Article inspection device

Abstract

An article inspection device is provided with: an image storage unit for storing an inspection image obtained by capturing an image of a conveyed article; and a determination unit that determines, as reliability, an inference value relating to the quality state for each predetermined unit region of the inspection image by using a learning model that learns using an image data set having the same imaging condition as the image, and determines the quality state of the article by comparing the value with a threshold value. The article inspection device is provided with: an inference value image generation unit that generates an inference value image having an inference value-equivalent pixel density corresponding to an inspection image; a projective image generation unit that generates a projective image by projecting the maximum value of the inference value equivalent density for each of a plurality of line image regions in the y-axis direction in the inference value image on the x-axis; and a display control unit that adds a line corresponding to the threshold value to the projective image and displays the projective image on a display unit.

Description

TECHNICAL FIELD

[0001] The present application relates to an article inspection apparatus, and particularly to an article inspection apparatus that inspects a quality state of an article using an inspection image obtained by photographing an article to be inspected and a learned model. BACKGROUND

[0002] Recently, an article inspection apparatus is known that applies a machine-learned model, so-called AI (Artificial Intelligence) model, to an inspection image or a sensor signal including a feature quantity corresponding to a quality state of an article, inspects the quality state of the article, and displays and outputs an image including inspection result information as the inspection image.

[0003] As such an article inspection apparatus, for example, there is an apparatus that, in order to improve article inspection accuracy, photographs a plurality of images different from each other under a prescribed imaging condition corresponding to each input system, acquires image data of the plurality of images of the article to be inspected and stores it in an image storage section, on the other hand, prepares a learned model for inspection determination that is machine-learned using learning image data acquired under the same imaging condition as the image data of the article to be inspected stored in the image storage section, and determines the quality state of the article to be inspected by comparing a quality defect degree calculated by processing the image data of the article to be inspected acquired at the time of actual inspection with a threshold value set in advance, by the learned model for inspection determination (for example, refer to Patent Literature 1).

[0004] Further, as an article inspection apparatus using a rule-based model that performs determination or a task according to an instruction or a rule defined by a person in advance, there is also known an apparatus that photographs an article to be inspected in conveyance by line scanning of an X-ray linear sensor every prescribed time and acquires an X-ray transmission image of the article to be inspected, and that prepares an X-ray inspection image of the article to be inspected by performing filter processing or the like for the acquired image, on the other hand, displays a projection drawing that associates the maximum pixel value among the pixel values of the line scanning image every prescribed time and the corresponding position in the conveyance direction of the article to be inspected in the imaging range on the screen together with the X-ray inspection image, and displays a bar chart (for example, refer to Patent Literature 2).

[0005] Patent Literature 1: Japanese Patent Application Laid-Open No. 2023-114828

[0006] Patent Literature 2: Japanese Patent Application Laid-Open No. 2016-180712

[0007] In the conventional article inspection apparatus described above, sometimes an index of a statistical scale, that is, a confidence is used to indicate to what extent the AI estimates the estimation result of whether or not it is an object to be detected with reliability. In this case, if the confidence exceeds a threshold value set in advance, it can be determined that the article or the portion to be detected is detected, and thus the reliability of the determination result can be improved.

[0008] However, in the image inspection function using AI, the size of the estimated value of each pixel of the inspected object image is generally represented by a heat map, and in this case, even if the size of the approximate confidence is known, there is a problem that the information is insufficient to set a specific threshold value.

[0009] Also, in the case where the information of the confidence is displayed together with the rectangle in the AI-based object detection system, if the numerical value or the rectangle is overlapped too much with the display image, there is a disadvantage that it is difficult to judge the size of the confidence on the position on the article in the portion where the candidates of the detection objects are dense.

[0010] On the other hand, in the article inspection apparatus, it is common that the inspected article is conveyed while the article is inspected for abnormalities, and notification output is performed when an abnormality is detected, and thus it is desirable to clearly display on the screen which position on the conveying direction of the inspected article the abnormality is detected, but if the candidate images of the plurality of abnormal positions in the inspection image are displayed together with the confidence, there is a problem that it is not easy to visually recognize or grasp the size of the confidence or whether the confidence exceeds the threshold value, and the like.

[0011] In response to this, it is considered that if the X-ray inspection image of the inspected object is displayed moving along one coordinate axis direction corresponding to the conveying direction, and the maximum value of the pixel value of the line scan image is projected and displayed on one coordinate axis, it is easy to visually recognize at which position of the inspected object the abnormality is detected.

[0012] However, even in this case, in the case where the maximum pixel value in each line scan image (for example, an X-ray transmission image) of the inspected object approaches the threshold value, there is a problem to be solved that it can lead to difficulty in visually recognizing at which position on the conveying direction the abnormality is detected. SUMMARY

[0013] Therefore, an object of the present application is to provide an article inspection apparatus capable of easily visually displaying the detection object portion or its candidate at which position on the conveying direction the abnormality or the like is detected in the display screen of the inspection image, and also capable of clearly indicating the determination reference thereof.

[0014] To achieve the above object, (1) the article inspection apparatus according to the present application is characterized by comprising: an image storage section that stores an inspection image of an inspected article obtained by photographing the inspected article being conveyed; a determination section that, by using a learning model that is learned in advance using an image data set in which a photographing condition is the same as that of the inspected article, calculates an inference value related to a quality state of the inspected article for each unit region of a predetermined number of pixels of the inspection image stored in the image storage section, and determines the quality state of the inspected article by comparing the inference value with a threshold value set in advance; an inference value image generation section that generates an inference value image that is a two-dimensional density image in which the inference value corresponding to the inspection image is set as a density; a projection image generation section that generates a bar graph-like projection image in which a maximum value of the density of a plurality of line image regions in another coordinate axis direction adjacent to one coordinate axis direction in the inference value image is projected onto a corresponding line image region in the one coordinate axis direction in a length corresponding to the maximum value of the density; and a display control section that displays the projection image to which a line corresponding to the threshold value is added on a display section.

[0015] According to this structure, in the present application, an inference value related to a quality state of an inspected article is calculated for each unit region of a predetermined number of pixels in an inspection image, and a two-dimensional density image in which the inference value is set as a density is generated as an inference value image. Furthermore, for a plurality of line image regions in another coordinate axis direction adjacent to one coordinate axis direction in the inference value image, a maximum pixel value of each line image region is associated with a conveying direction of the inspected article or the like and is sequentially projected onto either coordinate axis and a projection image is displayed. Therefore, it is easy to visually recognize at a glance which position in the conveying direction of the inspected article or which position in a line scanning direction in which an abnormality is detected. As a result, it is possible to easily visually recognize and display a detection target portion in which an abnormality is detected or a candidate thereof in which an abnormality is detected in which position in the conveying direction, and it is also possible to explicitly indicate a determination reference.

