Vial inspection apparatus and method based on hyperspectral image and AI model

Abstract

According to the present invention, by using hyperspectral images and machine learning models, rather than simply determining whether an arbitrary detection area is normal or abnormal, the invention determines whether the detection area is normal or abnormal in a first inspection and performs a second inspection on the area determined to be abnormal to determine the type of abnormality, thereby providing a technical effect of more accurately and quickly distinguishing between normal and defective products.

Description

Technology Field

[0001] The present invention relates to a quality control (QC) method for a product manufacturing process, and more specifically, to an inspection device and method for determining whether a product manufactured in a product manufacturing process is normal or abnormal based on hyperspectral images and an artificial intelligence model. Background Technology

[0002] In general, it is essential for products manufactured in industries such as pharmaceuticals and beverages to undergo quality control (QC) immediately after production. For example, since drugs such as various blood products produced in the pharmaceutical industry are injected directly into the human bloodstream, it is required to determine whether they are defective during the quality control process after production. Typically, these drugs are produced in small glass bottles called vials, and in the conventional QC process, workers manually determined the condition of each vial. In other words, previously, workers visually inspected the vial to be tested within approximately 1 to 2 seconds to check for foreign substances and determined whether it was normal or defective accordingly.

[0003] However, in this conventional method, operators manually inspect a large number of vials, which can lead to errors in determining whether a vial is normal or defective and results in significant variations depending on the operator. Furthermore, cross-verification is required to prevent this, incurring additional costs. Additionally, the Ministry of Food and Drug Safety has recently recommended increasing the inspection time for manual vials from 1 second to 5 seconds or more per vial, raising concerns about increased inspection time and consequent additional costs in the future.

[0004] In addition, although there is a method for detecting abnormalities or defects using cameras, it identifies foreign substances using only a single wavelength rather than multiple wavelengths. This frequently leads to cases where non-foreign substances are identified as foreign or foreign substances are deemed normal, resulting in reduced inspection accuracy. Furthermore, even when a substance is identified as a foreign substance, it is difficult to specifically identify what that substance is. The problem to be solved

[0005] The present invention aims to provide an inspection method and apparatus capable of rapidly and accurately determining whether a product is normal or defective by replacing conventional manual product inspection with hyperspectral imaging and AI-based inspection. means of solving the problem

[0006] According to one embodiment of the present invention, a product inspection method based on a hyperspectral image and an artificial intelligence model is disclosed, comprising: a step of receiving a hyperspectral image obtained by photographing a product to be inspected with a hyperspectral camera; a first inspection step of detecting an abnormal region in the hyperspectral image using a first machine learning model; and a second inspection step of determining the type of abnormality found in the first inspection step using a second machine learning model.

[0007] According to one embodiment of the present invention, a computer-readable recording medium is disclosed having a computer program for executing the product inspection method. Effects of the invention

[0008] According to the present invention, by using hyperspectral images and machine learning models, rather than simply determining whether an arbitrary detection area is normal or abnormal, the invention determines whether the detection area is normal or abnormal in a first inspection and performs a second inspection on the area determined to be abnormal to determine the type of abnormality, thereby providing a technical effect of more accurately and quickly distinguishing between normal and defective products. Brief explanation of the drawing

[0009] FIG. 1 is a flowchart illustrating a vial inspection method according to an embodiment of the present invention, FIG. 2 is a drawing showing an exemplary vial imaging image, FIG. 3 is a diagram illustrating an exemplary detection area, FIG. 4 is a diagram explaining an autoencoder, FIG. 5 is a flowchart illustrating a first inspection step according to one embodiment, FIG. 6 is a flowchart illustrating a secondary inspection step according to one embodiment, FIG. 7 is a diagram showing an exemplary hyperspectral spectrum of the normal region, FIG. 8 is a diagram showing an exemplary hyperspectral spectrum of an anomalous region, FIG. 9 is a drawing showing an albumin sample captured with a hyperspectral camera. FIG. 10 is a diagram showing values ​​calculated based on the spectral similarity of albumin data in the form of a heatmap. FIG. 11 is a diagram illustrating a vial inspection method according to an alternative embodiment. Specific details for implementing the invention

[0010] The above objects, other objects, features, and advantages of the present invention will be easily understood through the following preferred embodiments associated with the accompanying drawings. However, the present invention is not limited to the embodiments described herein and may be embodied in other forms. Rather, the embodiments introduced herein are provided to ensure that the disclosed content is thorough and complete, and to ensure that the spirit of the invention is sufficiently conveyed to those skilled in the art.

