Method and apparatus for inspecting hot glass containers for defects

Abstract

The present invention relates to a method for inspecting glass containers (2) at high temperatures, the method comprising: for each container, obtaining at least one transmission image (It) of the container illuminated by a source (14) of light passing through the container, and at least one infrared radiation image (Ir) of the container, analyzing the at least one transmission image and the at least one infrared radiation image, checking a match between at least a portion of the transmission image and at least a portion of the infrared radiation image, and classifying the container from the at least one transmission image and the at least one infrared radiation image matched to each other, and identifying at least one defect type for the container.

Description

[Technical field]

[0001] The present invention relates to the technical field of online inspection of transparent or translucent containers, such as glass bottles, pots or vials, for the purpose of their quality control, to detect and identify defects that may affect these containers.

[0002] The object of the present invention is to find a particularly advantageous application for analyzing the physical properties of containers in order to identify non-conforming physical properties corresponding to defects such as, for example, surface defects such as wrinkles or cracks, internal defects in the material such as cracks, inclusions or air bubbles, and also dimensional defects of the container such as glass distribution. [Background technology]

[0003] For the production of glass containers, a manufacturing method that includes melting glass and then transporting it to a forming unit is known to be carried out by a manufacturing facility that includes a melting furnace and a forehearth to transport the molten glass to a forming machine of the type generally called IS machine. The IS forming machine includes a distributor that forms glass droplets called parisons and distributes them to the forming section through a conduit called "discharge conduit". At the output of the forming section, the containers have a high temperature, typically between 300°C and 600°C. The containers just formed in the forming machine are arranged in succession on an output conveyor to form a row of containers. The containers are carried in a single file by the conveyor and transported successively to the various processing stations. In particular, the formed containers are transported to a melting furnace where they are heated and then cooled in a controlled manner that eliminates the thermal constraints caused by the forming process. Other forming methods for glass containers are known for tabletop glassware, insulators, syringes, and light bulbs. For example, forming machines such as rotary and sequential presses, rather than parallel aligned sections as in IS machines. There are also machines that process the preforms into tubes, especially monocalcium glass tubes, to manufacture syringes and light bulbs specifically for pharmaceutical use.

[0004] It is known to systematically inspect all containers discharged from the melting furnace using various inspection devices, in particular wall inspection systems as described in patents EP 2082216, EP 2145175 or EP 2856122. Thus, a transparent wall inspection technique is known, in which a light source is arranged on one side of the conveyor and at least one camera (typically 2 to 6, 12 or 24) is arranged on the other side to obtain at least one image formed by light transmitted through the wall of the container. Patent EP 1 109 008 describes a method for analyzing images of containers for cold inspection. In a segmentation step, features in the image and the area around the features are detected. Identification parameters of the area are calculated and combined with fuzzy logic methods to determine the most probable type of feature in the list of features corresponding to a possible defect. The suitability of the area is then determined by applying different criteria depending on the type of feature stored. For example, wrinkle type defects are rejected for a certain surface, while inclusion type defects are rejected even if they have a small surface. The patent specifically teaches that not all defects are of equal severity, justifying an attempt to determine the nature of the defect before deciding to reject the container.

[0005] In general, it would be beneficial to identify defects as early as possible in the manufacturing process of glass containers so that they can be corrected as soon as possible at the manufacturing facility. Thus, it would be advantageous to detect defects in containers that are directly related to the settings in the forming process, in particular, in order to correct the forming process as soon as possible if drift occurs. However, this remains complicated due to the nonlinearity of the system, the number of parameters of the process, and because one cause can produce multiple complex effects and multiple causes can explain the same effect.

