (0: apparel, 1: shoes), inspection result (0: OK, 1: NG), presence / absence of foreign objects (0: no foreign object, 1: foreign object present), inspection speed (msec), and inspection exclusion area (offset values ​​from the left edge, top, right edge, and bottom). AI operation result data is generated for the first and second imaging data in a predetermined file format (e.g., CSV format). When a foreign object is overlooked or overdetected, the operator 80 can touch the first or second result monitor 9b or 9d to save a record of the overlooked or overdetected foreign object as part of the AI ​​operation result data. In other words, this third function collects AI operation result data in a predetermined file format (e.g., CSV format). This allows the operator 80 of the X-ray inspection device 2a to easily identify only the first imaging data and the second imaging data that have been determined to be oversights or overdetections. Furthermore, the number of oversights or overdetections per month, for example, can be easily ascertained. Furthermore, the history of oversights or overdetections of foreign objects can be easily ascertained. Furthermore, these results can be shared on the Internet.

[0044] [Second Workstation 40] The second workstation 40 is located in a different location (remote location) from the X-ray inspection apparatus 2a and the first workstation 11 and is an information processing device capable of various information processing and information transmission and reception, such as a general-purpose computer. The second workstation 40 has an AI model file creation program. The AI ​​model file creation program is a program that performs AI learning or AI additional learning using the first imaging data and the second imaging data to create or update an AI inspection model file, which is an AI brain file that detects foreign objects.

[0045] The initial model file is generated based on the standard foreign substance sample image data using a model file creation program by an operator operating the second workstation 40. The standard foreign substance sample image data is generated externally using an acquisition method described below and is acquired by importing the data stored in a storage medium such as a HDD. When generating the initial model file, the operator selects the standard foreign substance sample image data to be used based on his or her own experience and knowledge, and uses only the selected standard foreign substance sample image data.

[0046] The basic model file is generated by an operator operating the second workstation 40, who uses a model file creation program to update the AI ​​inspection model file as needed based on the first imaging data and the second imaging data. The first imaging data and the second imaging data are data representing overlooked or overdetected foreign objects. The first imaging data and the second imaging data are generated externally and acquired by importing data stored in a storage medium such as a HDD (in this example, the foreign object information memory 17 of the first workstation 11). When generating the basic model file, the operator selects the first imaging data and the second imaging data to be used based on their own experience and knowledge, and uses only the selected first imaging data and the second imaging data.

[0047] In this way, the AI ​​inspection model file 16 is generated using only image data based on the operator's own experience and knowledge, thereby improving the foreign substance detection accuracy of the generated AI inspection model file 16. Furthermore, the basic model file is an AI inspection model file that has been updated (learned) from the initial model file. Therefore, when foreign substances are subsequently inspected using this basic model file, the same foreign substance will not be overlooked or overdetected. Therefore, each time a new basic model file is updated and generated, the foreign substance detection accuracy improves. A new basic model file is generated periodically (e.g., once a month). In other words, the basic model file is periodically learned (relearned). The second workstation 40 and the server 30 may be integrated and operated as a single workstation.

[0048] [Method for Acquiring Standard Contaminant Sample Images (Method for Acquiring Contaminant Sample Information)] The procedure for acquiring a standard contaminant sample image will be described. A sample holder 51 is prepared in advance, as shown in FIG. 4 . The sample holder 51 is, for example, a rectangular parallelepiped melamine sponge with sides of approximately 4 cm. Alternatively, a polyhedral sponge or foamed resin may be used. An example of a contaminant sample is a broken paper clip 50. It is preferable to prioritize contaminant samples for learning. This prevents forgetting to learn. For example, contaminants that are thought to be harmful to the human body if contaminated, such as glass fragments, needles, cutter blades, safety pins, nails, and screws, are given first priority for learning. Items that are mixed in due to factory or operator error, such as scissors, pens, and packing tape, are given second priority for learning. Small contaminants, such as grains of sand, are given third priority for learning.

[0049] As shown in Figure 4, the sample holder 51 has a slit 58 formed from the first surface 52 toward the inside. Next, a clip 50 is inserted into the slit 58 of the sample holder 51. The sample holder 51 is placed on a tray 91 (a sample tray made of the same material as the tray 91). In order to irradiate each of the six surfaces of the sample holder 51 with X-rays and obtain X-ray irradiation images, the sample holder 51 is first placed on the tray 91 with the first surface 52 of the sample holder 51 facing up, i.e., with the second surface 53 facing down, and X-rays are irradiated from the first surface 52.

