JP7250301B2 - Inspection device, inspection system, inspection method, inspection program and recording medium - Google Patents

Description

One aspect of the present invention relates to an inspection device, an inspection system, an inspection method, an inspection program, and a recording medium.

As a conventional inspection device, for example, the device described in Patent Document 1 is known. The inspection apparatus described in Patent Document 1 includes an X-ray irradiation unit that irradiates an article with X-rays and an X-ray detection unit that detects the X-rays, and X-rays generated based on X-ray detection results. Image processing is performed on the image to inspect the article.

WO2006/001107

2. Description of the Related Art In recent years, an inspection apparatus uses a learned model generated by machine learning to inspect a foreign object or the like. In image processing based on a trained model, all pixels of an image to be inspected are processed. As a result, the inspection takes time, and the throughput of the inspection apparatus may decrease.

An object of one aspect of the present invention is to provide an inspection apparatus, an inspection system, an inspection method, an inspection program, and a recording medium capable of improving processing performance.

An inspection apparatus according to one aspect of the present invention includes an irradiation unit that irradiates an article with an electromagnetic wave, a detection unit that detects the electromagnetic wave that has passed through the article, and a transmitted image created from the detection results of the detection unit. a processing unit that performs a reduction process to create a reduced image; an acquisition unit that obtains a processing result of processing the reduced image created by the processing unit using a trained model generated by machine learning; and an inspection unit that inspects the article based on the processing result acquired by the acquisition unit.

An inspection apparatus according to one aspect of the present invention uses a reduced image for processing using a trained model. As a result, in the inspection apparatus, the processing load due to the learned model is reduced, so that the processing result can be obtained quickly. Therefore, it is possible to improve the throughput of the inspection apparatus.

In one embodiment, the inspection unit may inspect the article based on the transmission image, and inspect the article based on the determination result of the inspection and the processing result. In this configuration, an article is inspected based on the result of determination based on the transmission image, for example, based on threshold determination of luminance values, and the result of processing. As described above, the inspection apparatus inspects an article based on two results, so that inspection accuracy can be improved.

In one embodiment, an output unit that outputs a sorting signal to a sorting device that sorts articles when an inspection unit detects an abnormality in an article; A setting unit for setting the reduction ratio of the reduced image based on the first time until the image reaches the minute device and the second time required for processing by the learned model. The time required for processing by the trained model changes according to the number of pixels in the reduced image, longer when the number of pixels is large, and shorter when the number of pixels is small. If the number of pixels is large, the processing may take a long time, and the processing may not be completed by the time the articles arrive at the sorting device. On the other hand, if the number of pixels is too small, the precision of processing by the trained model may decrease. The inspection apparatus sets the reduction ratio of the reduced image based on the first time and the second time. As a result, in the inspection device, it is possible to ensure the sorting by the sorting device, and to suppress the deterioration of the processing accuracy of the learned model.

In one embodiment, a generation unit that generates a trained model by machine learning using a teacher image may be provided. With this configuration, a trained model suitable for inspecting articles can be generated.

In one embodiment, a learning unit that processes the reduced image using the trained model generated by the generation unit may be provided, and the acquisition unit may acquire the processing result executed by the learning unit. In this configuration, the acquiring unit acquires the processing result obtained by inputting the reduced image to the trained model in the learning unit. Therefore, the inspection apparatus can more reliably perform inspections based on the processing results.

In one embodiment, the generator may generate a trained model including a neural network. With this configuration, the learned model can be made appropriate, and the inspection accuracy of the article can be improved.

In one embodiment, the inspection unit may inspect whether or not the article contains a foreign object. When an article includes a plurality of products, the products may overlap each other. In this case, it becomes difficult to distinguish between the portion where the products are overlapped and the foreign object based on the luminance value of the transmission image. Since the inspection device performs inspection based on the processing result of the processing by the learned model, it is possible to distinguish between the overlapping portions of the products and the foreign matter. Therefore, the inspection device is particularly effective in inspecting whether or not an article contains foreign matter.

In one embodiment, the irradiation section may irradiate the article with X-rays, and the detection section may detect the X-rays transmitted through the article. With this configuration, even if the article is wrapped, the article can be inspected without being affected by the packaging material or the printing applied to the packaging material.