[0011] Where terms such as "first," "second," etc. are used in this specification to describe components, these components shall not be limited by such terms. These terms are used merely to distinguish one component from another. The embodiments described and illustrated herein also include complementary embodiments.

[0012] In this specification, the singular form includes the plural form unless specifically stated otherwise in the text. Expressions used in this specification such as “comprising,” “consisting of,” and “consisting of” do not exclude the presence or addition of one or more other components in addition to the mentioned components.

[0013] In this specification, the term 'software' refers to technology that operates hardware in a computer, the term 'hardware' refers to a type of device or apparatus (CPU, memory, input device, output device, peripheral device, etc.) that constitutes a computer, the term 'step' refers to a series of processes or operations connected in a time series to achieve a predetermined objective, the terms 'computer program', 'program', or 'algorithm' refer to a set of instructions suitable for processing by a computer, and the term 'program recording medium' refers to a computer-readable recording medium that records a program used to install, execute, or distribute the program.

[0014] Terms such as 'part', 'module', 'unit', 'block', 'board', etc., used in this specification to refer to the components of the invention may mean a physical, functional, or logical unit that processes at least one function or operation, and may be implemented by one or more hardware, software, or firmware, or by a combination of hardware, software, and / or firmware.

[0015] In this specification, 'processing device', 'computer', 'computing device', 'server device', and 'server' may be implemented as a system equipped with an operating system such as Windows, Mac, or Linux, a computer processor, memory, an application program, and a storage device (e.g., HDD, SSD). A computer may be a device such as, for example, a desktop computer, a laptop, or a mobile terminal, but these are exemplary and are not limited thereto. A mobile terminal may be one of a mobile wireless communication device such as a smartphone, a tablet PC, or a PDA.

[0016] The present invention will be described in detail below with reference to the drawings. In describing the specific embodiments below, various specific details have been included to explain the invention more specifically and to aid in understanding. However, a reader with sufficient knowledge in the art to understand the invention will recognize that it can be used without these various specific details. Furthermore, it should be noted in advance that in describing the invention, parts that are known or commonly used technologies and are not significantly related to the invention are omitted to prevent confusion in explaining the invention.

[0017] In addition, in the following detailed description with reference to the drawings, a vial product containing a blood product is described as an example, but those skilled in the art will understand that the testing device and testing method according to the present invention are not limited thereto and can be applied to various products.

[0018] For example, the inspection device and inspection method according to the present invention are not limited to vials containing blood products, but may be applied to products such as liquids (e.g., various beverages), gases, solids, or gels contained in containers made of transparent or translucent materials such as glass or transparent plastic. In this case, the container may be a transparent or translucent container through which visible light passes, but is not limited thereto; any container capable of transmitting electromagnetic waves of a predetermined frequency band irradiated to obtain a hyperspectral image may be used.

[0019] In addition, the ‘foreign substance’ identified by the inspection device according to the present invention is defined in various ways depending on the specific embodiment of the present invention. For example, the ‘foreign substance’ mentioned in this specification may be a lump of protein, dust, or an insect depending on the specific embodiment of the invention, and is defined as ‘any substance or particle that should not be included or is preferable not to be included in the product being produced.’

[0020] FIG. 1 is a flowchart illustrating a vial inspection method according to an embodiment of the present invention. The flowchart of FIG. 1 may be performed by a data processing unit and an artificial intelligence model executed on a computer device. In one embodiment, the vial inspection method may include: a step of receiving a hyperspectral image of a vial captured by a hyperspectral camera (S10); a step of preprocessing the received hyperspectral image (S20); a first inspection step (S30) of detecting an abnormal region in the hyperspectral image using a first machine learning model; and a second inspection step (S50) of determining the type of abnormality found in the first inspection step using a second machine learning model.

[0021] At this time, if it is determined that there are no abnormalities as a result of detecting abnormalities in the hyperspectral image of the vial in the first inspection step (S30), it is determined to be a normal vial and shipped out. If one or more abnormal (abnormal) regions are found in the first inspection step (S30), the second inspection step (S50) is performed for each abnormal region to determine the type of abnormality. If the result of the second inspection of the abnormality determines, for example, that it is a simple protein fragment or a scratch on the surface of the vial, it is determined to be within the normal range and shipped out as normal, and if it is determined to be a foreign substance inside the vial, it is finally determined to be a defective vial.