[0050] Similarly, the sample holders 51 are placed on the tray 91 with the second surface 53, the third surface 54, the fourth surface 55, the fifth surface 56, and the sixth surface 57 of the sample holder 51 facing up, i.e., with the first surface 52, the fourth surface 55, the third surface 54, the sixth surface 57, and the fifth surface 56 facing down, and X-rays are irradiated from the second surface 53, the third surface 54, the fourth surface 55, the fifth surface 56, and the sixth surface 57, respectively. As a result, six X-ray irradiation images are acquired, as shown in FIGS. 5 to 10 . Since the sample holder 51 itself is unlikely to be captured in the X-ray irradiation images, clear X-ray irradiation images can be acquired from six directions. These six acquired X-ray irradiation images are associated with the item (in this case, the paperclip 50) and stored in the second workstation 40.

[0051] In this way, a standard impurity sample image can be obtained. Note that a standard impurity sample image is obtained for each impurity to be detected. Furthermore, even if the impurities are of the same type, impurities with different shapes, sizes, etc. are further classified by item and stored in the second workstation 40. In this way, thousands to tens of thousands of standard impurity sample images can be created.

[0052] [Generation of AI Operation Result Data] When detecting a foreign object in the foreign object detection unit 14, a foreign object may be overlooked or overdetected. In such cases, the presence or absence of a foreign object (0: no foreign object, 1: foreign object present) is recorded as part of the AI ​​operation result data for the first imaging data and second imaging data (image data) in which the foreign object was overlooked or overdetected. This recording of the presence or absence of a foreign object (0: no foreign object, 1: foreign object present) is performed by the operator 80 touching the screens of the first result monitor 9b and the second result monitor 9d. For example, if the foreign object detection unit 14 overlooks or overdetects a foreign object, the operator 80 touches the screens of the first result monitor 9b and the second result monitor 9d. Then, the AI ​​operation result data generating and assigning means 22 assigns a part of the AI ​​operation result data assigned to the image data corresponding to the touched screen.

[0053] The image data to which this AI operation result data has been added is stored in a foreign substance information memory 17. Because this foreign substance information memory 17 is, for example, an HDD, this foreign substance information memory 17 can be transported to and connected to the second workstation 40, which is an external device. The image data to which the AI ​​operation result data has been added is read by the second workstation 40, and only that image data selected by the operator of the second workstation 40 from the read image data is used (picked up) to generate a basic model file.

[0054] [Foreign Object Detection Procedure] Next, the foreign object detection procedure will be described with reference to FIG. 13 . As an example of foreign object detection, detection of a foreign object mixed in a shirt 90 will be described. The worker 80 starts up the foreign object detection device 2 (S1). Note that the server 30 is normally always running. Next, the worker 80 accesses the server 30 from the first workstation 11 and obtains authentication (S2). The worker 80 has been given an ID, password, etc. for access in advance through a subscription contract or the like.

[0055] During initial setup or when an update is required, the operator 80 downloads the latest AI inspection model file 16 from the server 30 to the first workstation 11 (S3). The downloaded latest AI inspection model file 16 is written and stored by the writing means 24 in the file storage unit 25 of the foreign object detection unit 14 of the first workstation 11. If an update is not required, step S3 is skipped. Next, the operator 80 operates the control panel 4 to start the conveyor belt 5a (S4). Next, the operator 80 places the tray 91 containing the shirts 90 on the first pin conveyor 5d (S5).

[0056] 1 and 13, next, the worker 80 places the tray 91 from the first pin conveyor 5d onto the conveyor belt 5a (on the conveyor path) (S6). Next, the worker 80 carries the tray 91 carrying the shirt 90 through the entrance 3a of the X-ray inspection device 2a (S7). When the tray 91 arrives below the X-ray source 6a on the conveyor belt 5a, X-rays are irradiated onto the shirt 90 from the first direction a and the second direction b (S8). The irradiated X-rays pass through the shirt 90 and reach the light-emitting units 6b and 6d below the conveyor belt 5a.

[0057] The light-emitting elements 6b and 6d each emit light in accordance with the amount of X-rays that reach them, and the first light-receiving element 6c and the second light-receiving element 6e each receive the emitted light.

[0058] The first monitor 9a displays an X-ray irradiation image (first image) of the shirt 90 from the first direction a based on a signal corresponding to the amount of light received from the first light receiving unit 6c (S9). The clip 50 is displayed as a foreign object in this image. This allows the worker 80 to view the X-ray irradiation image in which the clip 50 is displayed. This allows the worker 80 to recognize the clip 50 mixed in with the shirt 90 (S10).