An inspection system according to one aspect of the present invention is an inspection system comprising an inspection device and a server communicably connected to the inspection device, wherein the inspection device includes an irradiation unit that irradiates an article with electromagnetic waves. a detection unit that detects electromagnetic waves that have passed through an article; a transmission image created from the detection results of the detection unit is subjected to pixel number reduction processing to create a reduced image; and image information related to the reduced image is sent to a server. a processing unit that transmits image information, an acquisition unit that acquires from the server the result of processing by a trained model generated by machine learning in response to the image information being transmitted from the processing unit to the server, and an acquisition unit that acquires the result an inspection unit that inspects the article based on the processed result, and the server includes a learning unit that processes the reduced image using the trained model based on the image information transmitted from the inspection device. , and a communication unit that transmits the processing result to the inspection device.

An inspection system according to one aspect of the present invention uses a reduced image for processing using a trained model. As a result, in the inspection system, the processing load due to the learned model is reduced, so that processing results can be obtained quickly. Therefore, in the inspection system, it is possible to improve the throughput.

In one embodiment, a generation unit that generates a trained model by machine learning using a teacher image may be provided. With this configuration, a trained model suitable for inspecting articles can be generated.

An inspection method according to one aspect of the present invention is an inspection method performed by an inspection apparatus including an irradiation unit that irradiates an article with an electromagnetic wave and a detection unit that detects the electromagnetic wave that has passed through the article, wherein the detection unit A processing step of reducing the number of pixels of a transparent image created from the detection result to create a reduced image, and a learned model generated by machine learning for the reduced image created in the processing step. and an inspection step of inspecting the article based on the processing result acquired in the acquisition step.

In an inspection method according to one aspect of the present invention, a reduced image is used for processing using a trained model. As a result, in the inspection method, the processing load due to the learned model is reduced, so that processing results can be obtained quickly. Therefore, in the inspection method, it is possible to improve the throughput.

An inspection program according to one aspect of the present invention is a computer of an inspection apparatus comprising an irradiation unit for irradiating an article with electromagnetic waves, and a detection unit for detecting electromagnetic waves that have passed through the article. A processing unit that performs processing to reduce the number of pixels on a transparent image to create a reduced image, and a processing that performs processing using a trained model generated by machine learning on the reduced image created by the processing unit. An acquisition unit that acquires the result and an inspection unit that inspects the article based on the processing result acquired by the acquisition unit function.

An inspection program according to one aspect of the present invention uses a reduced image for processing by a trained model. As a result, in the inspection program, the processing load due to the learned model is reduced, so it is possible to quickly obtain processing results. Therefore, in the inspection program, it is possible to improve the throughput.

A recording medium according to one aspect of the present invention is a computer that records an inspection program to be executed by a computer of an inspection apparatus that includes an irradiation unit that irradiates an article with an electromagnetic wave and a detection unit that detects the electromagnetic wave that has passed through the article. A readable recording medium comprising: a processing unit for reducing the number of pixels of a transmission image created from the detection result of the detection unit by a computer to create a reduced image; an acquisition unit that acquires a processing result of performing processing on a reduced image using a trained model generated by machine learning; an inspection unit that inspects an article based on the processing result acquired by the acquisition unit; I have recorded a test program that works as

A recording medium according to one aspect of the present invention records the inspection program. The inspection program uses the reduced image for processing by the trained model. As a result, in the inspection program, the processing load due to the learned model is reduced, so it is possible to quickly obtain processing results. Therefore, the inspection program recorded on the recording medium can improve the throughput.

According to one aspect of the present invention, it is possible to improve the processing capability.

FIG. 1 is a diagram showing an inspection system according to one embodiment. FIG. 2 is a configuration diagram of the inside of the shield box shown in FIG. FIG. 3 is a diagram illustrating an example of a configuration of a control unit of the inspection apparatus; FIG. 4A is a diagram showing a transparent image, and FIG. 4B is a diagram showing a reduced image. FIG. 5 is a diagram showing a sorting device. FIG. 6 is a diagram showing an example of a neural network. FIG. 7 is a flow chart showing an example of the operation of the inspection system. FIG. 8 is a diagram showing an inspection program.

Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings. In the description of the drawings, the same or corresponding elements are denoted by the same reference numerals, and overlapping descriptions are omitted.

As shown in FIG. 1, the X-ray inspection system 1 includes an X-ray inspection device 10, a sorting device 30, and a server 50. The X-ray inspection system 1 inspects the article A using a learned model generated by machine learning.

The X-ray inspection apparatus 10 includes an apparatus body 11, a support leg 12, a shield box 13, a transport section 14, an X-ray irradiation section (irradiation section) 15, an X-ray detection section (detection section) 16, and a display. An operation unit 17 and a control unit 18 are provided.