[0006] For this reason, the prior art proposes methods for observing the containers in the hot state moving on the output conveyor of the forming machine, using equipment similar to that used under cold conditions for the inspection after the annealing furnace. For example, according to patents EP 0 177 004 and EP 3 516 377, the inspection of the walls is carried out using the same methods as for the cold inspection, and these patents describe in particular the means for adapting the inspection to the specific environment of the hot sector, by the arrangement of means for managing the high ambient temperatures. Thus, a light source arranged on one side of the trajectory of the containers on the output conveyor of the forming machine illuminates the containers, and at least one camera sensitive to the emitted light acquires at least one transmission image of each container. The images are analyzed to determine which containers have defects.

[0007] Patent US 6 584 805 also describes an installation for observing hot containers moving on the output conveyor of a molding machine. The installation includes an inspection station including a light source arranged on one side of the conveyor to illuminate the containers. The inspection station also includes a camera arranged on the other side of the conveyor to obtain a transmission image of the containers illuminated by the light source. The inspection station in particular makes it possible to measure the diameters of the containers at different heights. These measured diameters are compared with reference diameters to determine whether the containers are defective or not.

[0008] Since hot containers emit infrared radiation, the prior art also proposes observing hot containers moving on an output conveyor by means of an infrared camera based on the principle that thicker parts of the container emit more radiation. Patent EP 0 643 297 describes an apparatus for the analysis and diagnosis of a manufacturing process for glass products, which comprises a sensor sensitive to infrared radiation emitted by objects leaving the forming machine. The system also includes a digital processing device which compares the radiation with a mathematical reference model in order to determine existing deviations in the distribution of the glass and / or the causes of thermal stresses in the container.

[0009] Patent EP 0 679 883 describes a device for acquiring images by means of an infrared camera that collects radiation emitted by the hot containers at the output of the manufacturing machine. The infrared camera is synchronized with the movement of the container forming cavity. For inspection, processing software divides the container image captured by the infrared camera into inspection areas. By segmenting and measuring the bright or dark points in these areas, trapezoidal defects and inclusions can be identified.

[0010] According to patent DE10030649, by knowing the speed of the output conveyor and the sequence in which the containers emerge from the various sections, it is advantageously possible to link detected defects to their original cavities and to appropriately adjust the production method and correct the defects.

[0011] Therefore, the prior art teaches to inspect the hot container as soon as possible in order to quickly set or correct the forming method. The dimensional control of the diameter or height is performed through vision, and furthermore, according to patent application WO2021009456, the thickness of the glass can be measured by infrared imaging in at least two specific wavelength ranges. The dimensional deviation is itself an identified height, diameter, inclination or thickness defect, which can be assigned to at least one cause of drift in the manufacturing method.

[0012] Patent US 6 188 079 proposes to measure the thickness of the glass of a container based on infrared radiation, which comprises measuring a first intensity of radiation in a first spectral band emitted by a material between two surfaces, the outer and the inner one of the container. The first spectral band, the signal of which depends on both the temperature and the thickness of the glass, is preferably between 0.4 and 1.1 microns. The method also comprises measuring a second intensity of radiation in a second spectral band, in which the radiation is emitted substantially entirely from a single outer surface of the container. According to this patent, the second spectral band, the radiation of which depends only on the temperature, corresponds to the surface radiation and is preferably between 4.8 and 5 microns. The method comprises determining the thickness between the outer and inner surfaces of the container as a function of a combination of the first and second measured intensities. In other words, the thickness and the temperature are determined from two radiation measurements taken in the first and second spectral bands.

[0013] The situation is different for so-called "visual defects", i.e. any type of visual defect such as inclusions (ceramics, metals, etc. foreign bodies), bubbles, wrinkles, streaks (surface grooves), glazes (cracks), fins, trapezoids, oil stains, very thin areas, unmelted particles, etc. These visual defects appear in the image as local optical variations or pixels with deviations from the background. These visual defects can be of significant concern if they lead to a risk to the consumer, to the risk of container breakage or loss of function. However, the recognition of visual defects based on the image can be ambiguous. Due to the safety margin implemented during detection, a container may be considered defective even if it is compliant.