[0059] Similarly, the second monitor 9c displays an X-ray irradiation image (second image) of the shirt 90 from the second direction b based on a signal corresponding to the amount of light received from the first light receiving unit 6c (S9). This image displays the clip 50 tilted at 20 degrees as a foreign object. This allows the worker 80 to view the X-ray irradiation image in which the clip 50 tilted at 20 degrees is displayed. This allows the worker 80 to recognize the clip 50 tilted at 20 degrees that has been mixed into the shirt 90 (S10).

[0060] At the same time, the image generating unit 13 of the control unit 12 generates X-ray images of the shirt 90 from the light received by the light receiving units 6c and 6e. The foreign object detecting unit 14 of the control unit 12 then detects the presence or absence of a foreign object from the X-ray images. In this case, the foreign object detecting unit 14 recognizes the clip 50, so the foreign object indicator providing unit 20 provides a first foreign object indicator 92 and a second foreign object indicator surrounding the clip 50.

[0061] The foreign object detection unit 14 obtains an image represented by pixels by dividing the first image data or the second image data into regions, and detects a foreign object when the number of consecutive pixels detected as a foreign object in the image exceeds a threshold. In the image represented by the obtained pixels, pixels that are connected vertically, horizontally, or diagonally are considered to be a single block. For example, the threshold is set to "5." The threshold can be set to any value.

[0062] Then, a cluster of pixels 60 as shown in FIG. 11 has four pixels 60. That is, the number of pixels 60 is four. In this case, the foreign object detection unit 14 determines that there is no foreign object because the number of pixels 60 is less than the threshold. On the other hand, a cluster of pixels 60 as shown in FIG. 12 has seven pixels 60. That is, the number of pixels 60 is seven. In this case, the foreign object detection unit 14 determines that there is a foreign object because the number of pixels 60 is greater than the threshold. In this way, the foreign object detection unit 14 detects the presence or absence of a foreign object. In addition to the above determination, the foreign object detection unit 14 may also determine whether there is a foreign object based on the shape of the cluster of pixels 60.

[0063] An X-ray irradiation image of the shirt 90 taken from the first direction a is displayed on the first result monitor 9b (S9). The clip 50 is displayed as a foreign object in this image. A first foreign object indicator 92 is displayed around the clip 50 (see FIG. 14). This allows the operator 80 to view the X-ray irradiation image with the first foreign object indicator 92 displayed. The operator 80 therefore notices the blinking display of the first foreign object indicator 92 and can reliably identify the clip 50 mixed in with the shirt 90 (S10). The X-ray irradiation image displayed on the first result monitor 9b is the X-ray irradiation image displayed on the first monitor 9a with the first foreign object indicator 92 displayed.

[0064] Similarly, the second result monitor 9d displays an X-ray irradiation image of the shirt 90 from the second direction b (S9). In this image, the clip 50 is displayed as a foreign object tilted at 20 degrees. The clip 50 is surrounded by a second foreign object indicator. Therefore, the operator 80 can see the X-ray irradiation image with the second foreign object indicator. The operator 80 notices the blinking second foreign object indicator and can reliably identify the clip 50 mixed in with the shirt 90 (S10). The X-ray irradiation image displayed on the second result monitor 9d is the X-ray irradiation image displayed on the second monitor 9c with the second foreign object indicator added.

[0065] Next, the tray 91 placed on the conveyor belt 5a is discharged from the exit 3b of the foreign matter detector 2 (S11). The worker 80 takes out the shirt 90 and checks for any foreign matter (S12). Through this series of operations, the worker 80 can detect any foreign matter that has become mixed into the shirt 90.

[0066] Furthermore, during the above-described series of operations, it is possible that even though the clip 50 is displayed as a foreign object on the first monitor 9a, the first foreign object display 92 is not displayed on the first result monitor 9b. In other words, it is possible that the foreign object detection unit 14 has overlooked the foreign object. In this case, the operator 80 touches (inputs) the screen of the first result monitor 9b as described above. The AI ​​operation result data generation means 22 of the first workstation 11 then generates an AI operation result data file for the first imaging data corresponding to this touch. This generated AI operation result data file and the first imaging data are stored in the foreign object information memory 17.