The X-ray inspection apparatus 10 generates a transmission image G1 (see FIG. 4A) of the article A while conveying the article A, and inspects the article A based on the transmission image G1 (for example, inspection of the number of items stored, foreign matter contamination inspection, etc.). inspection, missing item inspection, crack inspection, etc.). In this embodiment, the X-ray inspection apparatus 10 inspects the article A for inclusion of foreign matter. The article A before inspection is carried into the X-ray inspection apparatus 10 by the carry-in conveyor 19 . After inspection, the article A is carried out to the conveyor 31 . Articles A determined to be defective (abnormal) by the X-ray inspection device 10 are sorted out of the production line by the sorting device 30 arranged on the conveyor 31 . Articles A determined to be non-defective by the X-ray inspection device 10 pass through the sorting device 30 as they are.

The device main body 11 accommodates the control unit 18 and the like. The support leg 12 supports the device main body 11 . The shield box 13 is provided in the device main body 11 . The shield box 13 prevents leakage of X-rays to the outside. Inside the shield box 13, there is provided an inspection area R in which the article A is inspected by X-rays. The shield box 13 is formed with a carry-in port 13a and a carry-out port 13b. The article A before inspection is carried into the inspection area R from the carry-in conveyor 19 through the carry-in entrance 13a. After the inspection, the articles A are carried out from the inspection area R to the conveyor 31 of the sorting device 30 through the outlet 13b. An X-ray shielding curtain (not shown) for preventing leakage of X-rays is provided at each of the carry-in port 13a and the carry-out port 13b.

The transport section 14 is arranged inside the shield box 13 . The transport unit 14 transports the article A along the transport direction D from the carry-in port 13a through the inspection area R to the carry-out port 13b. The conveying unit 14 is, for example, a belt conveyor stretched between the carry-in port 13a and the carry-out port 13b.

As shown in FIGS. 1 and 2, the X-ray irradiation unit 15 is arranged inside the shield box 13 . The X-ray irradiation unit 15 irradiates the article A transported by the transport unit 14 with X-rays. The X-ray irradiation unit 15 has, for example, an X-ray tube that emits X-rays, and a collimator that spreads the X-rays emitted from the X-ray tube in a fan shape within a plane perpendicular to the transport direction D.

The X-ray detector 16 is arranged inside the shield box 13 . The X-ray detection unit 16 is a line sensor composed of X-ray detection elements arranged one-dimensionally along the horizontal direction perpendicular to the transport direction D. As shown in FIG. The X-ray detector 16 detects X-rays that have passed through the article A and the conveyor belt of the conveyor 14 .

As shown in FIG. 1, the display/operation unit 17 is provided in the device main body 11 . The display operation unit 17 displays various information and accepts input of various conditions. The display operation unit 17 is, for example, a liquid crystal display, and displays an operation screen as a touch panel. In this case, the operator can input various conditions via the display operation section 17 .

The control unit 18 is arranged inside the device main body 11 . The control unit 18 controls operations of each unit of the X-ray inspection apparatus 10 . The control unit 18 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like. The X-ray detection result of the X-ray detection unit 16 is input to the control unit 18 . The control unit 18 creates a transmission image G1 based on the X-ray detection result.

As shown in FIG. 3 , the control unit 18 has a setting unit 20 , a processing unit 21 , an acquisition unit 22 , an inspection unit 23 and an output unit 24 .

The setting unit 20 sets the reduction ratio of the reduced image G2 (see FIG. 4B) created by the processing unit 21 . The setting unit 20 reduces the reduced image G2 based on the first time from when the transparent image G1 is created until the article A reaches the sorting device 30 and the second time required for processing by the learned model. set the rate. The first time is set according to the configurations of the X-ray inspection device 10 and the sorting device 30 . Specifically, the first time includes the dimension (length) of the article A, the distance between the X-ray detection unit 16 and the sorting device 30, the transport speed of the transport unit 14, the transport speed of the conveyor 31, and the like. is set as appropriate.

The second time is set according to the processing capability of the server 50 and the like. The second time is longer as the number of pixels of the image is larger. The setting unit 20 sets the reduction ratio so that the second time is shorter than the first time (first time>second time). The second time is, for example, 200 msec. From the viewpoint of processing accuracy by the learned model, it is preferable that the second time is not significantly different from the first time. As a result, it is possible to set a reduction ratio that achieves a predetermined processing accuracy while ensuring the sorting by the sorting device 30 . The setting unit 20 outputs reduction information indicating the set reduction ratio to the processing unit 21 .