[0014] Furthermore, it should be noted that in the images it is possible to distinguish acceptable artifacts, e.g. stamps or decorations, faintly marked mold joints, etc. It also appears necessary to accurately identify the nature of the visual defects and distinguish them from other defects in order to identify the critical defects. To avoid critical defects, it is necessary to improve the reliability of the classification of visual defects. In addition to improving the identification of defects increases production efficiency, the identification of defects allows to determine their potential causes and to operate the manufacturing equipment according to the category of the detected defects. Indeed, without a reliable identification of the detected defects, no safe decision can be made to correct the manufacturing method, whether manual or automatic. Summary of the Invention [Problem to be solved by the invention]

[0015] The present invention aims to overcome the drawbacks of the prior art by proposing a method for controlling the quality of hot glass containers, which method is designed to detect defects more effectively, in particular visual defects, to ensure their safe and reliable identification and to provide more complete information regarding the corrections to be made to the control parameters of the method for manufacturing glass containers by the manufacturing facility. [Means for solving the problem]

[0016] The object of the present invention is a method for inspecting hot glass containers leaving a manufacturing facility and identifying the type of defects on the containers, the method comprising the steps of: obtaining at least one transmission image of the container illuminated by a source of light passing through the container and at least one infrared radiation image of the container; - analyzing at least one transmission image and at least one infrared radiation image; - determining a match between at least a portion of the transmission image and at least a portion of the infrared radiation image; classifying the container from the matched at least one transmission image and at least one infrared radiation image to identify at least one defect type for the container; Includes.

[0017] According to one advantageous feature of the embodiment, the container is illuminated by a light source having an emission spectrum in the wavelength range below 0.8 μm and an infrared radiation image of the container in the wavelength range above 0.8 μm is obtained.

[0018] For example, an infrared radiation image of the container is taken with the light source turned off.

[0019] According to another example embodiment, the infrared radiation image is acquired at a viewing direction such that the light emitted by the light source is not captured by the infrared radiation of the container.

[0020] According to one aspect of the invention, to confirm a match between the transmission image and the infrared radiation image, the method detects candidate regions in the transmission image and the infrared radiation image, the method including, for each container: Matching candidate regions of the transmission image or candidate regions of the infrared radiation image with corresponding regions of the infrared radiation image and the transmission image, respectively, depending on their location on the container; Or matching candidate regions of the transmission image with candidate regions of the infrared radiation image; Check.

[0021] According to one variant of embodiment, the method comprises, as matching, determining a combination of a transmission image and an infrared radiation image to obtain a composite image, the method comprising the steps of: Extracting classification features from the composite image to provide classification criteria for transmission and radiation; and classifying containers using transmission and radiation classification criteria; Check.

[0022] According to another variant of embodiment, the method comprises, as matching, determining a combination of a transmission image and an infrared radiation image to obtain a composite image, the method comprising the steps of: Segmenting the composite image to detect candidate regions for synthesis; Extracting classification features from synthesis candidate regions to indicate transmission and radiation classification criteria; Classifying the container using the transmission and radiation classification criteria of the synthesis candidate area; Check.

[0023] According to one aspect of the present invention, The classification criteria for transparency are extracted from the transparent images, Radiation classification criteria are extracted from infrared radiation images, · Containers are classified using transmission and radiation classification criteria.

[0024] Advantageously, radiation classification criteria are selected for the infrared radiation images, transmission classification criteria are selected for the transmission images and / or combined criteria are selected taking into account the properties of a logical or mathematical combination of the transmission images and the infrared radiation images, these radiation and transmission classification criteria being position, size, shape or photometric criteria.

[0025] Advantageously, the method comprises classifying the containers by a supervised learning classifier, the input data of which are: Radiation and transmission classification criteria, or radiographic and x-ray images, or a part of a radiographic image and a part of a radiographic image, It is.