[0067] Conversely, it is also possible that the foreign object detection unit 14 mistakenly detects a foreign object as a foreign object even though the shirt 90 does not contain any foreign object, and the foreign object indicator assigning unit 20 assigns the first foreign object indicator 92. In other words, it is also possible that the foreign object detection unit 14 overdetects a foreign object. In this case, the operator 80 touches (inputs) the screen of the first result monitor 9b as described above. The AI ​​operation result data assigning means 22 of the first workstation 11 then generates AI operation result data for the first imaging data corresponding to this touch. The generated AI operation result data is assigned to the first imaging data, and the first imaging data is stored in the foreign object information memory 17. This stored first imaging data is periodically (e.g., monthly) transported to the second workstation 40.

[0068] The first monitor 9a and the second monitor 9c display X-ray irradiation images with an X-ray irradiation angle from the X-ray source 6a that differs by 20°. Therefore, a foreign object that is not visible on the first monitor 9a may be visible on the first result monitor 9b, and vice versa. This prevents foreign objects from being overlooked. The same applies to the first result monitor 9b and the second result monitor 9d.

[0069] Furthermore, in the X-ray inspection device 2a, the voltage related to X-ray irradiation is set to a predetermined value. Changing this voltage changes the grayscale range, for example, 256 levels. When changing the threshold, a voltage that minimizes the difference due to voltage is preferable. Changing the brightness does not change the grayscale range. Therefore, the brightness of the screen is not adjusted by adjusting the voltage. For example, at or below a first predetermined voltage, the difference in grayscale value between a non-metallic foreign object such as plastic and an overlapping portion of clothing becomes small. At a second predetermined voltage higher than the first predetermined voltage, the image may white out depending on the brightness adjustment. Furthermore, high X-ray irradiation intensity may shorten the life of the X-ray sensor.

[0070] The blinking status and color of the first foreign object indicator 92 and the second foreign object indicator can be set arbitrarily to prevent eye strain on the operator 80. The light, audio, and warning sound issued by the alarm device 19 can be set to ON or OFF. Area division is used to determine foreign objects by the foreign object detection unit 14. A threshold is set for the number of adjacent pixels detected as foreign objects vertically, horizontally, and diagonally, and the foreign object detection unit 14 performs foreign object determination processing using that number of pixels (threshold).

[0071] The foreign object detection device 2 allows selection of objects such as clothing and miscellaneous goods. It may be configured to read out an AI inspection model file 16 according to the selected object. There may be one or more objects. An AI may be selected for each inspection object. Changing the contrast of the X-ray irradiation image makes it easier for the worker 80 to see. The boundary of the X-ray line sensor can also be adjusted. The server 30 manages files of various X-ray irradiation images and has a function for creating reports for X-ray inspections. AI determines the position information of foreign objects. Foreign objects are identified on the lower screen of the two screens, the first monitor 9a and the first result monitor 9b. These images are in black and white, making it easy to grasp the shape, etc.

[0072] A deadline may be set for the authentication of the subscription contract of the first workstation 11. Also, a deadline for maintenance may be set for the first workstation 11. The authentication system, the server 30, and the basic model file can be upgraded.

[0073] The hexahedral sample holder 51 (resin holder) is a rectangular parallelepiped, measuring approximately 3.2 cm x 4 cm x 4 cm. A slit 58 is made in the sample holder 51, and a foreign object is inserted into it. By irradiating each of the six sides with X-rays, data can be obtained from six directions. The polyethylene or urethane tray and the sample holder 51 appear similar in X-ray images. Therefore, only a faint line appears, and the AI ​​learning is not affected. The foreign object to be detected by X-ray inspection is created, or the actual object is procured and embedded in the sample holder 51. By placing a foreign object in the sample holder 51 and photographing it, six different shapes can be photographed from a single foreign object (foreign objects with different areas can be obtained).

[0074] When retraining the basic model file, the AI ​​learns and corrects the detection failures, foreign object detections, and normal objects from actual work. During correction, to check the accuracy of the AI ​​learning, the created foreign objects with a high priority are used to check for a decrease in detection accuracy. The priority of the created foreign objects is ranked according to their frequency at the site and the degree of danger to the human body. A foreign object list is created for the initial model file only, based on the AI ​​learning priority. An initial model is created that is trained on the basic NG (general foreign objects that may be mixed in) and OK (clothing with no foreign objects at all). For example, approximately 40,000 images are inspected, and any overdetections or missed detections that occur are additionally trained. The initial model learning is ranked. Scissors, pens, and packing tape are given high rankings. Sand grains used in processing jeans are given a low ranking. The program is run using the training data.