[0016] In a preferred embodiment of the present application, (2) it can be configured such that the display control section displays the inspection image or the inference value image and the bar graph-like projection image on the display section after aligning positions in the one coordinate axis direction of the plurality of line image regions and the corresponding line image regions according to a conveying speed of the inspected article in the one coordinate axis direction.

[0017] In this case, in the display unit, the inspection image or inference value image and the bar chart-like projective image are aligned with the positions of each line image area along one coordinate axis according to the conveying speed of the inspected object in one coordinate axis direction. Therefore, it is possible to easily visually identify from the bar chart-like projective image with threshold display where an abnormality was detected in the inspected object being conveyed, or where no abnormality was detected at any position.

[0018] In a preferred embodiment of the present invention, (3) can be configured as follows: the display control unit moves the inspection image or the inference value image in the display unit according to the conveying speed of the inspected object in the direction of the first coordinate axis, and displays the inspection image or the inference value image and the bar-shaped projective image on both sides of the other coordinate axis direction of the inspection image.

[0019] In this case, in the display unit, when the inspection image or inferred value image is moved and displayed according to the conveying speed of the inspected object in one coordinate axis direction, the inspection image or inferred value image and its corresponding bar chart-like projective image are displayed on one side and the other side in another coordinate axis direction. Therefore, when the inspection image or inferred value image of the inspected object is moved and displayed, the corresponding bar chart-like projective image moves and displays in the same direction accordingly. For each inspected object, it is possible to easily visually identify whether there is an abnormality in the quality status from the bar chart-like projective image with threshold display.

[0020] In a preferred embodiment of the present invention, (4) can be configured as follows: the projective image generation unit generates at least one of the following projective images: a bar-shaped first projective image obtained by projecting the maximum concentration of a plurality of first line image regions adjacent to the other coordinate axis direction in the inference value image onto the corresponding first line image region on the one coordinate axis with a length corresponding to the maximum concentration; and a bar-shaped second projective image obtained by projecting the maximum concentration of a plurality of second line image regions adjacent to the other coordinate axis direction in the inference value image onto the corresponding second line image region on the other coordinate axis with a length corresponding to the maximum concentration.

[0021] In this case, the first projective image is projected onto one coordinate axis of the inferred value image, and the second projective image is projected onto the other coordinate axis of the inferred value image. Therefore, the first projective image and / or the second projective image can be associated with the inspection image or the inferred value image on either side of the vertical direction and / or on either side of the horizontal direction and displayed. For each inspected item, it is easy to visually identify whether there is an abnormality in the quality status or the location of the abnormality from the bar chart-like projective image with threshold display.

[0022] In a preferred embodiment of the present invention, (5) can be configured as follows: the display control unit aligns at least one of the inspection image or the inference value image and the first projection image and the second projection image with the position of any one or each coordinate axis direction corresponding to the first line image area and the corresponding first line image area and the second line image area and the corresponding second line image area, and then displays it on the display unit.

[0023] In this case, in the display unit, the inspection image or inference value image and the bar chart-like projective image are displayed after aligning multiple line image regions of the inspection image or inference value image and multiple corresponding line image regions of the first projective image to a position along one coordinate axis, and / or after aligning multiple line image regions of the inspection image or inference value image and multiple corresponding line image regions of the second projective image to a position along another coordinate axis. Therefore, it is possible to easily visually identify from the bar chart-like projective image with threshold display which inspected item in the transport, at what position, or whether no abnormality was detected at any position.

[0024] In a preferred embodiment of the present invention, (6) can be configured as follows: the display control unit moves the inspection image or the inferred value image in the display unit according to the conveying speed of the inspected object in the direction of the first coordinate axis, and displays at least one of the inspection image or the inferred value image and the first projection image and the second projection image in three adjacent display areas in the direction of the other coordinate axis of the inspection image or the inferred value image.

[0025] In this case, when the inspection image or inferred value image moves in the display unit according to the conveying speed of the inspected object in one coordinate axis direction, the inspection image or inferred value image and at least one of the first projective image and the second projective image are displayed in three adjacent display areas in another coordinate axis direction. Therefore, it is possible to configure the inspection image or inferred value image and at least one of the first and second projective images corresponding to the other coordinate axis direction, and to move and display them synchronously along one coordinate axis direction. Even during the moving display, it is easy to visually identify at which position of the inspected object an abnormality was detected when it was detected.

[0026] In a preferred embodiment of the present invention, (7) can be configured as follows: the display control unit moves the inspection image or the inferred value image in the display unit according to the conveying speed of the inspected object in the direction of the first coordinate axis, and displays at least one of the inspection image or the inferred value image and the first projection image and the second projection image in three adjacent display areas in the direction of the other coordinate axis of the inspection image or the inferred value image.

[0027] Invention Effects

[0028] According to the present invention, an article inspection device is provided that can easily and visually identify in the display screen of the inspection image which the detected object part or candidate is located in the transport direction where an abnormality is detected, and can also clearly indicate the judgment criteria. Attached Figure Description

[0029] Figure 1 This is a schematic structural diagram of an article inspection device according to one embodiment of the present invention.

[0030] Figure 2 This is an explanatory diagram of the display operation unit in an anomaly detection item inspection apparatus according to an embodiment of the present invention. It shows a first embodiment of the inspection image display in which an inspection image or a corresponding inference value image is arranged vertically on the display screen, and a bar graph-like projection image is obtained by projecting the maximum pixel density of each region in a plurality of y-axis line image regions in the inference value image onto the x-axis.

[0031] Figure 3 This is an explanatory diagram of the display operation unit in an article inspection apparatus according to an embodiment of the present invention. It shows a second embodiment of the inspection image display in which an inspection image or a corresponding inference value image is arranged left and right on the display screen, and a bar graph-like projection image is obtained by projecting the maximum pixel density of each region in a plurality of x-axis line image regions in the inference value image onto the y-axis.

[0032] Figure 4 This is an explanatory diagram of the display operation unit in an article inspection device according to an embodiment of the present invention. It shows a third embodiment of the inspection image display in which an inspection image or a corresponding inference value image is arranged in a display screen in a corresponding relationship with the top, bottom, left and right sides, and the maximum pixel density of each region in a plurality of x-axis and y-axis line image regions in the inference value image is projected onto the x-axis and y-axis respectively to obtain first and second projected images.

[0033] Figure 5 This is an explanatory diagram of the display operation unit in an article inspection device according to an embodiment of the present invention. It shows a fourth embodiment of the inspection image display in which an inspection image or a corresponding inference value image is arranged in three display areas on the upper and lower screens, and the maximum pixel density of each area in a plurality of x-axis and y-axis line image areas in the inference value image is projected onto the x-axis and y-axis respectively to obtain first and second projected images.