The processing unit 21 performs image processing on the transparent image G1. The processing unit 21 reduces the number of pixels of the transmissive image G1 to create a reduced image G2. The reduction process is a process for reducing the number of pixels of the transparent image G1, and a known method can be used. The processing unit 21 creates the reduced image G2 using, for example, the nearest neighbor method, the bilinear method, or the bicubic method. The processing unit 21 creates a reduced image G2 based on the reduction ratio set by the setting unit 20. FIG. The processing unit 21 may perform various types of processing, such as contrast adjustment, color change, and format change, on the transparent image G1. The processing unit 21 transmits image information related to the reduced image G2 to the server 50 via a communication unit (not shown).

The acquisition unit 22 acquires a processing result of processing by a learned model generated by machine learning. The acquisition unit 22 receives the processing information transmitted from the server 50 according to the image information transmitted from the processing unit 21 and acquires the processing result. The acquisition unit 22 outputs the acquired processing result to the inspection unit 23 .

The inspection unit 23 inspects the article A. The inspection unit 23 determines whether or not the article A contains a foreign substance F based on the processing result acquired by the acquisition unit 22 and the determination result based on the transparent image G1.

The inspection unit 23 uses an image processing algorithm set for each of a plurality of sensitivity levels (eg, 1 to 7) to process (image process) the transmission image G1 to generate a processed image. An image processing algorithm is a type indicating a processing procedure of image processing to be applied to the transparent image G1. An image processing algorithm is composed of one image processing filter or a combination of a plurality of image processing filters. A plurality of image processing algorithms can be obtained externally via a network such as the Internet. A plurality of image processing algorithms can also be acquired from an external storage medium such as a USB memory or removable hard disk. At least one of the plurality of image processing algorithms employs a genetic algorithm (GA), which is a technique that applies the mechanism of heredity and evolution in the living world, and the specifications of the X-ray inspection apparatus 10 It can be automatically generated from a plurality of image processing filters based on inspection conditions and the like. At least part of the plurality of image processing algorithms can also be appropriately set by the operator via the display operation unit 17 .

The inspection unit 23 performs predetermined image processing on the transparent image G1 based on an image processing algorithm, and determines whether or not the article A contains foreign matter F based on the determination image obtained by the image processing. . The predetermined image processing is, for example, binarization processing of the transparent image G1. The inspection unit 23 determines that the foreign matter F is included in the article A when the luminance value exceeds the threshold value in the determination image. The threshold value is appropriately set by testing or the like according to the properties of the article A. The inspection unit 23 stores the determination result in a storage unit (not shown).

The inspection unit 23 inspects whether or not the foreign matter F is included in the article A based on the processing result acquired by the acquisition unit 22 and the determination result. If the processing result indicates that the article A does not contain the foreign matter F and the determination result indicates that the article A does not include the foreign matter F, the inspection unit 23 determines that the article A does not include the foreign matter F. judge. The inspection unit 23 determines that the article A contains the foreign matter F when the processing result indicates that the article A includes the foreign matter F and the determination result indicates that the article A includes the foreign matter F. do. If one of the processing result and the judgment result indicates that the article A contains the foreign matter F, and the other of the processing result and the judgment result indicates that the article A does not contain the foreign matter F, the inspection unit 23 determines whether the processing result to adopt. Specifically, for example, if the processing result indicates that the article A contains the foreign matter F, and the determination result indicates that the article A does not include the foreign matter F, the inspection unit 23 determines that the article A does not include the foreign matter F. It is determined that The inspection unit 23 outputs inspection information indicating inspection results to the output unit 24 .

The output unit 24 outputs various signals based on the inspection information output from the inspection unit 23 . The output unit 24 outputs a display signal to the display operation unit 17 for displaying the inspection result on the display operation unit 17 based on the inspection information. The output unit 24 outputs a sorting signal instructing sorting of the article A to the sorting device 30 when the article A contains the foreign matter F in the inspection information.

As shown in FIG. 5 , the sorting device 30 is provided downstream of the X-ray inspection device 10 . The sorting device 30 is provided on the conveyor 31 . The sorting device 30 sorts the articles A based on the sorting signal output from the X-ray inspection device 10 . The sorting device 30 has a photoelectric sensor 32 and an arm 33 .