[0075] When training the AI's initial model file, standard foreign objects that are highly likely to cause harm to humans when mixed in are trained. This includes even those that occur in factory accidents. The AI ​​is trained in the order in which foreign objects accidentally mixed in by workers are learned. The learning order is determined based on the priority of the foreign objects to be detected, preventing high-priority foreign object detection objects from being forgotten to be trained. If the AI's learning procedure and priority include high and medium priorities, the high priority has already been trained by the AI. To eliminate any impact on the high priority, the learning conditions for the medium priority are changed, and the detection accuracy is improved. By repeating these operations, the AI's initial model file is created.

[0076] The basic model file is a file created by using the initial model file in actual inspection work and revising the initial model file using the model file creation program. The basic model file is generated based on learning data selected by the operator of the model file creation program from the first and second image data generated during actual inspection work, and the AI ​​is trained on the data for actually occurring contaminants to improve accuracy. Assuming an oversight occurs during actual inspection work, contaminant A is overlooked in the image (overseen image). 1. The AI ​​is trained using several normal images of the actual object and an image of contaminant A (overseen image). 2. After training, it is confirmed whether contaminant A can be detected. 3. The detection accuracy after AI training is confirmed using high-priority contaminant samples. To confirm whether there is any decrease in accuracy for contaminants other than those learned, step 3 can be performed using high-priority artificial contaminants, and can be used to check for malfunctions after AI training.

[0077] In additional learning, if an object is overlooked in both the first-direction X-ray image and the second-direction X-ray image, it is considered a "missed object." The current inspection model has a fairly high detection accuracy for small granular foreign objects. However, on the other hand, larger foreign objects (nails, screws, cutter chips) may be overlooked depending on their shape. This is because clothing accessories (buttons, hooks, zippers, buckles) that are learned as normal items often have a shape similar to that of larger foreign objects depending on the angle from which they are captured.

[0078] Large foreign objects can be seen at a glance on the first monitor 9a and the second monitor 9c, so even if they are overlooked, they are not used as "overlooked" images for learning purposes. This approach is based on AI's ability to detect small foreign objects that are difficult to notice. Color-coding by item is also possible. For example, it can be used to detect the number of buttons on a shirt 90 to prevent people from forgetting to button them. It can also be used to check whether nails inside high heels are broken. In this case, it is sufficient to determine whether the nails are normal, and this can be determined by setting a threshold value.

[0079] Foreign objects / contamination refers to unwanted objects that have been mixed into the product being inspected. Over-detection occurs when the AI ​​detects a foreign object in an area that is not actually a problem. In other words, it refers to the AI ​​detecting a foreign object in an area that is not actually a problem. A missed detection occurs when the AI ​​fails to detect a foreign object even though it is present. An AI model (including the program, foreign object information, and thresholds) is a stored version of the AI's brain that has already been trained. It can also store files that output the results of using a learning tool to learn "foreign object" and "normal" areas using X-ray images. Internally, it also remembers which part of which image was trained.

[0080] The foreign object detection system 1 described above detects the presence or absence of a foreign object based on X-rays irradiated onto the shirt 90 from two directions, the first direction a and the second direction b. Therefore, compared to detecting the presence or absence of a foreign object based on X-rays irradiated from a single direction, the system can reduce the risk of overlooking or overdetecting a foreign object. Furthermore, an X-ray irradiation image of the shirt 90 from the first direction a and an X-ray irradiation image in which the first foreign object indicator 92 is superimposed on the X-ray irradiation image are displayed side by side, allowing the operator 80 to compare these two X-ray irradiation images. Therefore, by comparing the first monitor 9a and the first result monitor 9b, the operator 80 can easily determine whether the foreign object detection unit 14 has overlooked or overdetected a foreign object. The same applies to the X-ray irradiation image of the shirt 90 from the second direction b. Furthermore, the foreign object detection unit 14 includes an initial model file or a basic model file that is an updated version of the initial model file. The basic model file is trained (updated) based on the first or second imaging data in which a foreign object was overlooked or overdetected, thereby preventing the same foreign object from being overlooked or overdetected again in the future. This improves the accuracy of detecting the presence or absence of foreign objects. Furthermore, training can be simplified.

[0081] Furthermore, the AI ​​inspection model file 16 is generated by the second workstation 40 using a model file creation program. Therefore, an AI inspection model file with high detection accuracy can be generated. Furthermore, because the AI ​​inspection model file 16 is generated by the second workstation 40, the involvement of the user of the foreign object detection device 2 in generating the AI ​​inspection model file 16 can be reduced.