[0034] Figure 6 This is an explanatory diagram of the display operation unit in an object inspection device of an object detection system according to another embodiment of the present invention. It shows an embodiment in which an inspection image is arranged in a corresponding relationship with the top, bottom, left and right sides of the display screen, a rectangle is surrounded by a shape defect in the inspection image as the inspection object, and the maximum pixel density of each region of the line image area in the x-axis direction and y-axis direction of the inspection image in the rectangle is projected onto the x-axis and y-axis respectively and displayed together with the defect determination threshold, and the inspection image is displayed by moving them synchronously along the top, bottom or left and right directions.

[0035] Figure 7 This is an explanatory diagram of the display operation unit in the item inspection device of an object detection system according to another embodiment of the present invention, showing its use with... Figure 6 The inspection images shown are identical to those in one embodiment, but the defect determinations in the first and second projective images are adjusted using a threshold compared to... Figure 6 The situation shown is the display state after the higher credibility side. Detailed Implementation

[0036] Hereinafter, the methods for carrying out the present invention will be described with reference to the accompanying drawings.

[0037] (One implementation method)

[0038] Figures 1 to 5 An article inspection device according to one embodiment of the present invention is shown.

[0039] First, let's explain the structure.

[0040] likeFigure 1 As shown, the article inspection apparatus 1 of this embodiment includes a conveying unit 10 for conveying the article to be inspected, i.e., the article W; a camera unit 20 for capturing images of the article W during conveying; a control unit 30 for controlling the conveying unit 10 and the camera unit 20; and a display / operation unit 60 such as a touch panel. Furthermore, the article inspection apparatus 1, through the camera unit 20, simultaneously irradiates the article W conveyed by the conveying unit 10 with X-rays and detects image data corresponding to the distribution of transmitted X-rays, and inspects the quality status of the article W based on this image data. The quality status referred to here includes whether the article W meets the required quality or physical quantity standards as a product, such as the presence or absence of foreign matter, the presence or absence of missing parts, the conformity of the shape / size / accommodation of the contents, and the distribution of density / thickness / volume or mass.

[0041] The conveying unit 10 winds the annular conveyor belt 11 around the conveyor roller 12 on the drive side and the conveyor roller 13 on the driven side, and feeds items W, which are sequentially fed into the upward section 11a of the conveyor belt 11 from the upstream side, in the following direction. Figure 1 The conveyor, which transports the object to the right and moves it through the camera section of the camera unit 20 to the downstream side onto the conveyor 14, is supported by a frame (not shown).

[0042] Although detailed specifications are not illustrated, the imaging unit 20 includes, for example, an X-ray generator (X-ray source) that generates X-rays of a specified energy band that transmit through the article W conveyed by the conveyor unit 10, and an X-ray detector disposed directly below the upward section 11a of the conveyor belt 11. Furthermore, the imaging unit 20 is not limited to a component that irradiates the article W with X-rays to acquire an inspection image Dpx; for example, it could be a component that uses NIR (near-infrared) camera images of the appearance or transmission as inspection images, or it could be a component that uses color images obtained by photographing the appearance of the article with other light sources such as visible light as inspection images.

[0043] The X-ray generator of the camera unit 20 can generate X-rays with wavelengths and intensities corresponding to its tube current and tube voltage using a known X-ray tube, and irradiate the article W on the conveyor belt 11 with fan-shaped beams of X-rays directed toward the main observation direction orthogonal to the article transport direction of the conveyor unit 10 through the X-ray window of the peripheral equipment (details not shown).

[0044] Furthermore, although details are not shown, the X-ray detector of the camera unit 20 is, for example, an X-ray linear sensor camera, and is positioned at a predetermined position in the transport direction corresponding to the X-ray irradiation position from the X-ray generator. The X-ray linear sensor camera is arranged in an array along the width direction of the transport path of the transport unit 10, with detection elements, such as phosphors (scintillators) and photodiodes or charge-coupled devices, at predetermined intervals, and outputs a detection signal Lx at a predetermined resolution equivalent to the X-ray transmission amount.

[0045] That is, the imaging unit 20 can detect X-rays transmitted from the X-ray generator to the article W in each predetermined transmission area corresponding to the detection element, convert them into an electrical signal corresponding to the transmission amount of the X-rays, and output an X-ray detection signal for generating an X-ray transmission image with the transmission direction of the X-rays as the observation direction. Here, the X-rays irradiated from the X-ray generator or the X-rays detected by the X-ray detector are set to have a constant radiation quality (energy, wavelength) determined according to the quality of the article W, but so-called dual-energy or multi-energy X-ray images can also be generated by multiple X-rays with different radiation qualities.

[0046] The control unit 30 has the functions of a conveying control mechanism that controls the conveying speed or conveying interval of the article W based on the conveyor belt 11 in the conveying unit 10, and an inspection control mechanism that controls the X-ray irradiation intensity or irradiation period in the camera unit 20, or controls the X-ray detection cycle and detection period of each article W in the X-ray linear sensor of the X-ray detector corresponding to the conveying speed of the article W.

[0047] The control unit 30 includes an inspection image storage unit 31, an image processing unit 32, and a learning model 33, which are the main mechanisms for performing the functions of the inspection control mechanism, and also includes a display control unit 50 for display control of the display / operation unit 60.

[0048] The inspection image storage unit 31 sequentially reads X-ray detection signals from the X-ray detector of the camera unit 20, temporarily stores image data representing the X-ray transmittance distribution of each item W in the memory, and outputs the image data as the inspection image Dpx.

[0049] The image processing unit 32 sequentially reads the image data of the inspection image Dpx output from the inspection image storage unit 31, and performs image analysis processing to extract global or local features of the image (e.g., extraction of feature quantities of local regions based on pixel values ​​or brightness gradients, extraction of frequency feature quantities of the image as a whole such as spatial frequency spectrum) through prescribed preprocessing or filtering. It then performs a first inspection image processing step, based on the result of this image processing and using a prescribed image processing algorithm, to determine whether the quality state of the item W is normal. The prescribed filtering processing mentioned here is used to detect or emphasize image features (e.g., edges or spots) that indicate a deviation from the normal state in quality state, i.e., a degree different from the normal level, through the aforementioned prescribed image processing algorithm.

[0050] In addition to performing the first inspection image processing function described above, the image processing unit 32 also performs a second inspection image processing function, which, in cooperation with the learning model 33, performs deep learning-based classification or anomaly detection based on the inspection image of the item W acquired by the inspection image storage unit 31 or the aforementioned filtered inspection image, thereby determining with confidence whether the quality status of the item W is normal. Here, classification is a process that extracts features from the input image and learns decision boundaries, for example, to determine the type of the inspected item. Anomaly detection is a process that detects abnormal parts in the input image, such as partial defects in the contents of the inspected item or unevenness outside the normal range, as abnormalities.

