Patent Application
20230252621
The present invention relates to an inspection device that inspects a quality state required for an inspection object, a learned model generation method, and an inspection method.
For example, an X-ray inspection device disclosed in the following Patent Document 1 is known as a device for inspecting a quality state required for an inspection object using X-rays. The X-ray inspection device disclosed in Patent Document 1 stores X-ray image data from an X-ray detector in an X-ray image storage unit, generates, using a pseudo image generation model, pseudo X-ray image data in another energy band including a predetermined energy band for various learning target article type, based on a learned result of the X-ray image data in a plurality of different energy bands, creates, by an image creation unit, a pseudo transmission image in another energy band using the pseudo image generation model based on the X-ray image data in a predetermined energy band of the inspection object, and determines, by a determination unit, a quality state of the inspection object based on the X-ray image data in the predetermined energy band of the inspection object and the pseudo transmission image in another energy band created an image creation unit.
However, in the conventional inspection device disclosed in Patent Document 1, the transmission image used for determination is one grayscale image, and an amount of information for determination is small, resulting in limitation to improvement in accuracy of inspecting a quality state of the inspection object.
Accordingly, the present invention has been made in view of the above problems, and an object of the present invention is to provide an inspection device capable of improving accuracy of inspecting a quality state of an inspection object, a learned model generation method, and an inspection method.
In order to achieve the object, an inspection device according to a first aspect of the present invention includes: an image storage unit 21 that stores, as pseudo RGB images, three inspection images having different transmission characteristics and obtained by capturing an inspection object W; and
a determination unit 24 that obtains a defective quality degree for the pseudo RGB image stored in the image storage unit based on a learned model 22 created in advance by learning using an image having a same format as the pseudo RGB image, and determines a quality state of the inspection object by comparison between the defective quality degree and a preset threshold.
An inspection device according to a second aspect of the present invention is the inspection device according to the first aspect
in which the learned model is learned for each type of the inspection object with respect to the image having the same format as the pseudo RGB image including at least images with a defective quality.
An inspection device according to a third aspect of the present invention is the inspection device according to the first aspect
in which the pseudo RGB images are three inspection images obtained by spectroscopy of light transmitting through the inspection object.
In addition, an inspection device according to a fourth aspect of the present invention is the inspection device according to the second aspect in which
the pseudo RGB images are three inspection images obtained by spectroscopy of light transmitting through the inspection object.
A learned model generation method according to a fifth aspect of the present invention includes: a learning image acquisition step of acquiring a non-defective image of an inspection object W and an image with only defective quality of the inspection object as learning images;
a step of creating a learning defective quality synthesis image in which the image with only defective quality is synthesized with the non-defective image of the inspection object using the learning image and a learning defective quality label showing a defective quality position in the learning defective quality synthesis image; and
a step of creating a learned model 22 by performing machine learning of the learning defective quality synthesis image,
in which the learning image acquired in the learning image acquisition step is a pseudo RGB image including three inspection images having different transmission characteristics and obtained by capturing the inspection object.
An inspection method according to a sixth aspect of the present invention includes: a step of obtaining a defective quality degree for pseudo RGB images of an inspection object W including three inspection images having different transmission characteristics and obtained by capturing the inspection object based on the learned model 22 created by the learned model creation method of the fifth aspect, and determining a quality state of the inspection object by comparison between the defective quality degree and a preset threshold.
According to the present invention, it is possible to improve the accuracy of inspecting the quality state of the inspection object by learning determination using the three channels (hereinafter, “3ch”) of inspection image based on transmission corresponding to the RGB color image.
Hereinafter, a mode for carrying out the present invention will be described in detail referring to the accompanying drawings.
As shown in
The three inspection images having different characteristics are having different transmission characteristics to the inspection object W and capable of distinguishing the foreign matter from other matters, that is, the grayscale images having different shades (changes), and processed as pseudo RGB images having the same position information with respect to the inspection object W like an RGB image, which is captured by a color camera and divided into three primary colors for each pixel.
The inspection device 1 performs processing the pseudo RGB images for each pixel based on a learned model (calculation formula) corresponding to the type of the inspection object W to be inspected, obtaining a defective quality degree showing a probability of a defective quality (for example, foreign matter degree, shape defect degree, or the like), and determining a quality state of the inspection object W by comparison between the obtained defective quality degree and a preset threshold.
In addition, the “quality state” in this example means suitability of quality, physical quantity, and the like required for the inspection object W as a product. Specifically, examples of the “quality state” include presence/absence of foreign matter (bone, metal, or the like) contained in the inspection object W, excess or deficiency of contents, different types of contents, and shape defect of contents.