[0082] Furthermore, the first foreign object display 92 or the second foreign object display is a frame that surrounds the foreign object and is displayed in a flashing manner, which reduces the likelihood that the operator 80 monitoring the first result monitor 9b or the second result monitor 9d will overlook the foreign object.

[0083] Furthermore, the foreign object detection unit 14 obtains an image represented by pixels by dividing the X-ray irradiation image from the first direction a or the X-ray irradiation image from the second direction b into regions. The foreign object detection unit 14 then detects a foreign object when the number of consecutive pixels detected as foreign objects in the image exceeds a threshold. This allows the foreign object detection unit 14 to easily detect the presence or absence of a foreign object.

[0084] One aspect of the present disclosure relates to a foreign substance information acquisition method for acquiring standard foreign substance sample image data for use in a foreign substance detection device 2 for clothing and miscellaneous goods. In the foreign substance information acquisition method, a foreign substance sample (e.g., a paperclip 50) is inserted into a hexahedral resin sample holder 51. Next, the sample holder 51 is placed with its first surface 52 facing downwards on a resin tray 91 that is placed on the transport path 5 and on which a shirt 90 is placed. X-rays are then irradiated onto the sample holder 51. Next, the sample holder 51 is placed on the tray 91 with its second surface facing downwards. X-rays are then irradiated onto the sample holder 51. Next, the sample holder 51 is placed on the tray 91 with its third surface facing downwards. X-rays are then irradiated onto the sample holder 51. Next, information on the foreign substance sample from at least three directions is acquired based on signals from both light-receiving units 6c and 6e that receive the X-rays. Therefore, by utilizing at least three surfaces of the hexahedron, foreign substance sample information is stored from at least three perspectives. Therefore, information on the foreign substance sample can be obtained quickly and easily, and the accuracy of foreign substance detection based on the information on the foreign substance sample is improved.

[0085] Furthermore, the foreign object detection system 1 including the foreign object detection device for clothing and miscellaneous goods has a server 30 to which the AI ​​inspection model file 16 is uploaded from the second workstation 40. The writing means 24 downloads the AI ​​inspection model file 16 from the server 30 and writes the AI ​​inspection model file 16 to the file storage unit 25 of the foreign object detection device 2 for clothing and miscellaneous goods. Therefore, when the AI ​​inspection model file 16 stored in the server 30 is updated, the AI ​​inspection model file 16 stored in the file storage unit 25 of the foreign object detection device 2 can also be updated. Therefore, the AI ​​inspection model file 16 stored in the foreign object detection device 2 can be updated via the server 30 without having to go to the installation location of the foreign object detection device 2. This makes it possible to easily update the AI ​​inspection model file 16.

[0086] Although the present disclosure has been described with reference to the above structure, it will be apparent to those skilled in the art that many alternatives, modifications, and variations are possible without departing from the scope of the present invention. Therefore, the present disclosure may include all alternatives, modifications, and variations that do not depart from the spirit and scope of the appended claims. For example, the present disclosure is not limited to the specific structure described above, and may be modified as follows:

[0087] The embodiments described in detail above with reference to the accompanying drawings are exemplary of the present disclosure and are not intended to limit the present disclosure. The detailed description teaches those skilled in the art how to make, use, and / or implement various aspects of the present teachings, but is not intended to limit the scope of the present disclosure. Furthermore, each of the additional features and teachings described above may be applied and / or used separately or together with other features and teachings to provide an improved foreign object detection device 2 for clothing and accessories.

[0088] In the embodiment, the predetermined angle θ has been described as, for example, 20°. Alternatively, it may be any angle in the range of 5 to 30°. Also, in the embodiment, the X-ray sources 6a in the first direction a and the second direction b have been described as having the same configuration (composed of an integrated member). Alternatively, these X-ray sources 6a may have different configurations. Also, in the embodiment, the sample holder 51 has been described as being hexahedral. Alternatively, the sample holder 51 may be any polyhedral shape as long as it is an polygonal shape.

[0089] Furthermore, the foreign object detection device 2 in the above embodiment includes an X-ray inspection device 2a, a first workstation 11, and a display / input unit 9. The X-ray inspection device 2a and the first workstation 11 are configured separately. Alternatively, the X-ray inspection devices 2a and 11 may have the same configuration. For example, if they have different configurations, the foreign object detection system 1 can be constructed by retrofitting the first workstation 11 to an existing X-ray inspection device 2a. If they have the same configuration, the first workstation 11 can be incorporated into the X-ray inspection device 2a. This eliminates the need to provide the first workstation 11 separately, resulting in a simple configuration.