[0026] According to one exemplary embodiment, the method includes classifying the container using a supervised learning classifier that uses as input data at least one composite image obtained by combining at least one radiographic image of the container with at least one transmission image, or by combining a region of at least one radiographic image with a corresponding region of at least one transmission image.

[0027] According to another exemplary embodiment, the method includes classifying containers with a supervised learning classifier trained with a training database including a set of records, each of which includes, for an observed exemplary container: at least one radiographic image of the container, at least one radiographic image of the container, and at least one label for assigning at least one classification of the object to the exemplary container from a list of possible classifications, such as a type of defect; Or, at least one radiographic image region of the exemplary container, at least one transillumination image region of the exemplary container, and at least one label that assigns at least one classification of the object from a list of possible classifications, such as a type of defect, to a corresponding region of the exemplary container; Includes.

[0028] According to one advantageous feature of the invention, the method aims to classify each container according to at least one classification of objects from a list of possible classifications including at least the type of defect, the list of possible classifications including at least no defects, trapezoidal shape, inclusions, air bubbles.

[0029] According to one advantageous feature of the method, a step is carried out of taking into account at least one type of detected defects in order to estimate adjustment information for at least one control parameter of the manufacturing facility.

[0030] Another object of the present invention is to provide an apparatus for inspecting hot glass containers exiting a manufacturing facility and identifying types of defects in the containers, the apparatus comprising: a system for acquiring a transmission image of the container and an infrared radiation image of the container; an information processing unit connected to the image acquisition system; Including, The information processing unit is a system for analyzing at least one transmission image and at least one infrared radiation image of the container; a system for matching at least one region of a transmission image of the container with at least one region of at least one infrared radiation image; a container classifier for identifying at least one type of defect for the container based on at least one region of the at least one transmission image and at least one region of the at least one infrared radiation image that are matched to each other; The device is configured to include:

[0031] According to one exemplary embodiment, a system for acquiring a transmission image of a container and an infrared radiation image of the container includes, on the one hand, a camera provided with a lens and sensitive to infrared radiation emitted from the container, and, on the other hand, a light source passing through the container and a camera provided with a lens and sensitive to light transmitted through the container.

[0032] Advantageously, the image acquisition system includes a system for selecting light emitted from the light source, said system being arranged to remove light emitted from the light source from the radiation captured by the infrared sensitive camera.

[0033] For example, a system for acquiring a transmission image of a container and an infrared radiation image of a container may include: A light source to illuminate the container, A sensor sensitive to infrared radiation emitted by the container; A sensor sensitive to light emitted from a light source and transmitted through the container; a common optical lens for recovering the infrared radiation emitted by the container and the light transmitted through the container from the radiation received by the infrared-sensitive sensor, said optical lens being associated with an optical separation and filtering system for removing the light emitted by the light source; Includes.

[0034] According to one aspect of the embodiment, the information processing unit comprises: - an ejection device to control the ejection of containers identified as defective; and / or A display device for presenting to the operator the identified defects, a transmission image of the container and an infrared radiation image; is connected to

[0035] Typically, the information processing unit is connected to a production ECU that manages the manufacturing equipment, receiving from the production ECU the containers, their images and time information that allows associating their detected defects with a mold number or a molding cavity; Sending the identified defects and the results of the performed measurements to a production ECU, enabling the production ECU to automatically estimate adjustment information for at least one control parameter of the manufacturing facility. [Brief description of the drawings]

[0036] [Figure 1] FIG. 1 is a simplified diagram of an apparatus according to the present invention capable of inspecting hot glass containers exiting an exemplary manufacturing facility.

[0037] [Diagram 2] FIG. 2 shows one exemplary embodiment of an image acquisition system implemented in an inspection apparatus according to the invention for containers exiting a manufacturing facility.

[0038] [Diagram 3] FIG. 3 shows another exemplary embodiment of an image acquisition system for a container being discharged from a manufacturing facility, implemented in an inspection device according to the invention.