The photoelectric sensor 32 is a sensor that detects passage of the article A. FIG. The photoelectric sensor 32 is installed on the upstream side of the arm 33 and detects the carry-in of the article A to the sorting device 30 . The photoelectric sensor 32 has a light projector 32a and a light receiver 32b. The photoelectric sensor 32 transmits a signal to the X-ray inspection apparatus 10 assuming that the article A has reached the sorting device 30 when the light emitted from the light projector 32a to the light receiver 32b is blocked by the article A.

The tip of the arm 33 swings about the proximal side as a base axis by driving force of a motor or the like. The arm 33 pushes the article A to one side in the width direction of the conveyor 31 and sorts the article A out of the production line.

As shown in FIG. 1, the server 50 is a device that generates a learned model through machine learning and performs processing using the learned model. The server 50 includes a CPU (Central Processing Unit), a ROM (Read Only Memory), a RAM (Random Access Memory), and the like.

The server 50 is managed by the user of the X-ray inspection apparatus 10, for example. As shown in FIG. 1, the X-ray inspection apparatus 10 and the server 50 are communicably connected by a wired or wireless network N such as the Internet or a telephone network, and can exchange information with each other. .

The server 50 includes a communication unit 51 , a trained model generation unit 52 and a learning unit 53 .

The communication unit 51 communicates with the X-ray inspection apparatus 10 . The communication unit 51 receives image information transmitted from the X-ray inspection apparatus 10 and outputs the image information to the learning unit 53 . The communication unit 51 transmits the processing information output from the learning unit 53 to the X-ray inspection apparatus 10 .

The trained model generating unit 52 acquires learning data used for machine learning, performs machine learning using the acquired learning data, and generates a trained model. Learning data includes teacher images and other data. The teacher image is, for example, a transmitted image acquired by the X-ray inspection apparatus 10, and includes a transmitted image including the foreign matter F, a transmitted image not including the foreign matter F, and the like. Other data includes, for example, data related to areas where products overlap each other when product A includes a plurality of products in a transmission image acquired by the X-ray inspection apparatus 10, and data included in product A. data and the like related to the area where the foreign matter F that is trapped exists.

The learned model generation unit 52 uses each pixel value of the teacher image as an input value to the neural network, and performs machine learning using the processing information corresponding to the teacher image as the output value of the neural network to create a neural network NW (Fig. 6 ). When a pixel value is used as an input value, it is used as an input value of a neuron associated with each pixel (pixel position on the image). The above machine learning itself can be performed similarly to conventional machine learning algorithms.

The learned model outputs information indicating the presence/absence of the foreign matter F in the reduced image G2. A trained model includes a neural network NW. A trained model may include a convolutional neural network. Furthermore, the trained model may include neural networks of multiple layers (e.g., 8 layers or more). That is, a trained model may be generated by deep learning.

As shown in FIG. 6, the neural network NW includes, for example, a first layer that is an input layer, second, third, and fourth layers that are intermediate layers (hidden layers), and an output layer. and the fifth layer. The first layer outputs an input value x=(x ₀ , x ₁ , x ₂ , . . . , x _p ) having p parameters as elements to the second layer. Each of the second, third, and fourth layers converts total input to output by an activation function and passes the output to the next layer. The fifth layer also transforms the total input into an output by an activation function, which is the output value y=(y ₀ , y ₁ ) of the neural network whose elements are two parameters.

In this embodiment, the neural network NW inputs the pixel value of each pixel of the reduced image G2 and outputs information indicating the presence or absence of the foreign matter F. FIG. The input layer of the neural network NW is provided with as many neurons as the number of pixels of the reduced image G2. The output layer of the neural network NW is provided with neurons for outputting information indicating the presence or absence of foreign matter F. FIG. Information indicating the presence or absence of foreign matter F can be determined based on the output values of neurons in the output layer. The output value of a neuron is a value between 0 and 1, for example. In this case, the larger the neuron value (the closer the value is to 1), the more foreign matter F is included in the article A (abnormal). This indicates that A does not contain foreign matter F (is normal).

The learning unit 53 inputs the reduced image G2 based on the image information output from the communication unit 51 to the trained model. The learning unit 53 outputs to the communication unit 51 processing information indicating a processing result including the output value output from the neural network NW of the trained model. The processing result includes information indicating the presence/absence of the foreign matter F in the article A and, if the article A includes the foreign matter F, information indicating the area (position) of the article A in which the foreign matter F is included. is

Next, the operation (inspection method) of the X-ray inspection system 1 will be described with reference to FIG. As shown in FIG. 7, the X-ray inspection apparatus 10 creates a transmission image G1 from the detection result of the X-ray detection unit 16 (step S01). Subsequently, the X-ray inspection apparatus 10 performs reduction processing on the transmitted image G1 to create a reduced image G2 (step S02, processing step). X-ray inspection apparatus 10 transmits reduced image G2 to server 50 (step S03). Also, the X-ray inspection apparatus 10 determines whether or not the article A contains foreign matter F based on the transmission image G1 (step S04).