The transport unit 2 sequentially transports the inspection object W at a predetermined interval in a transport direction A, and has a configuration in which, for example, a loop-shaped transport belt 11 is wound around a plurality of transport rollers 12, and a conveyor, which can sequentially transport the inspection object W to the right in
The image acquisition unit 3 acquires a learning image used in a learning phase to be described later or an inference image of the inspection object W used in an inference phase to be described later, and serves as an inspection unit that inspects a quality state of the inspection object W transported by the transport unit 2.
The image acquisition unit 3 includes an X-ray generator 14 that generates, for example, X-rays in a predetermined energy band that transmits through the inspection object W transported by the transport unit 2, when the learning image and the inference image are acquired as an X-ray image, and an X-ray detector 15 that is disposed immediately below the upper running section 13 of the transport belt 11 of the transport unit 2 to face the X-ray generator 14.
The X-ray generator 14 generates X-rays having a wavelength and intensity corresponding to a tube current and a tube voltage of a known X-ray tube 16, and allows the X-rays to pass through an X-ray window 17a of an enclosure 17, so as to irradiate the inspection object W or a sample (non-defective work of the inspection object W, or foreign matter) on the transport belt 11 with fan-beam X-rays orthogonal to the transport direction of the transport unit 2.
The tube current and tube voltage of the X-ray tube 16 has a set value that may be adjusted according to a material or size (in particular, dimension in a direction in which the X-rays transmit through) of the inspection object W to be inspected or the sample, and has a set value that is determined and selected so that an appropriate contrast can be obtained by test imaging using the inspection object W for a new type.
The X-ray detector 15 has, for example, detection elements, which includes a scintillator, which is a phosphor, and a photodiode or a charge-coupled element, arranged in an array at a predetermined pitch in a width direction of a transport path of the transport unit 2, includes an X-ray line sensor camera so as to detect X-rays at a predetermined resolution, and is arranged at a predetermined position in a transport direction that corresponds to a position where X-rays are irradiated from the X-ray generator 14.
The X-ray detector 15 detects the X-rays, which is irradiated from the X-ray generator 14 and transmits through the inspection object W or the sample, for each predetermined transmission region corresponding to the detection element, and converts the X-rays into an electric signal according to a transmission amount of the X-rays to output an X-ray detection signal for each transmission region.
Here, a method for acquiring a learning image from an X-ray image when presence/absence of a foreign matter on the inspection object W is inspected will be described. First, a tube current and tube voltage of the X-ray tube 16 for a certain type of the inspection object W are set by test imaging, and the transport unit 2 is driven and controlled. That is, acquisition conditions for acquiring an X-ray image for a certain type of the inspection object W are set, and when 1,000 non-defective images and 50 images of only foreign matter (bone) are acquired as learning images, first, a low-energy image (non-defective product)_(1ch) and a high-energy image (non-defective product)_(1ch) are acquired at the same timing by irradiating non-defective work of the inspection object W with the X-rays from the X-ray generator 14, for example, from a position at a predetermined height. This operation is executed 1,000 times to acquire 1,000 low-energy images (non-defective products) and 1,000 high-energy images (non-defective products). As a result, one low-energy image (non-defective product) and one high-energy image (non-defective product) can be acquired in one execution.
Similarly, a low-energy image (only one foreign matter)_(1ch) and a high-energy image (only one foreign matter)_(1ch) are acquired at the same timing by irradiating work of only foreign matter (for example, only one piece of bone) with the X-rays from the X-ray generator 14, for example, from a position at a predetermined height (the same position as the non-defective image). This operation is executed 50 times to acquire 50 low-energy images (only one foreign matter) and 50 high-energy images (only one foreign matter). As a result, one low-energy image (only one foreign matter) and one high-energy image (only one foreign matter) can be acquired in one execution.
In this way, learning images are acquired under predetermined acquisition conditions for a certain type of the inspection object W.
Next, a 3ch pseudo RGB image is created from the learning image described above. In this case, the low-energy and high-energy images of a non-defective work, and one database image of a foreign matter are prepared.
Here, the low-energy and high-energy images of a non-defective work are the pair of images obtained at the same timing by irradiating the non-defective work of the inspection object W with the X-rays from the X-ray generator 14, and the database image is the image obtained by cutting from a plurality of low-energy and high-energy images obtained by irradiating only foreign matter at the same timing.