[0039] [Figure 4] FIG. 4 is a simplified functional block diagram of a first exemplary embodiment of an inspection device according to the present invention, which performs what is referred to as conventional processing of the information contained in the transmission image and the infrared radiation image.

[0040] [Diagram 5]FIG. 5 is a simplified functional block diagram of a first exemplary embodiment of an inspection apparatus according to the present invention, which performs a segmentation operation on a composite image obtained by matching a transmission image and an infrared radiation image.

[0041] [Figure 6] FIG. 6 is a simplified functional block diagram of a first exemplary embodiment of an inspection apparatus according to the present invention, which implements a convolutional neural network having as input data the matched candidate regions resulting from the segmentation operations of the transmission image and the infrared radiation image.

[0042] [Figure 7] FIG. 7 is a simplified functional block diagram of a first exemplary embodiment of an inspection apparatus according to the present invention, which implements two convolutional neural networks having as input data the candidate regions resulting from a segmentation operation of a transmission image or an infrared radiation image.

[0043] [Figure 8] FIG. 8 is a simplified functional block diagram of a second exemplary embodiment of an inspection apparatus according to the present invention implementing a convolutional neural network having as input data a composite image of a transmission image and an infrared radiation image.

[0044] [Figure 9] FIG. 9 is a simplified functional block diagram of a second exemplary embodiment of an inspection apparatus according to the present invention that implements a convolutional neural network having as input data all or part of a transmission image and an infrared radiation image without prior segmentation.

[0045] [Figure 10A] FIG. 10A is a simulated transmission image showing the defect. [Figure 10B] FIG. 10B is a simulated infrared image showing the same defect when the defect is a filled cavity. [Figure 10C]FIG. 10C is a simulated infrared image showing the same defect when the defect is stuck glass.

[0046] [Figure 11A] FIG. 11A is a transmission image showing the defect. [Figure 11B] FIG. 11B is a simulated infrared image showing the same defect where the defect is a piece of glass. [Figure 11C] FIG. 11C is a simulated infrared image showing the same defect when the defect is an oil stain.

[0047] [Figure 12A] FIG. 12A is an infrared image. [Figure 12B] FIG. 12B is a transmission image showing an example where a pebble defect may be confused with a blister defect.

[0048] [Figure 13A] FIG. 13A is an infrared image. [Figure 13B] FIG. 13B is a transmission image showing another example where a bleb defect may be confused with a pebble defect.

[0049] [Figure 14A] FIG. 14A is an infrared image. [Figure 14B] FIG. 14B is a transmission image showing another example where a blister defect may be confused with a pebble defect. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0050] 1 shows an apparatus 1 according to the invention for inspecting hot glass containers 2 exiting various types of known manufacturing or forming facilities 3. The inspection apparatus 1 aims to detect for each container whether the container has defects and, for defective containers, to identify the type of defect among a group of defects.

[0051] At the output of the production facility 3, the containers 2, such as glass bottles or vials in the illustrated example, have a high temperature, typically in the range of 300° C. to 600° C. In a known manner, the containers 2 freshly formed by the facility 3 are handled by an output conveyor 5, on which, in the illustrated example, they are successively placed, forming a row of containers. The containers 2 are conveyed by the conveyor 5 in a row along the conveying direction and are transported successively to various treatment stations, in particular an annealing furnace. Upstream of the annealing furnace is arranged a coating hood 6, which generally constitutes the first of the post-forming treatment stations. Advantageously, the inspection device 1 according to the invention inspects the containers in the hot state at the first surface treatment station, i.e. upstream of the coating hood 6.

[0052] The manufacturing facility 3 is known per se, but an example will be briefly described to make it possible to understand the interaction between the inspection device 1 according to the invention and the manufacturing facility 3.