[0022] The vial inspection method according to the present invention is performed by a data processing unit and an artificial intelligence model, etc., executed on a computer device. For example, the computer device may include a data preprocessing unit that performs data preprocessing (S20) and an artificial intelligence model that performs inspection steps (S30, S50) by first and second machine learning models, and each of these components may be implemented as software that is programmed to be executable on the computer device (or combined with firmware, hardware, etc. as necessary).

[0023] Now, the specific steps of the vial inspection method of FIG. 1 will be explained below with reference to FIG. 2 to FIG. 9.

[0024] First, in step (S10) of FIG. 1, the computer device receives a hyperspectral image of a vial to be inspected, captured by a hyperspectral camera. A hyperspectral image is an image that combines spatial information and spectral information for a captured image to include three-dimensional information for each pixel. For example, the hyperspectral image includes data in frequency bands such as ultraviolet (UV), near-infrared (NIR), and short-wave infrared (SWIR) as well as the visible light region of the object to be inspected (e.g., data on the intensity of the reflection spectrum).

[0025] After receiving a hyperspectral image of a subject to inspection, the computer device can perform preprocessing on the image (S20). Preprocessing is performed to remove noise from the data or to speed up computer processing in subsequent inspection steps. For example, since hyperspectral images have a large amount of data, it may be desirable to compress the data before applying it to a machine learning model, and to this end, the data can be compressed by performing principal component analysis (PCA) on the hyperspectral image in the preprocessing step (S20).

[0026] The hyperspectral image contains hyperspectral spectrum information in units of a predetermined detection area of ​​a predetermined size, and the subsequent first inspection step (S30) and second inspection step (S50) determine whether the area is abnormal or normal in units of these predetermined detection areas.

[0027] For example, FIG. 2 is a drawing showing an exemplary vial image, and an exemplary visible light image is shown. In this image, for example, areas (a) and (b) are normal areas without foreign substances or scratches, and areas (c) and (d) are abnormal areas where scratches or foreign substance shapes are visible. FIG. 3 is an enlarged view of area (a) of FIG. 2, and is assumed to have an exemplary size of 5x5 pixels.

[0028] At this time, each pixel, that is, an area of ​​1x1 pixel size, can be set as a first detection area (100), and a hyperspectral spectrum (10) can be obtained for each first detection area (100). That is, in FIG. 3, a total of 25 first detection areas (100) are shown, and the hyperspectral spectrum (10) of the corresponding detection area is displayed by overlapping it with some of the first detection areas.

[0029] Meanwhile, since abnormalities appearing in the vial, such as glass scratches, protein coagulation, and foreign substances, appear over an area much larger than a single first detection area (100), a second detection area larger than the first detection area may be required to cover a single abnormality. For example, in FIG. 3, the second detection area (200) is set to an area consisting of 5x5 pixels, and in this case, it will be understood that 25 first detection areas (100) constitute one second detection area (200).

[0030] However, it goes without saying that the sizes of the first detection area (100) and the second detection area (200) described above may vary depending on the specific embodiment. For example, the first detection area (100) may be defined as 3x3 pixels or 5x5 pixels, and the second detection area (200) may be defined as 5x5 first detection areas (100) in width and height, or as 10x10 first detection areas (100). However, for convenience of explanation in this specification, it is assumed that the first detection area (100) is 1x1 pixel in size and the second detection area (200) is 5x5 pixel in size.

[0031] Referring again to FIG. 1, a first inspection step (S30) using a first machine learning model can be performed after data preprocessing. In one embodiment, the first machine learning model is a model that receives data of a hyperspectral image in a first detection area unit and determines whether it is normal, and can be implemented, for example, as an autoencoder.

[0032] Figure 5 is a schematic diagram illustrating the configuration of an autoencoder. An autoencoder is a type of machine learning model that reduces the dimensionality of input data, compresses it, and then restores it to its original size. Referring to Figure 5, an autoencoder generally consists of an encoder and a decoder. The encoder is a model that reduces the dimensionality of input data to compress it into a latent vector form and outputs it, while the decoder is a model that receives the latent vector as input, upsamples it back to its original size, and outputs the original data.

[0033] If an autoencoder is trained using only normal data and then inputs abnormal (abnormal) data, the autoencoder's output will not be fully restored to normal data, so the restoration error will inevitably increase; therefore, if the restoration error of a data exceeds a predetermined threshold, that data can be determined as abnormal.