[0090] In the embodiment, the procedure for acquiring a standard foreign substance sample image is to place the sample holder 51 on the tray 91 with the first surface 52 to the sixth surface 57 of the sample holder 51 facing up, and to irradiate X-rays from the first surface 52 to the sixth surface 57, respectively. Alternatively, the sample holder 51 may be placed on the tray 91 with the first surface 52 to the sixth surface 57 of the sample holder 51 facing down, and to irradiate X-rays from the opposite surfaces, respectively. Furthermore, the number of directions in which X-rays are irradiated is not limited to six, and may be at least three.

[0091] In the above embodiment, if the output of the foreign object detection unit 14 indicates an oversight or overdetection of a foreign object, the operator 80 touches the screen of the first result monitor 9b or the second result monitor 9d. Image data corresponding to the touched screen is stored in the foreign object information memory 17 by the AI ​​operation result data generating and assigning means 22. That is, these processes are performed in the first workstation 11. Alternatively, if the output of the foreign object detection unit 14 indicates an oversight or overdetection of a foreign object, the processes may be performed in the second workstation 40. In this case, the image data (first image data and second image data) acquired by the first workstation 11 is sent to the second workstation 40 via a line or a recording medium. The operator uses various devices of the second workstation 40 (e.g., input means, monitor, AI operation result data generating and assigning means, foreign object information memory, etc.) to store the image data in the foreign object information memory, etc. Based on the information stored in the foreign object information memory, the basic model file is updated from the initial model file to an AI inspection model file. This basic model file is sent from the second workstation 40 to the first workstation 11 via a line or a recording medium.

Claims

1. A foreign object detection device for clothing and miscellaneous goods that detects foreign objects contained in clothing and miscellaneous goods, comprising: a transport path for transporting the clothing and miscellaneous goods; an X-ray source that irradiates X-rays from a first direction and a second direction onto the clothing and miscellaneous goods on the transport path; an upstream sensor that is provided below the transport path and receives the X-rays from the first direction; a downstream sensor that is provided downstream of the upstream sensor and below the transport path and receives the X-rays from the second direction; a file storage unit that stores an AI inspection model file; a foreign object detection unit that uses the AI ​​inspection model file to detect a foreign object based on first imaging data from the upstream sensor or detects a foreign object based on second imaging data from the downstream sensor; a first monitor that displays a first image based on the first imaging data; and a first result monitor that is provided in parallel with the first monitor and displays a first foreign object display related to the location of a foreign object detected in the first imaging data superimposed on the first image. a foreign object detection unit for detecting a foreign object detected in the second imaging data by superimposing the second image on the second image; an input unit that is operated by an operator when the foreign object detection unit overlooks or overdetects a foreign object; and a foreign object information memory that stores information about the overlooked or overdetected foreign object based on a signal from the input unit, wherein the AI ​​inspection model file is an initial model file generated based on standard foreign object sample image data or a basic model file obtained by updating the initial model file based on information stored in the foreign object information memory, the first imaging data, and the second imaging data, and the first monitor and the first result monitor are arranged side by side on the upstream side of a main body that accommodates the X-ray source, and the second monitor and the second result monitor are arranged side by side on the downstream side of the main body.

2. A foreign object detection device for clothing and miscellaneous goods that detects foreign objects contained in clothing and miscellaneous goods, comprising: a transport path for transporting the clothing and miscellaneous goods; an X-ray source that irradiates the clothing and miscellaneous goods on the transport path with X-rays from a first direction and a second direction; a first sensor that receives the X-rays from the first direction; a second sensor that receives the X-rays from the second direction; a file storage unit that stores an AI inspection model file; a foreign object detection unit that uses the AI ​​inspection model file to detect a foreign object based on first imaging data from the first sensor or detects a foreign object based on second imaging data from the second sensor; a first monitor that displays a first image based on the first imaging data; a first result monitor that is juxtaposed to the first monitor and that displays a first foreign object indication related to the location of a foreign object detected in the first imaging data superimposed on the first image; and a second monitor that is provided in a location visible from a location where the first monitor is visible and that displays a second image based on the second imaging data. the foreign object detection unit detects a foreign object by detecting a location of the foreign object detected in the second image data and displaying the foreign object information on the second image data; an input means operated by an operator when the foreign object detection unit overlooks or overdetects a foreign object; and a foreign object information memory for storing information about the overlooked or overdetected foreign object based on a signal from the input means, wherein the AI ​​inspection model file is an initial model file generated based on standard foreign object sample image data or a basic model file obtained by updating the initial model file based on information stored in the foreign object information memory, the first image data, and the second image data, and wherein the standard foreign object sample image data is generated by inserting a foreign object sample into a polyhedral sample holder made of resin, placing the sample holder with a first surface facing down on a resin tray on which clothing and miscellaneous goods are placed and which is placed on the transport path, or on a sample tray made of the same material as the tray, and irradiating the sample holder with X-rays, placing the sample holder on the tray or the sample tray with the second surface facing down, and irradiating the sample holder with X-rays;a foreign body detection device for clothing and miscellaneous goods, the device placing the sample holder on the tray or the sample tray with the third surface facing down, irradiating the sample holder with X-rays, and acquiring information on the foreign body sample in at least three directions as foreign body information based on a signal from a sensor that receives the X-rays.