The 3ch pseudo RGB images are created based on the low-energy image, the high energy image and the difference image of the two images. The low-energy image and high-energy image are created by synthesizing the database images of a foreign matter while being shifted to random positions on one non-defective image. The creation of the plurality of pseudo RGB images from the one non-defective image is performed on the prepared plurality of non-defective images, for example, several tens of/non-defective images. Furthermore, a plurality of other low-energy and high-energy images, both of which are captured at the same timing, of only a foreign matter is prepared, and 3ch pseudo RGB images having the same position information and different transmission characteristics are created using the same method.
Further, a method for creating a 3ch pseudo RGB image will be described with specific numerical examples. First, as learning images, 1,000 low-energy images (non-defective product)_(1ch), 1,000 high-energy images (non-defective product)_(1ch), 50 low-energy images (only one foreign matter)_(1ch), and 50 high-energy images (only one foreign matter)_(1ch) are obtained respectively.
Then, 50 low-energy images (only one foreign matter)_(1ch) and 50 high-energy images (only one foreign matter)_(1ch) are used to create a database image of a foreign matter, which is obtained by cutting only a region where the foreign matter is shown, with the low energy and the high energy, respectively. That is, one low-energy image (50 foreign matters are shown)_(1ch) and one high-energy image (50 foreign matters are shown)_(1ch) are created.
Next, a foreign matter synthesis image is created by synthesizing one low-energy image (which is obtained by synthesizing a foreign matter with a non-defective product)_(1ch) and one high-energy image (which is obtained by synthesizing a foreign matter with a non-defective product)_(1ch) from one low-energy image (non-defective product)_(1ch), one high-energy image (non-defective product)_(1ch), one low-energy image (50 foreign matters are shown)_(1ch), and one high-energy image (50 foreign matters are shown)_(1ch).
When the foreign matter is synthesized with the non-defective image, coordinate data (labels) of the foreign matter are simultaneously output under conditions, such as foreign matter position: random position on non-defective work and the same position as the (low- and high-) energy images, foreign matter rotation angle: random angle from 0 to 360 degrees, and the number of foreign matters: randomly selected foreign matter (one or a plurality of foreign matters randomly selected from 50 foreign matters).
Here,
Then, a difference between the two is made using one low-energy image (synthesis of a foreign matter with a non-defective product)_(1ch) and one high-energy image (synthesis of a foreign matter with a non-defective product)_(1ch). As a result, one difference image (difference between the low-energy image and the high-energy image obtained by synthesizing a foreign matter with a non-defective product)_(1ch) is created.
Further, for example, one pseudo RGB image (R component: low energy, G component: high energy, and B component: difference) (3ch) including a foreign matter (a portion indicated by diagonal lines in the drawing) as shown in
The above operation is executed 1,000 times to acquire 1,000 pseudo RGB images (R component: low energy, G component: high energy, and B component: difference) (3ch) and 1,000 foreign matter coordinate data (positions of the low energy and high energy are common) as the learning foreign matter synthesis image i4 and the learning foreign matter label r.
Next, a method for acquiring an inference image from an X-ray image will be described. When an inference image is acquired from an X-ray image, it is required that the case of acquiring a learning image described above and the acquisition conditions (configurations and positions of the X-ray generator 14 and the X-ray detector 15, X-ray irradiation conditions, X-ray detection conditions, transport speed, and the like) of an image by the image acquisition unit 3 are the same. To specifically describe acquisition of the low-energy image and the high-energy image, for example, the inspection object W is irradiated with X-rays from one X-ray source of the X-ray generator 14, for example, from directly above at a predetermined height, the X-rays transmitted through the inspection object W is detected using two X-ray line sensor camera with X-ray filters having different ray qualities (which refers to a type or energy of radiation during irradiation), and images detected by the two X-ray line sensor camera are treated as a low-energy image and a high-energy image, respectively, thereby creating a difference image from the two images.
Alternatively, after using X-ray filters, the inspection object W is irradiated with the X-rays having different ray qualities from the X-ray generator 14, for example, from directly above at a predetermined height, the X-rays transmitted through the inspection object W are detected using two X-ray line sensor cameras for different energies, and images detected by the two X-ray line sensor cameras are treated as a low-energy image and a high-energy image, respectively, thereby creating a difference image from the two images (see JP-A-2002-168803 and Japanese Patent No. 5297087).