[0034] FIG. 6 is a flowchart illustrating a first inspection step (S30) using an autoencoder. Referring to FIG. 6, in step (S310), a hyperspectral spectrum in the first detection area unit is input to the autoencoder as input data. The autoencoder performs encoding and decoding processing on the input data and outputs restored data. In step (S320), the output data of the autoencoder is compared with the input data to determine whether there is an abnormality in the hyperspectral spectrum in the first detection area unit. That is, if the error between the input spectrum and the output spectrum is below a preset threshold, the corresponding detection area is considered normal, and if it exceeds the threshold, it is determined that there is an abnormality in the corresponding detection area.

[0035] Referring again to FIG. 1, for the first detection area (100) identified as an anomaly as a result of the inspection in the first inspection step (S30) using the first machine learning model, a second inspection step (S50) using the second machine learning model can be performed. In one embodiment, the second inspection step (S50) is a step of determining the type of anomaly found in the first inspection step using the second machine learning model.

[0036] In one embodiment, if any first detection area (100) is determined to be abnormal, a second inspection is performed on a second detection area (200) covering this area. For example, if multiple adjacent first detection areas (100) are determined to be abnormal due to a single scratch or foreign substance in the first inspection step (S30), a second detection area (200) covering this scratch or foreign substance is selected. For example, in FIG. 2, area (c) and area (d) may each be a second detection area (200) covering the first detection areas (100) that were determined to be abnormal as a result of the first inspection.

[0037] In the second inspection step (S50), a machine learning model that classifies images using spectral similarity values ​​can be used to determine the type of anomaly. For example, in the second inspection step (S50) according to one embodiment, a convolutional neural network (CNN) model is used to determine the type of anomaly. A convolutional neural network (CNN) is a machine learning model useful for finding and classifying image recognition patterns. Generally, a CNN model can classify images by repeating a convolution layer and a pooling layer one or more times after the input layer, and then passing through a fully-connected layer and a softmax function.

[0038] FIG. 6 is a flowchart illustrating an exemplary method of a second inspection step (S50) according to one embodiment.

[0039] In step (S510), for a second detection area covering a first detection area determined to be abnormal in the first inspection step (S30), a hyperspectral spectrum representing the detection area (hereinafter also simply referred to as the “representative hyperspectral spectrum”) is determined.

[0040] The representative hyperspectral spectrum is determined using the hyperspectral spectrum of each first detection area (100) within the corresponding second detection area (200). For example, one of the hyperspectral spectra of all first detection areas (100) within the second detection area (200) may be selected as the representative hyperspectral spectrum, or a new hyperspectral spectrum may be generated using the hyperspectral spectra of all first detection areas (100) within the second detection area (200) and designated as the representative hyperspectral spectrum. For example, the mode or average value in each wavelength band may be calculated using the hyperspectral spectra of all first detection areas (100) within the second detection area (200), and the representative hyperspectral spectrum may be selected from therein.

[0041] In this regard, FIGS. 7 and 8 show exemplary graphs of the hyperspectral spectrum when the second detection area (200) is normal and when it is abnormal.

[0042] FIGS. 7(a) and FIGS. 7(b) respectively show enlarged hyperspectral spectra of different first detection regions determined to be normal. For example, FIG. 7(a) shows the hyperspectral spectra of each of the first detection regions (100) within an arbitrary second detection region (200) overlaid on a single graph, and also shows a representative hyperspectral spectrum (11) selected (or newly generated) from the hyperspectral spectra of all the first detection regions (100) thus overlapped.

[0043] For example, as shown in FIG. 3, the first detection area (100) is 1X1 pixel and the second detection area (200) is 5X5 pixel and includes 25 first detection areas (100). The second detection area (200) includes 25 hyperspectral spectra, and FIG. 7(a) shows the 25 hyperspectral spectra for any second detection area (200) overlaid and displayed, and the graph among them marked with a red line is the representative hyperspectral spectrum (11).

[0044] FIGS. 8(a) and FIGS. 8(b) show typical spectra of scratches on a glass vial among the abnormalities. For example, when the first detection area (100) is 1x1 pixels and the second detection area (200) is 5x5 pixels and includes 25 first detection areas (100), FIG. 8(a) shows the area (15) represented by overlapping 25 hyperspectral spectra for any second detection area (200), and the red line represents a representative hyperspectral spectrum (11) selected using the 25 hyperspectral spectra. FIGS. 8(c) and FIGS. 8(d) show typical spectra of foreign substances inside the vial among the abnormalities.