The server 50 processes the reduced image G2 transmitted from the X-ray inspection apparatus 10 using the learned model (step S05). Server 50 transmits the result of processing by the learned model to X-ray inspection apparatus 10 (step S06).

Subsequently, the X-ray inspection apparatus 10 acquires the processing result transmitted from the server 50 (acquisition step), and based on the processing result, inspects whether or not the article A contains foreign matter F (step S07, inspection step). When the X-ray inspection device 10 determines that the foreign matter F is included in the article A, it transmits a sorting signal to the sorting device 30 (step S08). If the X-ray inspection apparatus 10 does not determine that the foreign matter F is included in the article A, the process ends.

Next, an inspection program P for realizing the X-ray inspection system 1 will be described. As shown in FIG. 8, the inspection program P can be recorded on a computer-readable recording medium 100. FIG. The inspection program P stored in the recording medium 100 includes a setting module P1, a processing module P2, an acquisition module P3, an inspection module P4 and an output module P5. The inspection program P functions by causing the computer to execute the setting module P1, the processing module P2, the acquisition module P3, the inspection module P4, and the output module P5. Functions realized by executing the setting module P1, the processing module P2, the acquisition module P3, the inspection module P4, and the output module P5 are a setting unit 20, a processing unit 21, an acquisition unit 22, an inspection unit 23, and an output unit 24, respectively. is similar to the function of

The inspection program P is recorded in a program recording area of the recording medium 100 . The recording medium 100 is, for example, a recording medium such as a CD-ROM, DVD, ROM, semiconductor memory, or the like. The test program P may be provided over a communication network as a computer data signal superimposed on a carrier wave.

As described above, in the X-ray inspection system 1 according to this embodiment, the reduced image G2 is used for processing by the learned model. As a result, in the X-ray inspection system 1, the processing load due to the learned model is reduced, so that processing results can be obtained quickly. Therefore, in the X-ray inspection system 1, it is possible to improve the throughput.

Further, in the X-ray inspection system 1 according to the present embodiment, the processing capability can be improved by using the reduced image G2 for processing by the learned model. Therefore, in the X-ray inspection system 1, the processing capacity can be improved without increasing the performance of the server 50 (such as CPU specifications). Therefore, since the X-ray inspection system 1 does not require expensive equipment, cost reduction can be achieved.

In the X-ray inspection system 1 according to this embodiment, the inspection unit 23 of the X-ray inspection apparatus 10 inspects the article A based on the transmission image G1, and based on the determination result and the processing result of the inspection, Inspect item A. In this configuration, the article A is inspected based on the determination result of the luminance value threshold determination based on the transmission image G1 and the processing result. As described above, the X-ray inspection apparatus 10 inspects the article A based on the two results, so that inspection accuracy can be improved.

In the X-ray inspection system 1 according to the present embodiment, the X-ray inspection apparatus 10 sends a sorting signal to the sorting device 30 that sorts the articles A when the inspection unit 23 detects an abnormality in the article. Based on the output unit 24 to output, the first time from the creation of the transparent image G1 until the article reaches the sorting device 30, and the second time required for processing by the learned model, the reduced image G2 and a setting unit 20 for setting a reduction ratio. The time required for processing by the trained model changes according to the number of pixels of the reduced image G2, longer when the number of pixels is large, and shorter when the number of pixels is small. If the number of pixels is large, the processing may not be completed until the article A reaches the sorting device 30 because the processing takes time. On the other hand, if the number of pixels is too small, the precision of processing by the trained model may decrease. The X-ray inspection apparatus 10 sets the reduction ratio of the reduced image G2 based on the first time and the second time. Thereby, in the X-ray inspection apparatus 10, the sorting by the sorting device 30 can be reliably performed, and the deterioration of the processing accuracy by the learned model can be suppressed.

In the X-ray inspection system 1 according to this embodiment, the server 50 includes a learned model generation unit 52 that generates a learned model by machine learning using teacher images. With this configuration, a trained model suitable for inspecting articles can be generated. The trained model generation unit 52 generates a trained model including a neural network. With this configuration, the learned model can be made appropriate, and the inspection accuracy of the article can be improved.