[0051] The learning model 33 is a multi-layer neural network used to enable the image processing unit 32 to perform classification or anomaly detection functions based on deep learning, according to the aforementioned inspection image or the inspection image after specified filtering.

[0052] In the learning phase, the learning model 33 uses only image data of qualified product images without any abnormalities such as foreign objects as the image dataset for learning to learn the features of qualified product images. A specified number of qualified product images (e.g., about 1000 images) are input into the learning model 33, and the learning model adjusts the parameters using parameters such as the weights between layers of the neural network, for example, the weighted weights in the hidden layer (intermediate layer) relative to any i-th neuron of the input layer and the weighted weights in the output layer of any k-th neuron relative to any j-th neuron of the hidden layer (which may include a threshold).

[0053] The dataset of qualified product images used in the learning model 33 is, for example, a dataset in which OK labels are automatically attached to the sample images of qualified products as annotation information for anomaly detection. However, it can also be a dataset in which qualified products with shapes or configurations that are closest to the categories of anomalous products, such as qualified products with defects that are difficult to observe with the naked eye or with bumps that are close to the normal limit, are assigned a score that indicates a greater degree of abnormality.

[0054] The learning model 33 uses image data of each qualified product image for learning, and adjusts parameters such as inter-layer weights in a manner such that the output value of the neural network is distributed in the normal attribute region of the main feature quantity distribution of each qualified product image in the feature space based on the global or local features of the aforementioned image.

[0055] When the learning model 33, which has completed the parameter adjustment, inputs the image data of the inspection image Dpx of the qualified product from the inspection image storage unit 31 during the inference phase, it determines the centroid or distribution pattern of the feature quantity distribution of the inspection image and its various judgment pixel regions in the feature space based on the global or local features of the aforementioned image within the normal attribute region of the distribution of the qualified product image used for learning, based on the output value of the neural network.

[0056] The image processing unit 32 performs a second inspection image processing function in cooperation with the learning model 33 to perform classification or anomaly detection based on the aforementioned deep learning. It has an AI processing unit 41 that performs the second inspection image processing function and a determination unit 45 that comprehensively determines whether the quality status of the item W is normal or makes a determination based on the results of the first inspection image processing and the results of the second inspection image processing in the AI ​​processing unit 41, or makes a determination based on the effective inspection image processing results of the variety.

[0057] The AI ​​processing unit 41 is configured to perform classification or anomaly detection functions based on the aforementioned inspection image or the inspection image after specified filtering and using the learning model 33, and includes a projective image generation unit 42 and an inference value image generation unit 43.

[0058] When the image data of the inspection image Dpx is input from the inspection image storage unit 31, the inference value image generation unit 43 compares the distribution of the output value of the learning model 33 relative to each image data with the centroid or distribution pattern of the distribution of multiple qualified product images that have been learned in the same feature space for the corresponding judgment pixel region for each specified number of judgment pixel regions, and outputs an inference value as the confidence level, which is the probability that the distribution pattern of the feature quantity of the input image is not distributed in the normal attribute region.

[0059] More specifically, the inference value image generation unit 43 is configured as an image generation neural network corresponding to the multi-layer neural network of the learning model 33, for example, using a convolutional autoencoder with feature parameters or weighting coefficients learned by the learning model 33. This autoencoder has the property that when image data of the inspection image Dpx is input from the inspection image storage unit 31, the error in reconstructing the defective or abnormal parts of the image increases compared to the case of inputting image data of a qualified product, especially when the image data is not of a qualified product. Therefore, the reconstruction error of the qualified product image in the inference value image generation unit 43 can be considered a value representing the degree of abnormality in the distribution pattern of the feature quantities of the input image.

[0060] Therefore, the inference value image generation unit 43 outputs an inference value as confidence for each pixel of the reconstructed image based on the autoencoder of the image data of the inspection image Dpx from the inspection image storage unit 31. This inference value represents the probability that the distribution of feature quantities in the aforementioned feature space deviates from the normal attribute region, for each of the multiple predetermined number of decision pixel regions constituting the inspection image Dpx.

[0061] The inference value image generation unit 43 is further configured to generate image data of a two-dimensional density image with the density of the image as the output value of the output value of the confidence level corresponding to the multiple decision pixel regions of the inspection image Dpx, as the image data of the inference value image Dcf. Here, the density of each pixel (pixel density) can be directly obtained by using the output value of the confidence level, but when the range of values ​​that the output value of the confidence level can take is much different from the range of values ​​that the specified value of the pixel density can take, a conversion formula can be used to obtain the value of the density corresponding to the output value of the confidence level. For example, when the range of values ​​that the output value of the confidence level can take, Dw, is 0 to 100 (Dw=100) and the range of values ​​that the specified value of the pixel density, Cw, is 0 to 255 (Cw=255), the conversion formula [1] can be used.

[0062] Pixel density = confidence level × Cw / Dw……[1]

[0063] Furthermore, when the image being examined is a color image, the saturation of the two-dimensional saturation image can be pixel values ​​that only produce saturation variations for a specific color component (R, G, B). Moreover, the learning model 33, while also functioning as an object detection model, can learn whether a rectangle in the image contains an object or background. If it contains an object, it learns in a way that reduces the error of the ground truth label for the object type within the rectangle. In this case, the inference value image generation unit 43 can have a two-stage network structure: a stage for acquiring a feature map of the input image and a stage for generating multiple rectangles on that feature map and proposing candidate rectangle regions for calculating the classification inference within each rectangle and its inference error.

[0064] like Figure 2As shown, when the projection image generation unit 42 inputs the image data of the two-dimensional density image, namely the image data of the inference value image Dcf, from the inference value image generation unit 43, it projects the maximum concentration Cpv corresponding to the confidence level of each line image region xpi in the direction of the other coordinate axis y onto the coordinate axis x for multiple adjacent line image regions xpi in the direction of the coordinate axis x in the inference value image Dcf. This generates a bar graph-shaped projection image Dpr in the corresponding line image regions xpi' with a length corresponding to the maximum concentration Cpv of the confidence level, and outputs it to the display control unit 50.

[0065] The display control unit 50, based on the conveying speed Vc of the item W towards one coordinate axis x, causes the inspection image Dpx or the inference value image Dcf to be displayed along one side of the inspection image display area 63 of the display / operation unit 60. Figure 2 The x-axis direction is shifted, and the inference value image Dcf and the bar chart-like projective image Dpr are arranged so that their respective coordinate axes, i.e., the x-axis, are parallel to each other and aligned along the other coordinate axis, i.e., the y-axis, of the two coordinate axes x and y of the inference value image Dcf.