Alternatively, the inspection object W is irradiated with low-energy X-rays and high-energy X-rays from the two X-ray generators 14, respectively, for example, from directly above at a predetermined height, and the X-rays transmitted through the inspection object W are detected using the corresponding two X-ray line sensor cameras, respectively, thereby creating a difference image from the low-energy image and the high-energy image detected by the two X-ray line sensor cameras (see Japanese Patent No. 5706724 and Japanese Patent No. 5775406).
Alternatively, a single photon counting-type X-ray detector capable of simultaneously acquiring a low-energy image and a high-energy image may be used.
When using two X-ray line sensor cameras, the acquisition conditions may include at least an arrangement interval of the two X-ray line sensor cameras and a transport speed of the transport unit 2 (transport belt 11), or a value determined based on the arrangement interval and the transport speed. Accordingly, a difference in an imaging position of the inspection object W in the image acquired by each X-ray line sensor camera can be grasped. If the difference in the imaging position can be quantitatively grasped, various calculation processing can be performed to reduce an influence of the difference, such that it is technically synonymous with “having the same position information”.
In addition, when using the photon counting-type X-ray detector, the acquisition conditions may include at least an energy threshold for counting photons. Since the photon counting-type X-ray detector outputs images showing different transmission characteristics for each energy band according to the set energy threshold, it is convenient in principle because the imaging position of the inspection object W is the same in the low-energy image and the high-energy image. That is, the images obtained by photon counting-type X-ray detectors are included in the expression “having the same position information”.
As shown in
The control unit 4 comprehensively controls each unit for inspecting a quality state of the inspection object W, and includes an image storage unit 21, a learned model 22, an image processing unit 23, and a determination unit 24. Various parameters for operating the transport unit 2, the inspection unit 3 (image acquisition unit 3), and the display unit 5 are stored, and various data are processed. Particularly, as for the operation of the transport unit 2 or the inspection unit 3 (image acquisition unit 3), an inspection parameter for each type of the inspection object W to be inspected is set in association with a product type number showing each product type.
The inspection parameter includes the learned model 22 (calculation formula) applied to an image obtained when the inspection object W is captured, or a threshold for determining a quality state.
The image storage unit 21 stores the inference image i5 acquired by the image acquisition unit 3 using the inspection parameter of the product type set for the inspection object W to be inspected. As described above, the inference image i5 consists of a 3ch pseudo RGB image, in which for example, the low-energy image is regarded as R component of the RGB color image, the high-energy image is regarded as G component of the RGB color image, and the difference image (difference between the low-energy image and the high-energy image) is regarded as B component of the RGB color image.
The learned model 22 is a trained model created by executing a learning phase, which will be described later, on an external terminal device such as a personal computer, using the 3ch pseudo RGB image based on the learning image. This learned model 22 is used when calculating a foreign matter degree: K as a defective quality degree of the inference image i5 of the inspection object W in an inference phase to be described later. A foreign matter degree is a value showing a probability that the foreign matter is present. Data transfer of the learned model 22 between the inspection device 1 and the external terminal device is performed, for example, via a network or an external storage medium such as a USB memory.
The image processing unit 23 executes image processing for predetermined determination processing on image data of the inference image i5 (3ch pseudo RGB image) stored in the image storage unit 21, based on the learned model 22 corresponding to a type of product to be inspected (learned model created from a learning image under the same acquisition conditions as the inference image i5).
The determination unit 24 executes the determination processing of a quality state of the inspection object W (for example, determination processing of presence/absence of a foreign matter) based on the image data processed by the image processing unit 23.
Although not shown, the control unit 4 includes transport control means for controlling a transport speed or transport interval of the inspection object W by the transport belt 11 of the transport unit 2, and detection control means for controlling an X-ray irradiation intensity or irradiation period of the inspection unit 3 or controlling an X-ray detection period a detection period of each inspection object W using the X-ray line sensor of the X-ray detector 15 according to the transport speed of the inspection object W.
The display unit 5 includes various displays such as a liquid crystal display, and displays and outputs a determination result of the determination unit 24.
Next, a learning phase executed to create the learned model 22 will be described with reference to a flowchart of
In the learning phase, learning is performed using the 3ch pseudo RGB image consisting of an image having the same format as the inference image i5, that is, three transmission images having the same position information and different transmission characteristics (for example, R component: low-energy image, G component: high-energy image, and B component: difference image). This learning is performed for each product type of the inspection object W to be inspected.
As shown in
Next, a learning foreign matter synthesis image i4 and a learning foreign matter label r are created (ST2). In creating the learning foreign matter synthesis image i4 and the learning foreign matter label r, when a foreign matter detection algorithm is created using deep learning, an NG image (an image including a foreign matter) and coordinate data (labels) of a foreign matter are required as learning teacher data.