In the X-ray inspection system 1 according to this embodiment, the inspection unit 23 of the X-ray inspection apparatus 10 inspects whether or not the article A contains foreign matter F. When the article A includes a plurality of products, the products may overlap each other. In this case, it becomes difficult to distinguish between the portion where the products are overlapped and the foreign object based on the luminance value of the transparent image G1. Since the X-ray inspection apparatus 10 performs inspection based on the processing result of the processing using the learned model, it is possible to distinguish between the overlapping portions of the products and the foreign matter. Therefore, the X-ray inspection apparatus 10 is particularly effective in inspecting whether or not the article A contains foreign matter F.

Although the embodiments of the present invention have been described above, the present invention is not necessarily limited to the above-described embodiments, and various modifications can be made without departing from the scope of the invention.

In the above-described embodiment, the X-ray detection unit 16 of the X-ray inspection apparatus 10 has been described as an example having one line sensor. However, the X-ray detector 16 may have two line sensors. In this configuration, the two line sensors are arranged facing each other in the vertical direction. One (upper) line sensor detects low-energy band X-rays that have passed through the article A and the conveyor belt of the conveyor unit 14 . The other (lower) line sensor detects the high-energy band X-rays that have passed through the article A, the conveyor belt of the conveyor unit 14, and one of the line sensors.

In the above-described embodiment, the inspection unit 23 of the X-ray inspection apparatus 10 inspects the article A based on the determination result based on the transmission image G1 and the processing result of the process performed by the learned model generated by machine learning. has been described as an example. However, the inspection unit 23 may inspect the article A based only on the processing result of the processing by the learned model.

In the above-described embodiment, an example in which the X-ray inspection system 1 includes the X-ray inspection apparatus 10 and the server 50 has been described. However, the server 50 may not be provided. In this case, the X-ray inspection apparatus 10 only needs to include a learned model generation section and a learning section. Alternatively, the X-ray inspection apparatus 10 may acquire a trained model generated by another device (computer) and store it in the storage unit.

In the above-described embodiment, an example in which the X-ray inspection system 1 includes the X-ray inspection device 10 and the sorting device 30 has been described. However, the sorting device 30 may be included as part of the configuration of the X-ray inspection device 10 .

In the above-described embodiment, an example has been described in which the server 50 includes the trained model generation unit 52 . However, the server 50 does not have to include the trained model generator 52 . In this case, the server 50 acquires a trained model generated by another device, stores it in the storage unit, and processes the reduced image G2 using the trained model stored in the storage unit. .

In the above embodiment, the trained model generation unit 52 obtains learning data used for machine learning, performs machine learning using the obtained learning data, and generates a trained model. However, the method of generating trained models is not limited to this. In addition to the teacher image and other data, other data may be used as the learning data.

In the above embodiment, the neural network NW has five layers (four layers when the input layer is excluded) as an example. However, the number of neural network layers forming the trained model generation unit 52 is not limited at all. For example, the trained model generation unit 52 may use a neural network having an arbitrary number of layers of 3 or more, and may use a neural network having an arbitrary number of intermediate layers of 1 or more. means. Also, the configuration of each layer of the neural network (for example, the number of neurons) and the connections between neurons are not limited to the configurations shown in the above embodiments.

In the above-described embodiment, an example in which the inspection apparatus is the X-ray inspection apparatus 10 has been described. However, it is not limited to the X-ray inspection apparatus, and any inspection apparatus that inspects articles using electromagnetic waves may be used. That is, in the present invention, electromagnetic waves are X-rays, near-infrared rays, light, and other electromagnetic waves. In addition, the present invention is not limited to inspecting the presence or absence of foreign matter contained in an article, and is not limited to an article that contains contents such as food in a package such as a film packaging material and is shipped. It may also be an inspection for contents biting into parts, breakage of contents in the package, contamination of foreign matter in the package, and the like. Also, the type of article is not particularly limited, and various articles can be inspected. Similarly, the type of foreign matter is not particularly limited, and various foreign matter can be inspected.

DESCRIPTION OF SYMBOLS 1... X-ray inspection system, 10... X-ray inspection apparatus, 15... X-ray irradiation part (irradiation part), 16... X-ray detection part (detection part), 20... Setting part, 21... Processing part, 22... Acquisition part , 23... inspection unit, 24... output unit, 50... server, 51... communication unit, 52... learned model generation unit, 100... recording medium, A... article, F... foreign matter, P... inspection program, P2... processing module (Processing unit), P3... Acquisition module (acquisition unit), P4... Inspection module (inspection unit).