[0066] That is, the display control unit 50 displays the checked image Dpx or the inferred value image Dcf at the same position on one side of the y-axis direction. Figure 2 The upper side of the image is used as the upper viewing image, and the other side along the y-axis is... Figure 2 The lower side of the image shows the projection image Dpr as an observation image that allows observation of the distribution of the maximum value Cpv (referred to as confidence) among multiple decision pixel regions of each line image region xpi from the front of the inspection image Dpx or the inference value image Dcf.

[0067] Furthermore, the projective image generation unit 42 is configured to generate at least one of the following projective images: for multiple first-line image regions xpi adjacent in the x-direction and y-direction in the inspection image Dpx or the inference value image Dcf, the maximum confidence value Cpv (equivalent to the concentration of the inferred confidence value) corresponding to the pixel with the maximum concentration in each first-line image region xpi is generated; Figures 2 to 5 The first projected image Dpr1, abbreviated as confidence level Cpv, is obtained by projecting the confidence level Cpv onto a coordinate axis, i.e., the x-axis. This results in a first projected image Dpr1 with a length Li corresponding to the maximum confidence level Cpv in the corresponding first-line image region xpi´. For multiple adjacent second-line image regions xqk in the y-direction of the inspection image Dpx or the inference image Dcf, the maximum confidence level Cpv of each second-line image region xqk is projected onto another coordinate axis, i.e., the y-axis. This results in a length Lk corresponding to the maximum confidence level Cpv in the corresponding second-line image region xqk. Figures 3 to 5The second projective image Dpr2 (in the case of Lk=Lj) is shown.

[0068] like Figures 2 to 5 In the first to fourth embodiments of the inspection image display, the display control unit 50 displays, for example, an inspection image Dpx that is easy to identify the item W on the display / operation unit 60, and aligns the position xi in the x-axis direction and / or the position yi in the y-axis direction of the first line image area xpi and the second line image area xqk of the inference value image Dcf with the inspection image Dpx or the inference value image Dcf, respectively, at least one of the first projective image Dpr1 and the second projective image Dpr2 that can be obtained from the inference value image Dcf at the same position as the inspection image Dpx displayed therein, based on the movement of the item W through the camera interval (including temporary stops for a predetermined time).

[0069] That is, in Figure 2 In the first embodiment of the inspection image display shown, in the inspection image display area 63 of the display / operation unit 60, each first line image region xpi´ of the first projective image Dpr1 is displayed relative to the inspection image Dpx (or the inferred value image Dcf: about Figures 2 to 5 (The same applies below) The corresponding first-line image regions xpi are always displayed in a way that moves synchronously with each other on the same straight line.

[0070] Furthermore, in Figure 3 In the second embodiment of the inspection image display shown, in the inspection image display area 63 of the display / operation unit 60, the second line image regions xqk' of the second projective image Dpr2 are synchronously moved along the x-direction (in this case, the y-direction) relative to each other in such a way that the corresponding second line image regions xqk of the inspection image Dpx are always displayed on the same straight line. Furthermore, the number m and width (y-direction width) of the second line image regions xqk' can be the same as or different from the number n and width (x-direction width) of the first line image regions xpi', and may also change depending on the image size and display size of the inspection image Dpx. Figure 3 In subsequent embodiments of the inspection image display, the width of each line image region is shown as a ratio. Figure 2 The example diagram is large.

[0071] Moreover, in Figure 4In the third embodiment of the inspection image display shown, in the inspection image display area 63 of the display / operation unit 60, each first line image region xpi' of the first projective image Dpr1 is displayed synchronously along the x-direction so that the corresponding first line image regions xpi of the inspection image Dpx are always displayed on the same straight line. Each second line image region xqk' of the second projective image Dpr2 is displayed synchronously along the x-direction (or y-direction) so that the corresponding second line image regions xqk of the inference value image Dcf are always displayed on the same straight line.

[0072] Or, such as Figure 5 The fourth embodiment of the inspection image display illustrates that, in the inspection image display area 63 of the display / operation unit 60, within display area A1 of the three adjacent display areas A1, A2, and A3 in the y-axis direction, the inspection image Dpx and its coordinate axis directions x and y are described. Figure 1 The display may be moved along with or in conjunction with a more appropriate grid display. Alternatively, it may be as follows: within display area A2 directly below display area A1 of the three display areas A1, A2, and A3, each first-line image region xpi' of the first projective image Dpr1 is moved synchronously along the x-direction relative to the corresponding first-line image region xpi of the inspection image Dpx, ensuring they are always displayed on the same straight line. Similarly, each second-line image region xqk' of the second projective image Dpr2, which has a different viewing direction relative to the first projective image Dpr1, is moved synchronously along the x-direction relative to each first-line image region xpi' of the first projective image Dpr1, ensuring they are always displayed on the same straight line.

[0073] That is, the display control unit 50 can move the inspection image Dpx or the inference value image Dcf to the left or right side in the inspection image display area 63 of the display / operation unit 60 according to the conveying speed of the item W in the x-axis direction, and display at least one of the inspection image Dpx or the inference value image Dcf and the first projective image Dpr1 and the second projective image Dpr2 in three adjacent display areas A1, A2, and A3 in the y-axis direction of the inspection image Dpx or the inference value image Dcf.

[0074] In the display / operation unit 60, the inspection image Dxp or the inference value image Dcf is displayed in the wide inspection image display area 63 below the inspection status display area 61 and the common information display area 62. In the right-hand area, namely the inspection information display areas 63a and 63b, the variety number corresponding to the current set variety, the inspection judgment result of the latest inspected item W displaying the main inspection information, and the total inspection result relative to the set variety are displayed.

[0075] Furthermore, in the operation area 64 below the inspection image display area 63, various operation buttons constituting the operation input function section of the display / operation unit 60 are arranged from left to right, including a menu button 64a, a display switching button 64b, and a setting / adjustment button 64c. Moreover, on the right side of the operation area 64 are a stop button 71 (STOP in the figure) for requesting to stop the inspection and a start button 72 (START in the figure) for requesting to start the inspection operation.

[0076] exist Figure 2 and Figure 3 The projected image Dpr shown is Figure 4 and Figure 5 In the first projective image Dpr1 and the second projective image Dpr2 shown, the determination threshold Thr, which is used to determine whether the quality status of the item W is normal in the determination unit 45 based on the result of the first inspection image processing in the image processing unit 32 and the result of the second inspection image processing in the AI ​​processing unit 41, is displayed parallel to the projective axis, i.e., the x-axis or the y-axis.

[0077] The determination threshold Thr is generated by the projected image generation unit 42 and is automatically preset in advance in the following manner: for example, when the image is captured by the camera unit 20 with an additional... Figure 2 When the image data of the multiple foreign object samples Ct1, Ct2, Ct3, and Ct4 with different sphere diameters shown are input from the inspection image storage unit 31 to the image processing unit 32, the foreign object sample Ct1 with the smallest diameter among these multiple foreign object samples Ct1, Ct2, Ct3, and Ct4 is regarded as the sample with the smallest foreign object or other relatively mild abnormal part that is detected with the required detection accuracy for the article W. The maximum value of the confidence (concentration) Cpvi in ​​the specific first line image region xpi corresponding to the foreign object sample Ct1 is judged as abnormal according to the judgment threshold Thr, or corresponds to the result of the first inspection image processing.