[0053] The manufacturing facility 3 includes a production ECU 7 for managing the various functions of the molding facility 3. Conventionally, the manufacturing facility 3 includes several different molding sections that operate in parallel and continuously supply at least one glass container. In the example of an IS machine, the different molding sections each include at least one blank mold and at least one blow mold that receives a glass parison. In known methods, the molding sections, the blank mold and the blow mold from which each container 2 emerges can be identified and the sequence of container movement for a given production run is known.

[0054] The inspection system 1 according to the invention comprises a system 10 for acquiring a transmission image It of the container 2 and an infrared radiation image Ir of the container 2, and an electronic information processing unit 11 connected to the acquisition system 10. This electronic information processing unit 11 is any type of computer system, including a computer, external peripherals (display unit, keyboard, etc.), programs, databases, etc. This information processing unit 11 is connected to the production ECU 7 and receives from the production ECU 7, if necessary, the containers 2, their images and time information for relating their detected defects to a mould number or a moulding cavity. Typically, the operation of the image acquisition system 10 is synchronized with the operation of the container moulding cavity. Furthermore, this information processing unit 11 transmits the identified defects and the results of the performed measurements to the production ECU 7, enabling the production ECU to automatically estimate adjustment information for at least one control parameter of the manufacturing facility 3. Such adjustment of the control parameter is performed manually or automatically. Finally, the information processing unit 11 is connected to an ejection device for controlling the ejection of containers identified as defective and / or to a display device for presenting to an operator the identified defects, a transmission image and an infrared radiation image of the container.

[0055] The acquisition system 10 makes it possible to observe each container 2 in two ways: infrared radiation of the hot container and transmission of light passing through the same container. The system 10 for acquiring transmission images and infrared radiation images of a container can be made in any suitable manner. According to the example shown in Fig. 1, the acquisition system 10 comprises, on the one hand, a camera 13 sensitive to the infrared radiation emitted by the container 2 and provided with a lens 13a, and, on the other hand, a light source 14 passing through the container and a camera 15 sensitive to the light transmitted through the container and provided with a lens 15a.

[0056] According to one advantageous feature of the embodiment, the acquisition system 10 comprises a system for selecting the light emitted by the light source 14, said system 10 being arranged to remove said light from the radiation captured by the infrared-sensitive camera 13. In other words, the acquisition device 10 is arranged such that the infrared-sensitive camera 13 captures only infrared radiation from the inspected container. Of course, this aim can be achieved in different ways.

[0057] For example, the infrared radiation image of the container is acquired with the light source 14 turned off. According to another exemplary embodiment, the infrared radiation image is acquired in a viewing direction such that the infrared radiation of the container does not capture the light emitted by the light source 14, as in the example shown in Fig. 1. According to another exemplary embodiment, the light source 14 does not emit within the sensitivity spectrum of the sensor of the infrared-sensitive camera 13. According to one feature of the invention, the radiation spectrum of the light source 14 is in the wavelength range below 0.8 μm, and the sensor of the infrared-sensitive camera 13 captures infrared radiation in the wavelength range above 0.8 μm.

[0058] The acquisition system 10 may also include optical filters so that the infrared sensitive camera 13 captures only the infrared light from the inspected container. These optical filters may be mounted anywhere between the light source 14 and the infrared sensitive camera 13.

[0059] 2 illustrates another exemplary embodiment of an acquisition system 10 including a light source 14 that illuminates the container 2, a sensor 13b sensitive to infrared radiation emitted by the container 2, and a sensor 15b sensitive to light emitted by the light source 14 and transmitted through the container. The acquisition system 10 also includes a common optical lens 18 for collecting the infrared radiation emitted by the container and the light transmitted through the container. The optical lens 18 is associated with an optical separation and filtering system 19 for removing the light emitted by the light source 14 from the radiation received by the infrared sensitive sensor 13b.