Claims (10)

an irradiation unit that irradiates an article with electromagnetic waves;
a detection unit that detects an electromagnetic wave that has passed through the article;
a processing unit that creates a reduced image by performing pixel number reduction processing on the transparent image created from the detection result of the detection unit;
an acquisition unit that acquires a processing result of performing processing by a trained model generated by machine learning on the reduced image created by the processing unit;
an inspection unit that inspects the article based on the processing result acquired by the acquisition unit;
an output unit that outputs a sorting signal to a sorting device that sorts the articles when the inspecting unit detects an abnormality in the article;
A reduction ratio of the reduced image is set based on a first time from when the transparent image is created until the article reaches the sorting device and a second time required for processing by the learned model. a setting unit;
The inspection device, wherein the processing unit generates the reduced image using the reduction rate at which processing by the learned model is completed before the article reaches the sorting device. 2. The inspection apparatus according to claim 1, wherein said inspection unit inspects said article based on said transmission image, and inspects said article based on a determination result of said inspection and said processing result. 3. The inspection apparatus according to claim 1 , further comprising a generator that generates the learned model by machine learning using a teacher image. a learning unit that processes the reduced image using the trained model generated by the generating unit;
The inspection device according to claim 3 , wherein the acquisition unit acquires the processing result executed by the learning unit. The inspection device according to claim 3 or 4 , wherein said generation unit generates said trained model including a neural network. The inspection device according to any one of claims 1 to 5 , wherein the inspection unit inspects whether or not the article contains foreign matter. The irradiation unit irradiates the article with X-rays,
The inspection apparatus according to any one of claims 1 to 6 , wherein the detection unit detects the X-rays that have passed through the article. An inspection method performed by an inspection apparatus comprising an irradiation unit that irradiates an article with electromagnetic waves and a detection unit that detects the electromagnetic waves that have passed through the article,
a processing step of reducing the number of pixels of a transmission image created from the detection result of the detection unit to create a reduced image;
an acquisition step of acquiring a processing result of processing the reduced image created in the processing step by a trained model generated by machine learning;
an inspection step of inspecting the article based on the processing result acquired in the acquisition step;
an output step of outputting a sorting signal to a sorting device that sorts the articles when an abnormality in the article is detected in the inspection step;
A reduction ratio of the reduced image is set based on a first time from when the transparent image is created until the article reaches the sorting device and a second time required for processing by the learned model. a configuration step;
In the processing step, the inspection method includes generating the reduced image using the reduction ratio at which processing by the learned model is completed before the article reaches the sorting device. A computer of an inspection device comprising an irradiation unit for irradiating an article with electromagnetic waves and a detection unit for detecting the electromagnetic waves transmitted through the article,
a processing unit that creates a reduced image by performing pixel number reduction processing on the transparent image created from the detection result of the detection unit;
an acquisition unit that acquires a processing result of performing processing by a trained model generated by machine learning on the reduced image created by the processing unit;
an inspection unit that inspects the article based on the processing result acquired by the acquisition unit;
an output unit that outputs a sorting signal to a sorting device that sorts the articles when the inspecting unit detects an abnormality in the article;
A reduction ratio of the reduced image is set based on a first time from when the transparent image is created until the article reaches the sorting device and a second time required for processing by the learned model. function as a setting part ,
The inspection program, wherein the processing unit generates the reduced image using the reduction rate at which processing by the learned model is completed before the article reaches the sorting device. A computer-readable recording medium recording an inspection program to be executed by a computer of an inspection apparatus comprising an irradiation unit for irradiating an article with electromagnetic waves and a detection unit for detecting the electromagnetic waves transmitted through the article,
said computer,
a processing unit that creates a reduced image by performing pixel number reduction processing on the transparent image created from the detection result of the detection unit;
an acquisition unit that acquires a processing result of performing processing by a trained model generated by machine learning on the reduced image created by the processing unit;
an inspection unit that inspects the article based on the processing result acquired by the acquisition unit;
an output unit that outputs a sorting signal to a sorting device that sorts the articles when the inspecting unit detects an abnormality in the article;
A reduction ratio of the reduced image is set based on a first time from when the transparent image is created until the article reaches the sorting device and a second time required for processing by the learned model. function as a setting part ,
The processing unit generates the reduced image using the reduction ratio at which processing by the learned model is completed before the article reaches the sorting device, and reads the inspection program. Possible recording media.

Images