[0078] Furthermore, by referring to the display content of the inspection image Dxp or the inference value image Dcf and the first projective image Dpr1 and / or the second projective image Dpr2 displayed in the inspection image display area 63 of the display / operation unit 60, the setting / adjustment button 64c and the like can be used to fine-tune the judgment threshold Thr in the direction of increasing or decreasing confidence Cpv.

[0079] In the projective image Dpr, the first projective image Dpr1, and the second projective image Dpr2 displayed in the inspection image display area 63 of the display / operation unit 60, the display range in the direction of increasing or decreasing confidence (density) of each can be appropriately set, for example, according to the usage range of display image density in the display / operation unit 60 (equivalent to the brightness range from minimum brightness to maximum brightness), the background image density in the inspection image Dxp or the inference value image Dcf, and the display image density range of the item W, or it can be appropriately set according to the differences in display methods such as the first to fourth embodiments of inspection image display.

[0080] Next, the function will be explained.

[0081] In this embodiment configured as described above, firstly, during the learning phase of the learning model 33, the features of the qualified product images are learned using only image data of qualified product images without any abnormalities such as foreign objects as the image dataset for learning, and the parameters such as the weights between layers of the neural network constituting the learning model 33 are used for learning.

[0082] If the parameter adjustment is completed, during the inference stage of the learned model 33, the image data of the inspection image Dpx of the qualified product is input from the inspection image storage unit 31.

[0083] For example, firstly, in order to set the inspection conditions for the type of object to be inspected, if the camera unit 20 captures a virtual defective product with multiple foreign matter samples Ct1, Ct2, Ct3, and Ct4 of different spherical diameters attached, and the image data of the inspection image Dpx is input from the inspection image storage unit 31 to the image processing unit 32, then the judgment threshold Thr is automatically set in such a way that the maximum value of the confidence (concentration) Cpvi in ​​the specific first-line image region xpi in the inference value image Dcf corresponding to the foreign matter sample Ct1 with the smallest diameter among these multiple foreign matter samples Ct1, Ct2, Ct3, and Ct4 exceeds the judgment threshold Thr and is judged as abnormal.

[0084] Alternatively, referring to the display content of the inspection image Dxp or inference value image Dcf and the first projection image Dpr1 and / or the second projection image Dpr2 displayed in the inspection image display area 63 of the display / operation unit 60, the operator operates the setting / adjustment button 64c, etc., to fine-tune the judgment threshold Thr in the direction of increasing or decreasing confidence Cpv.

[0085] Next, the inspected object, i.e., item W, is photographed at a specified conveying interval or conveying distance unit. The image data of the multiple items W that are put in are used as the image data of the inspection image Dpx and are sequentially input from the inspection image storage unit 31 to the image processing unit 32.

[0086] At this time, in the inference value image generation unit 43 of the image processing unit 32, based on the reconstruction error of each pixel of the reconstructed image of the image data of the inspection image Dpx from the inspection image storage unit 31, an inference value of the degree of abnormality is calculated as a confidence level for each of the multiple predetermined number of judgment pixel regions constituting the inspection image Dpx, for example, each pixel in each judgment pixel region. Moreover, based on the output value of each confidence level corresponding to the multiple judgment pixel regions of the inspection image Dpx, image data of a two-dimensional density image from the inference value image generation unit 43, with the density of the image respectively represented by the output value of these confidence levels, is generated as the image data of the inference value image Dcf.

[0087] Furthermore, when outputting the image data of the two-dimensional density image, i.e., the image data of the inference value image Dcf, from the inference value image generation unit 43, the projective image generation unit 42 uses the image data of the inference value image Dcf, for example, as... Figure 2 As shown, for each line image region xpi in the inference value image Dcf, the maximum value Cpv of the concentration corresponding to the confidence level of the line image region xpi is projected onto a coordinate axis x, and a bar-shaped projected image Dpr is generated in the corresponding line image region xpi´ to display the maximum value Cpv as the confidence level Cpv, and then output to the display control unit 50.

[0088] Furthermore, when outputting the image data of the inference value image Dcf from the inference value image generation unit 43, the determination unit 45 sets a determination threshold Thr for determining whether the quality status of the item W is normal based on the result of the first inspection image processing in the image processing unit 32 and the result of the second inspection image processing in the AI ​​processing unit 41.

[0089] Furthermore, at this time, the display control unit 50, based on the conveying speed Vc of the item W in the direction of a coordinate axis x, causes the inspection image Dpx or the inference value image Dcf to be displayed along the inspection image display area 63 of the display / operation unit 60. Figure 2 The image is moved along the x-axis, and the inference value image Dcf and the bar graph-like projective image Dpr, which is attached with a line corresponding to the judgment threshold Thr, are moved synchronously along the y-axis in such a way that the x-axis are parallel to each other and have the same display range in the x-direction.

[0090] In this embodiment, the inferred value related to the quality state of item W is calculated as the confidence level for each unit area with a specified number of pixels in the inspection image Dpx, and a two-dimensional density image with the inferred value as the density is generated as the inferred value image Dcf. Furthermore, for multiple line image regions xpi adjacent to the y-axis direction on one coordinate axis of the inferred value image Dcf, the maximum pixel value Cpv of each line image region xpi is correlated with the transport direction of item W, etc., and is sequentially projected onto any coordinate axis, and the projected image is displayed.

[0091] Therefore, it is easy to instantly visually identify at which position in the transport direction of item W or in the line scan direction an anomaly or foreign object is detected. As a result, in the inspection image display area 63, the location of the detected object or candidate for anomaly detected in the transport direction, i.e., the x-direction, can be easily visually identified, and the judgment criteria can also be explicitly indicated as the judgment threshold Thr.

[0092] Furthermore, in this embodiment, it is possible to easily visually identify from the bar graph-like projective image Dpr that displays a threshold that an anomaly is detected in which item W is being transported, at which position in the transport direction, or that no anomaly is detected at any position.

[0093] Furthermore, in this embodiment, when the inspection image Dpx or the inferred value image Dcf in the display / operation unit 60 moves to the left or right side according to the conveying speed Vc of the item W in the x-axis direction, the inspection image Dpx or the inferred value image Dcf and its corresponding bar graph-shaped projective image Dpr are displayed on the upper and lower sides (one side and the other side) in the y-axis direction. Therefore, for each item W, it is easy to visually identify whether there is an abnormality in the quality status from the bar graph-shaped projective image Dpr with the threshold display. Moreover, as the inspection image Dpx or the inferred value image Dcf of the item W moves and is displayed, the corresponding projective image Dpr moves and is displayed synchronously along the same direction, so it is easier and more accurate to visually identify the location of the abnormal part of the quality status.