[0045] As can be seen in FIGS. 7 and FIGS. 8, the distribution and deviation (dispersion) of the hyperspectral spectrum of the normal region and the hyperspectral spectrum of the abnormal region are different, and the distribution and degree of deviation of the hyperspectral spectra of the second detection region (200) are also different depending on the type of abnormality (e.g., scratches, foreign substances, etc.).

[0046] Meanwhile, FIG. 9 is a diagram showing an albumin sample captured by a hyperspectral camera as another example. In FIG. 9, the left area is an image captured by the camera, and in this image, two abnormal areas are marked with red circles, and a representative hyperspectral spectrum for each abnormal area is shown on the right. Of the two abnormal areas, the upper area was found to be a scratch and the lower area was found to be a foreign substance. At this time, it can be seen that the hyperspectral spectra (11a, 11b) of each abnormal area are not only different when compared to the normal hyperspectral spectrum (20), but also show different distributions between the hyperspectral spectra (11a, 11b) of the two abnormal areas.

[0047] Referring to FIG. 6, in step (S520), a spectral similarity value is generated using the representative hyperspectral spectrum and normal spectrum calculated in step (S510). In the second inspection step (S50), the type of the anomalous is determined using a machine learning model that classifies images (hereinafter also referred to as the 'second machine learning model'). In this case, as input data to be input to the second machine learning model, in one embodiment of the present invention, the spectral similarity value of the image spectrum is used.

[0048] The spectral similarity value can be calculated by a known method that numerically scales the similarity (e.g., brightness of light, similarity of spectrum shape, etc.) between the representative hyperspectral spectrum of the second detection area and a predefined normal hyperspectral spectrum. In one embodiment, the second inspection step (S50) uses a heatmap as the spectral similarity value. The spectral similarity-based heatmap according to the present invention visualizes the similarity between the representative hyperspectral spectrum of the second detection area and a predefined normal spectrum using color. The spectral similarity-based heatmap according to the present invention can be generated using a known method, such as a correlation heatmap, which visualizes the correlation between two variables.

[0049] For example, FIG. 10 is a diagram showing values ​​calculated based on the spectral similarity of albumin data in the form of a heatmap, FIG. 10(a) and FIG. 10(b) show a heatmap of a normal albumin sample, FIG. 10(c) shows a heatmap of a case where there is a scratch on the outer surface of the vial, and FIG. 10(d) shows a heatmap of a sample containing foreign substances inside the vial, respectively.

[0050] Referring again to FIG. 6, in step (S530), when a spectrum similarity value (e.g., a spectrum similarity-based heatmap) for the second detection area (200) to be inspected is calculated, it is input into a second machine learning model to determine the type of abnormality of the second detection area (200).

[0051] For example, the second machine learning model is a Convolutional Neural Network (CNN) model, and spectral similarity values ​​(e.g., a similarity-based heatmap) are input into the CNN as input data, and the CNN model outputs the results of the above classification as output data.

[0052] If the reason for the abnormality determined as a result of this second inspection (S50) is due to a simple protein fragment or a scratch on the surface of the vial, it is determined to be within the normal range and released as normal (S60_Yes in Fig. 1), and if the reason for the abnormality is determined to be a foreign substance inside the vial, it is finally determined to be a defective vial (S60_No in Fig. 1).

[0053] As such, the vial inspection method according to the present invention can determine whether an abnormal area is a scratch or a foreign substance by analyzing the hyperspectral spectrum of the abnormal area, and furthermore, can identify the type of foreign substance. That is, according to the present invention, by utilizing hyperspectral images and a machine learning model, rather than simply determining whether an arbitrary detection area is normal or abnormal, the normality or abnormality of the detection area is determined in a first inspection, and a second inspection is performed on the area determined to be abnormal to determine the type of abnormality, thereby having the technical effect of more accurately and quickly distinguishing normal / defective vials.

[0054] Meanwhile, AUROC (Area Under the Receiver Operating Characteristic Curve) is one of the important indicators used to evaluate the performance of a classification model, and it is a criterion indicating how well the model performs binary classification. The vial inspection method according to the present invention was derived after collecting 48,000 pixel-level spectral data from a total of 31 albumin samples and conducting multiple tests, and the AUROC value reaches an average of 96%. This means that the accuracy of foreign substance identification is very high.