3. The foreign object detection device for clothing and miscellaneous goods as described in claim 1 or 2, wherein the AI ​​inspection model file is generated by an external device using a model file creation program.

4. A foreign object detection device for clothing and miscellaneous goods as described in any one of claims 1 to 3, wherein the first foreign object indication or the second foreign object indication is a frame that surrounds the foreign object and is displayed in a flashing manner.

5. The foreign object detection device for clothing and miscellaneous goods as described in any one of claims 1 to 4, wherein the detection of the foreign object by the foreign object detection unit based on the first image capture data or the detection of the foreign object based on the second image capture data is performed by dividing the first image capture data or the second image capture data into regions to obtain an image represented by pixels, and the presence of the foreign object is determined when the number of consecutive pixels detected as the foreign object in the image exceeds a threshold value.

6. A foreign object detection system including the foreign object detection device for clothing and miscellaneous goods described in any one of claims 1 to 5, comprising: a server to which the AI ​​inspection model file is uploaded from an external device; and a writing means for downloading the AI ​​inspection model file from the server and writing the AI ​​inspection model file into the file memory unit of the foreign object detection device for clothing and miscellaneous goods.

7. A foreign object detection system for detecting foreign objects contained in clothing and miscellaneous goods, comprising: an X-ray inspection device for irradiating the clothing and miscellaneous goods being conveyed with X-rays to obtain imaging data; foreign object detection means for detecting foreign objects from the imaging data using an AI inspection model file; a video monitor for displaying an image based on the imaging data; a result monitor for displaying a foreign object indication related to the location of a foreign object detected from the imaging data superimposed on the image; input means operated by an operator when the foreign object detection section has overlooked a foreign object or overdetected a foreign object; and a foreign object information memory for storing information on the overlooked foreign object or overdetected foreign object based on a signal from the input means, wherein the AI ​​inspection model file is an initial model file generated by an external device, or a basic model file obtained by updating the initial model file by the external device, and the basic model file is generated by updating the initial model file based on the information stored in the foreign object information memory and the imaging data, the X-ray inspection device has an X-ray source which irradiates the transported clothing and miscellaneous goods with X-rays from a first direction and a second direction, an upstream sensor which receives X-rays from the first direction, and a downstream sensor which is provided downstream of the upstream sensor in the transport direction and receives X-rays from the second direction; the imaging data has first imaging data obtained based on X-rays from the upstream sensor and second imaging data obtained based on X-rays from the downstream sensor; the video monitor has a first video monitor which displays the image based on the first imaging data, and a second video monitor which displays the image based on the second imaging data; the result monitor has a first result monitor which displays a first foreign object indication related to the location of a foreign object detected in the first imaging data by superimposing it on the first image, and a second result monitor which displays a second foreign object indication related to the location of a foreign object detected in the second imaging data by superimposing it on the second image; the first video monitor and the first result monitor are provided in parallel upstream of the X-ray inspection device, The second image monitor and the second result monitor are disposed in parallel downstream of the X-ray inspection apparatus in a foreign object detection system.

Citation Information

Patent Citations

  • A equipment for automated inspection puts in order case finished shoes

    CN207198062U

  • Device for detecting broken needles of shoes and clothes based on artificial intelligence technology

    CN213779923U

  • Radioscopic unit

    JP1994265485A

  • X-ray foreign matter inspecting apparatus

    JP1999023494A

  • X-ray foreign matter detector

    JP2011196778A

Cited By

  • X-ray foreign matter detector with single radiation source for secondary screening

    CN120900970A