[0094] Furthermore, in this embodiment, the projective image generation unit 42 generates at least one of the following projective images: a first projective image Dpr1, which projects the maximum value Cpv of the inferred value concentration of each of the multiple adjacent first line image regions xpi in the x-axis direction onto the x-axis; and a second projective image Dpr2, which projects the maximum value Cpv of the inferred value concentration of each of the multiple adjacent second line image regions xqk in the x-axis direction onto the y-axis. Therefore, the first projective image Dpr1 and / or the second projective image Dpr2 can be displayed in a corresponding relationship with the inspection image Dpx or the inferred value image Dcf on either side of the vertical direction and / or either side of the horizontal direction. For each item W, it is possible to easily visually identify whether there is an abnormality in the quality status or the location of the abnormality from the bar-shaped projective image Dpr with a threshold display.

[0095] Furthermore, in this embodiment, the display control unit 50 checks at least one of the image Dpx or the inferred value image Dcf and the first projective image Dpr1 and the second projective image Dpr2, aligns the positions of each first line image region xpi and its corresponding first line image region xpi´ and each second line image region xqk and its corresponding second line image region xqk´ with the corresponding x-axis or y-axis direction or each coordinate axis direction, and then displays it on the display / operation unit 60. Therefore, it is easier to visually identify, from any timely display switching of the first projective image Dpr1 and the second projective image Dpr2 or bidirectional simultaneous projective display information, which item W in the transported items W, at what position, or no abnormality was detected at any position.

[0096] In this embodiment, when the inspection image Dpx or the inference value image Dcf is moved to the left or right side in the display / operation unit 60 according to the conveying speed of the article W in the x-axis direction, at least one of the first and second projective images Dpr1 and Dpr2 can be configured corresponding to the y-axis direction relative to the inspection image Dpx or the inference value image Dcf, and can be moved and displayed synchronously along the x-axis direction. Therefore, even during the moving display, when an anomaly is detected, it is easy to visually identify at which position of the article W the anomaly was detected, and the display area of ​​the moving image can be sufficiently ensured in the direction of movement.

[0097] Thus, according to this embodiment, an item inspection device 1 can be provided that can easily and visually identify in the display screen of the inspection image Dpx which position in the transport direction is where an abnormality or other defects are detected, or its candidate, and can also clearly indicate the judgment criteria.

[0098] (Other implementation methods)

[0099] Figure 6 and Figure 7 An embodiment of an inspection image display in an article inspection apparatus according to another embodiment of the present invention is illustrated.

[0100] Furthermore, the article inspection device of this embodiment has a device structure that is substantially the same as that of the article inspection device 1 of the aforementioned embodiment. Therefore, regarding the structure similar to that of the article inspection device 1 of the previous embodiment, the same... Figure 1 The same reference numerals are used in the embodiment shown, and to avoid repetition of the description, the differences from the embodiment will be described below.

[0101] In this embodiment, the learning model 33 is configured to also function as an object detection model. The control unit 30 uses the functions of the AI ​​processing unit 41, the projective image generation unit 42, and the inference value image generation unit 43 of the image processing unit 32, as well as the function of the determination unit 45, to perform the object detection function.

[0102] That is, the inference value image generation unit 43 calculates an inference value as the degree of abnormality for each pixel of the reconstructed image based on the autoencoder of the image data of the inspection image Dpx from the inspection image storage unit 31, for a plurality of predetermined number of determination pixel regions constituting the inspection image Dpx, as a confidence level. However, in this embodiment, the inference value is further calculated based on the relationship between this confidence level and the determination threshold Thr set by the determination unit 45. Figure 6 , Figure 7 The rectangles shown here, Bx1, Bx2, and Bx3, are displayed on the display / operation unit 60.

[0103] To perform this object detection, during the learning phase of the learning model 33, for example, for multiple decision pixel regions constituting a specified number of pixels in the inspection image Dpx, a search is conducted to determine whether to set them as candidates for rectangles (bounding boxes; here, containing information on position, type, and confidence) and to determine whether the rectangle contains an object or background. In the case of an object, for a playback image of the object within the rectangle, the error between the object's category and the ground truth label is calculated to determine the candidate rectangle region, and learning is performed in a way that reduces this error. Furthermore, the shape of the bounding box surrounding the object's extent is a typical rectangle, but it can be any shape including rectangles.

[0104] During the inference phase, the inference value image generation unit 43 obtains a feature map of the input image from the inspection image storage unit 31. Multiple rectangles are generated on this feature map, and inferences are performed to classify whether the object within each rectangle is an object or background, along with calculations of the object's inference error (e.g., the aforementioned reconstruction error). Furthermore, the image processing unit 32 and the display control unit 50 first display multiple rectangular candidate regions that exceed a pre-set, sufficiently small first threshold Thr1 (where Thr1 corresponds to a deliberate pixel density value exceeding 0), for example... Figure 6 After obtaining a sufficient number of candidate rectangles, such as the three rectangular candidate regions Bx1, Bx2, and Bx3 shown, for higher precision inspection, such as manually adjusting a second threshold Thr2 larger than the first threshold Thr1 by skilled personnel who are adept at visually identifying and inspecting malformations in the image Dpx or inference value image Dcf, for example, to perform higher precision inspection, the thresholds are adjusted to be more precise. Figure 7 As shown, when there is a shape defect to be detected, a rectangular candidate area Bx2 surrounding the shape defect can be accurately displayed. Of course, as a structure that automatically sets the second threshold Thr2 according to the required detection accuracy, it can be configured so that the user can effectively confirm whether the threshold setting is appropriate based on changes in the content displayed on the screen in the display / operation unit 60 as changes in the object detection area.

[0105] In this embodiment, the inferred value related to the quality state of item W is calculated as the confidence level for each unit area with a specified number of pixels in the inspection image Dpx, and a two-dimensional density image with the inferred value as the density is generated as the inferred value image Dcf. Furthermore, for multiple line image regions xpi adjacent to the y-axis direction on one coordinate axis (x-axis) of the inferred value image Dcf, the maximum pixel value Cpv of each line image region xpi is projected onto one and another coordinate axes (x-axis and y-axis) respectively, establishing a corresponding association with the transport direction of item W, and is displayed as the first projected image Dpr1 and the second projected image Dpr2.

[0106] Therefore, it is easy to instantly visually identify which position in the transport direction of item W or which position in the line scan direction an anomaly or foreign object is detected. As a result, in the inspection image display area 63, it is easy to visually identify which position in the transport direction, i.e., the x-direction, the detected object part or its candidate can be displayed, and its judgment criteria can also be clearly indicated as judgment thresholds Thr1 and Thr2.