[0055] FIG. 11 is a diagram illustrating a vial inspection method according to an alternative embodiment. According to the alternative embodiment, normal and abnormal can be classified based on a model that classifies foreign substance types in a single step without undergoing two stages of inspection (S30, S50) as in the above-described embodiment. For example, a method that best represents the spectrum of each foreign substance type is applied, such as a method that utilizes the probability distribution distance between the normal spectrum and the foreign substance spectrum as a measure of similarity, like SID, or a preprocessing method such as PCA.

[0056] For example, referring to the exemplary method of FIG. 11, assuming the first detection area (100) is 1x1 pixels, in step (a), the hyperspectral image data is compressed by performing dimensionality reduction by Principal Component Analysis (PCA) on the first detection area (100) unit. Next, in step (b), a 3D-Convolutional Network technique is applied to learn all pixel-specific hyperspectral features well, and then in step (c), a 2D-Convolutional Network is applied to learn the relationship between the target pixel and surrounding pixels, i.e., spatial features, based on the representative spectral shape of each pixel and the surrounding pixel spectra. Finally, in step (d), classification is performed, thereby probabilistically determining which of the predefined multiple types related to foreign object detection the final representation output from the classification step is.

[0057] As described above, those skilled in the art to which the present invention pertains will understand that various modifications and variations are possible from the description in this specification. Therefore, the scope of the present invention should not be limited to the described embodiments, but should be defined by the claims set forth below as well as equivalents thereof. Explanation of the symbols

[0058] 10, 11: Hyperspectral spectrum 100: 1st detection area 200: Second detection zone

Claims

Claim 1 A product inspection method based on a hyperspectral image and an artificial intelligence model, comprising: a step (S10) of receiving a hyperspectral image obtained by photographing a product to be inspected with a hyperspectral camera; a first inspection step (S30) of detecting an abnormal region in the hyperspectral image using a first machine learning model; and a second inspection step (S50) of determining the type of abnormality in the abnormal region using a second machine learning model, from a similarity map between the hyperspectral spectrum of a second detection region corresponding to the first detection region detected as the abnormal region in the first inspection step and the hyperspectral spectrum of a predetermined normal region. Claim 2 A product inspection method based on a hyperspectral image and an artificial intelligence model, characterized in that, in claim 1, it further includes a step (S20) of performing preprocessing on the hyperspectral image by performing principal component analysis (PCA) on the hyperspectral image received in step (S10) and compressing the hyperspectral image. Claim 3 A product inspection method based on a hyperspectral image and an artificial intelligence model according to claim 1, wherein the first inspection step (S30) comprises: a step (S310) of inputting a hyperspectral spectrum of a preset first detection area unit as input data to a first machine learning model; and a step (S320) of determining whether there is an abnormality in each detection area by comparing the output of the first machine learning model with the input data. Claim 4 A product inspection method based on hyperspectral images and an artificial intelligence model, characterized in that, in claim 3, the step of determining whether there is an abnormality in each detection area (S320) includes the step of determining whether there is an abnormality in each detection area based on whether the restoration error of the first machine learning model, determined by the difference between the input data and the output, exceeds a predetermined threshold value (S320). Claim 5 A product inspection method based on a hyperspectral image and an artificial intelligence model according to claim 3, wherein the second inspection step (S50) comprises: a step (S510) of determining a representative hyperspectral spectrum representing a second detection area covering a first detection area determined to be abnormal in the first inspection step; a step (S520) of generating a spectral similarity value between the representative hyperspectral spectrum and a normal spectrum; and a step (S530) of inputting the spectral similarity value into a second machine learning model to determine the type of abnormality of the second detection area, wherein at least a portion of the first detection area and the second detection area are different from each other. Claim 6 A product inspection method based on a hyperspectral image and an artificial intelligence model, characterized in that, in claim 5, the representative hyperspectral spectrum is determined as the hyperspectral spectrum of any one of the plurality of first detection regions included in the second detection region; or the statistical value of the hyperspectral spectra of the plurality of first detection regions included in the second detection region. Claim 7 A product inspection method based on hyperspectral images and an artificial intelligence model, characterized in that, in claim 1, the first machine learning model is an autoencoder and the second machine learning model is a convolutional neural network (CNN) model. Claim 8 A computer-readable recording medium having a computer program recorded thereon for executing a product inspection method according to any one of paragraphs 1 through 7.

Images