The coordinate data (label) of a foreign matter herein is data indicating a position of a foreign matter portion on the NG image, and refers to text file data in which the top left X-Y coordinates and the bottom right X-Y coordinates are recorded when a frame circumscribing the image of the foreign matter portion is created.
When the coordinate data (label) of a foreign matter is extracted from the NG image, it is required to manually specify a coordinate position of many foreign matters by a person, which is not realistic in deep learning for requiring thousands of images.
Therefore, in this example, the learning foreign matter synthesis image i4 is created by synthesizing the images using the non-defective image and a foreign matter image captured individually, and simultaneously, the coordinate data (label) of the foreign matter portion is created as the learning foreign matter label r, thus achieving work efficiency.
Therefore, a database image of a foreign matter obtained by cutting a region where the foreign matter is shown is created in advance from the 50 images of only foreign matter (one piece of bone) acquired in advance, and when the database image is synthesized with the non-defective image, one or a plurality of foreign matters are randomly selected from the foreign matters of the database image.
Then, when the database image of a foreign matter is synthesized with the non-defective image, a total of 1,000 foreign matter synthesis images are created by setting the foreign matter position to a random position on non-defective work and randomly fluctuating the foreign matter rotation angle from 0 to 360 degrees.
When creating a foreign matter synthesis image, both the low-energy image and the high-energy image consisting of the X-ray image are used, and the X-ray image is extended to 3ch to create the foreign matter synthesis image in the format of the 3ch pseudo RGB image, in which for example, the low-energy image is regarded as R component of the RGB color image, the high-energy image is regarded as G component of the RGB color image, and the difference image (difference between the low-energy image and the high-energy image) is regarded as B component of the RGB color image.
When the database image of a foreign matter is synthesized with the non-defective image, each image regarded as the same component of R, G, or B is respectively used. In addition, when synthesizing the database image of a foreign matter with the non-defective image, a foreign matter label can be obtained from the specified coordinate data of the database image of a foreign matter by a coordinate position in which the database image of a foreign matter is arranged. Furthermore, when the database image of a foreign matter is rotated, the coordinate data of the database image of a foreign matter is obtained by coordinate transformation.
Then, machine learning of the foreign matter synthesis image is performed using a plurality of pseudo RGB images created by the above method to create a learned model (ST3). That is, learning is performed by performing a convolution operation on the foreign matter synthesis image using the plurality of pseudo RGB images and outputting the resulting foreign matter degree (probability of foreign matter) K to create a learned model 22 consisting of a calculation formula for producing a better result.
Specifically, learning is performed using 1,000 pseudo RGB images (learning foreign matter synthesis image i4 extended to 3ch) and 1,000 coordinate data of a foreign matter (learning foreign matter label r) corresponding to respective pseudo RGB images, to create the learned model 22. In this case, when an image of a workpiece (with foreign matter) is input, learning is performed to output a foreign matter position (top left X-Y coordinates and bottom right X-Y coordinates) and a foreign matter degree K (0<K<1.0) (as the value of K is increased, it is regarded as a foreign matter).
Next, an inference phase for inferring a quality state of the inspection object W in the above inspection device 1 will be described with reference to a flowchart of
As shown in
Then, the image processing unit 23 calculates the foreign matter position (top left x-y coordinates and bottom right x-y coordinates) and the foreign matter degree (0<K<1.0) (as the value of K is increased, it is regarded as a foreign matter) based on the learned model created during the learning phase (ST12). That is, as shown in
Then, a determination threshold on a side of the inspection device 1 is set to S in advance, and when K≤S, the determination is output as OK (ST14), and when K>S, the determination is output as NG (ST15). More specifically, as shown in
In an example of
Meanwhile, in the above embodiment, the X-ray image consisting of the low-energy image (R component), the high-energy image (G component), and the difference image (B component) has been described as an example of the 3ch pseudo RGB image instead of the RGB color image, but the embodiment is not limited thereto. The 3ch pseudo RGB images may be three transmission images having the same imaging position information with respect to the inspection object W and having different transmission characteristics. For example, the 3ch pseudo RGB images may be a combination of an X-ray image obtained by converting a density of the image by a lookup table coefficient, an X-ray image before conversion, and a near-infrared image. In addition, an image of each channel of the pseudo RGB image may be subjected to image processing such as smoothing and standardization, and may correct a size, direction, and deviation of the image determined by positions and directions of an X-ray source or a light source, an optical component such as a lens, and the detector, as pre-processing. Furthermore, three inspection images, which are obtained by capturing the inspection object W with a multispectral camera (spectral camera) and spectroscopy of light (energy) transmitting through the inspection object W, can be used.