[0107] Furthermore, in this embodiment, it is possible to easily visually identify from the first and second projective images Dpr1 and Dpr2, which have threshold displays, which item W is being transported, at which position in the transport direction an anomaly is detected, or no anomaly is detected at any position.

[0108] As explained above, the present invention provides an article inspection device that can easily and visually identify in the display screen of the inspection image where an anomaly or other abnormality is detected in the transport direction, and can also clearly indicate the judgment criteria. The present invention is effective among all article inspection devices that use inspection images obtained by photographing the inspected object and a learned model to inspect the quality status of the article.

[0109] Symbol Explanation

[0110] 1-Item inspection device; 10-Conveying unit; 11-Conveyor belt; 11a-Upward section; 12-Conveyor roller; 13-Conveyor roller; 14-Conveyor (subsequent stage conveyor); 20-Camera unit; 30-Control unit; 31-Inspection image storage unit; 32-Image processing unit; 33-Learning model; 41-AI processing unit; 42-Projective image generation unit; 43-Inference value image generation unit; 45-Decision unit; 50-Display control unit; 60-Display / operation unit; 61-Inspection status display area. 62 - Common information display area, 63 - Inspection image display area, 63a, 63b - Inspection information display area, 64 - Operation area, 64c - Setting / adjustment button, 71 - Stop button, 72 - Start button, A1, A2, A3 - Display area, Bx1, Bx2, Bx3 - Rectangle (rectangle candidate area), Ct1, Ct2, Ct3, Ct4 - Foreign object sample, Cpv - Confidence (equivalent to the maximum pixel density of the confidence level of each line image area, inferred value), C pvi - Confidence (equivalent to the maximum pixel density of the confidence level of each first-line image region, inferred value), Cpvk - Confidence (equivalent to the maximum pixel density of the confidence level of each second-line image region, inferred value), Dcf - Inferred value image, Dpr - Projective image (first or second projective image), Dpr1 - First projective image, Dpr2 - Second projective image, Dpx - Inspection image (X-ray transmission image, detection image), Lx - Detection signal (equivalent to the detection signal of X-ray transmission), Thr, Thr1, Thr2 - Judgment threshold, xpi - First-line image region (the line image region along another coordinate axis, i.e., the y-axis), xpi´ - Corresponding line image region (the projective image region corresponding to the first-line image region), xqk - Second-line image region (the line image region along one coordinate axis, i.e., the x-axis), xqk´ - Corresponding line image region (the projective image region corresponding to the second-line image region), Vc - Conveying speed, W - Item (the object being inspected).

Claims

1. An article inspection apparatus, characterized by, Possessing: an image storage section (31) that stores an inspection image (Dpx) of an inspected object (W) obtained by photographing the inspected object (W) being conveyed; a determination section (45) that, by using a learning model (33) that is learned in advance using an image data set in which the imaging conditions are the same as the inspected object, calculates an inference value related to the quality state of the inspected object for each unit region of a predetermined number of pixels of the inspection image stored in the image storage section, and determines the quality state of the inspected object by comparing the inference value with a threshold value (Thr) set in advance; an inference value image generation section (43) that generates an inference value image (Dcf) that is a two-dimensional density image in which the inference value corresponding to the inspection image is set as the density; a projection image generation section (42) that generates a bar graph-shaped projection image (Dpr) in which the maximum value (Cpv) of the density of a plurality of line image regions (xpi) in the other coordinate axis (y) direction that are adjacent in the one coordinate axis (x) direction in the inference value image is respectively projected onto the corresponding line image region (xpi') in the one coordinate axis with a length corresponding to the maximum value of the density; and a display control section (50) that adds a line corresponding to the threshold value to the projection image and displays it on a display section (60).

2. The article inspection apparatus according to claim 1, characterized in that the display control section displays the inspection image or the inference value image and the bar graph-shaped projection image on the display section after aligning the position (xi) in the one coordinate axis direction of the plurality of line image regions and the corresponding line image region in accordance with the conveyance speed (Vc) of the inspected object in the one coordinate axis direction.

3. The article inspection apparatus according to claim 1 or 2, characterized in that the display control section moves the inspection image or the inference value image in the display section in accordance with the conveyance speed of the inspected object in the one coordinate axis direction, and displays the inspection image or the inference value image and the bar graph-shaped projection image on both sides in the other coordinate axis direction of the inspection image or the inference value image.

4. The article inspection apparatus according to claim 1 or 2, characterized in that the projection image generation section generates at least one of: a first projection image (Dpr1) that is a bar graph-shaped image obtained by respectively projecting the maximum value of the density of a plurality of first line image regions (xpi) in the other coordinate axis direction that are adjacent in the one coordinate axis (x) direction in the inference value image onto the corresponding first line image region in the one coordinate axis with a length corresponding to the maximum value of the density; and a second projection image (Dpr2) that is a bar graph-shaped image obtained by respectively projecting the maximum value of the density of a plurality of second line image regions (ypi) in the one coordinate axis direction that are adjacent in the other coordinate axis (y) direction in the inference value image onto the corresponding second line image region in the other coordinate axis with a length corresponding to the maximum value of the density. A bar chart-like 2nd projection image (Dpr2) in which maximum values of the concentration of a plurality of 2nd line image regions (xqk) of the one coordinate axis (x) direction adjacent in the other coordinate axis direction in the inference value image are respectively projected onto the corresponding 2nd line image region on the other coordinate axis as a length corresponding to the maximum value of the concentration.

5. The article inspection apparatus according to claim 4, wherein The display control section displays the inspection image or the inference value image (Dpx or Dcf) and at least one of the 1st projection image and the 2nd projection image (Dprl and / or Dpr2) on the display section in alignment with positions (xi and / or yi) in the corresponding any one or each coordinate axis (x and / or y) direction with respect to the 1st line image region and the corresponding 1st line image region and the 2nd line image region and the corresponding 2nd line image region.

6. The article inspection apparatus according to claim 5, wherein The display control section moves the inspection image or the inference value image in the display section in accordance with a conveyance speed of the inspected article in the one coordinate axis direction, and displays the inspection image or the inference value image and at least one of the 1st projection image and the 2nd projection image in three display regions (Al, A2, A3) adjacent in the other coordinate axis direction of the inspection image or the inference value image.

7. The article inspection apparatus according to claim 4, wherein The display control section moves the inspection image or the inference value image in the display section in accordance with a conveyance speed of the inspected article in the one coordinate axis direction, and displays the inspection image or the inference value image and at least one of the 1st projection image and the 2nd projection image in three display regions (Al, A2, A3) adjacent in the other coordinate axis direction of the inspection image or the inference value image.

Images