In addition, in the above embodiment, a case of determining presence/absence of the foreign matter as a quality state of the inspection object W has been described, but the embodiment is not limited thereto, and it is possible to determine, as a quality state, presence/absence of missing products in the inspection object W, pass or fail of a shape, size, and storage state of contents, density, thickness, volume, or mass distribution, or the like.
Hereinafter, a case of determining the presence/absence of a shape defect of contents in the inspection object W as a quality state will be described as an example. The determination of presence/absence of a shape defect of contents in the inspection object W basically has the same structure as the case of determining presence/absence of the foreign matter described above, in which types of the NG determination are increased to two or more types (types of shape defect (chipping and bending)), the determination threshold can also be set individually according to the types of shape defect (chipping and bending), and a combination with the above foreign matter inspection can also be used.
As in the case of determining presence/absence of the foreign matter, as a learning shape defect synthesis image from the learning image, an X-ray image obtained by converting the density (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch), the X-ray image before converting the density (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch), and a near-infrared image (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch) are created, and a learning shape defect label is created.
Then, one pseudo RGB image (R component: after converting the density, G component: before converting the density, and B component: near-infrared) (3ch) and coordinate data of the shape defect are created from one X-ray image obtained by converting the density (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch), one X-ray image before converting the density (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch), and one near-infrared image (synthesis of the shape defect (chipping and bending) with the non-defective product)_(1ch), and for example, as shown in
As in the learning phase described with reference to
Specifically, learning is performed using a plurality of pseudo RGB images (learning shape defect synthesis image i4 extended to 3ch) and coordinate data of the shape defect (learning shape defect label r) corresponding to respective pseudo RGB images. In this case, when an image of a workpiece (with shape defect) is input, learning is performed to output a shape defect position (top left X-Y coordinates and bottom right X-Y coordinates) and a shape defect degree K (0<K<1.0) (as the value of K is increased, it is regarded as a shape defect).
Then, when the shape defect of the inspection object W is inspected, the inference phase shown in
Then, the image processing unit 23 calculates the shape defect position (top left x-y coordinates and bottom right x-y coordinates) and the shape defect degree (0<K<1.0) (as the value of K is increased, it is regarded as a shape defect) based on the learned model 22 created during the learning phase (ST22). That is, as shown in
Then, a determination threshold on a side of the inspection device 1 is set to S in advance, and when K≤S, the determination is output as OK (ST24), and when K>S, the determination is output as NG (ST25). More specifically, determination thresholds S1 and S2 are set in advance on a side of the inspection device 1, and the calculated shape defect degrees K1 and K2 are compared with the determination thresholds S1 and S2. Here, as shown in
In addition, as shown in
The determination thresholds S1 and S2 can be set individually for each type of the shape defect (for example, chipping and bending). In addition, the above identification display may identify the NG determination location and a normal location, such as displaying the NG determination location by color and displaying the NG determination location by lighting/blinking, in addition to enclosing the NG determination location, and the display form is not limited thereto.
As described above, according to the present embodiment, since as an image instead of the 3ch RGB color image, the learned model is created from three inspection images having different transmission characteristics of the inspection object in transmission-based imaging, it is possible to improve inspection accuracy of a quality inspection of the inspection object by efficiently applying a learning library of the existing RGB color image, learning determination using the 3ch image corresponding to the RGB color image, and comparing determination of conventional one grayscale image to increase an amount of information for determination. When presence/absence of the foreign matter is inspected as a quality state for example, the learned model is created from three inspection images having the same position information and different transmission characteristics for the inspection object, and thus not only an appearance of the foreign matter differs depending on the type of the inspection object, but also information indicating a relationship between the foreign matter and the surroundings thereof (background) is increased, such that the foreign matter can be inspected with higher accuracy.
Although the best mode of the inspection device, learned model generation method, and inspection method according to the present invention have been described above, the present invention is not limited by the description and the drawings according to this mode. That is, it is a matter of course that other modes, examples, operation techniques, and the like made by those skilled in the art based on this mode are all included in the scope of the present invention.
| Number | Date | Country | Kind |
|---|---|---|---|
| 2022-017365 | Feb 2022 | JP | national |