[0060] 3 shows another variant of the embodiment of the acquisition system 10, which includes a camera with a common lens 18 for collecting infrared radiation emitted by the container and light transmitted through the container. The acquisition system 10 also includes two juxtaposed linear sensors arranged in the focal plane of the camera, one of the sensors 13b being sensitive to infrared radiation and the other sensor 15b being sensitive to light emitted by the light source 14. The linear sensors 13b, 15b are arranged along the axis of symmetry of the container, i.e. perpendicularly. Of course, the acquisition system 10 also includes a system for selecting the light emitted by the light source 14, arranged to remove it from the radiation captured by the infrared sensitive sensor 13b.

[0061] The acquisition system 10 takes, for each container 2, one or more transmission images It and infrared radiation images Ir, each of which is a two-dimensional image. The acquired images are processed by an information processing unit 11, which: a system for analyzing at least one transmission image It and at least one infrared radiation image Ir, A system for matching at least one region of a transmission image with at least one region of at least one infrared radiation image; a container classifier for identifying at least one type of defect on the container from at least one region of the at least one transmission image and at least one region of the at least one infrared radiation image that are matched to each other; The device is configured to include:

[0062] The information processing unit 11 is thus suitable for executing an inspection method for detecting defects in containers and for identifying one type of defect among a group of defects for the container. As shown in Figures 4 to 9, the inspection method according to the invention comprises executing, for each container, an operation Act of acquiring at least one transmission image It of the container 2 illuminated by a light source 14 of light transmitted through the container, and an operation Acr of acquiring at least one infrared radiation image of the container. Of course, this image acquisition operation can be performed for all containers or only for some of the containers.

[0063] The method according to the invention then comprises carrying out an operation of analysing at least one transmission image and at least one infrared radiation image.The method according to the invention then comprises confirming the execution of an operation or step MC of matching at least a portion of the transmission image It with at least a portion of the infrared radiation image Ir.The method according to the invention then comprises carrying out a classification step CI to identify at least one type of defect Dk for the container from the information contained in the at least one transmission image and the at least one infrared radiation image matched to one another.

[0064] According to a first embodiment, which is carried out in the exemplary embodiment of Figs. 4-7, the analysis operation aims at processing the image to extract regions corresponding to visible features or objects, if any, of this object. An object in a digital image is usually a set of related pixels that have common characteristics that are not present in adjacent sets. Thus, the object is enclosed by a closed contour and is recognized as an object only based on the characteristics of the image, i.e. grey level, etc.

[0065] The objects correspond to areas or regions of the image where defects may be present. Regions with objects, also called candidate regions, represent objects that may be classified as belonging to a class of objects from a list of classifications that may contain a particular type of defect. The list of object classes may be, for example, · Marks, shadows, decorations, emblems, strings, etc. of containers that form mold joints that are not defects Surface defects such as wrinkles, streaks, drips, orange peel or tears, Shape defects such as fins, trapezoids, plugged necks, Internal defects in the material, such as cracks, inclusions, blisters, pebbles, air bubbles, Thin spots, which are areas with poor glass distribution; It is.

[0066] According to this more accurate approach, the method according to the invention aims at classifying each container into at least one class belonging to a set of classes, some of which include the type of defect.

[0067] According to the invention, the classification of the container is performed by classification of the images or parts of the transmission image It and the infrared radiation image Ir. It should be noted that often there are several different defects on the same container. Apparently, according to the invention, a first region of the container can be classified according to a first part of the transmission image It and a first part of the infrared radiation image Ir corresponding to this first part of the container, and also, for the same container, a second region of the container can be classified according to a second part of the transmission image It and a second part of the infrared radiation image Ir corresponding to this second part of the container.

[0068] The invention aims to ensure production and to quickly correct errors in the process by classifying the container or the image or part of the image. This advantage is particularly important when a critical defect called trapezoid or bird feather occurs in the process. This defect is a glass thread inside the container, the end of which is connected to the inner wall. The most typical trapezoid is in the form of a loose string with a downward curved arc, crossing straight. This defect can break and lead to glass breakage in future bottled liquids. This is dangerous for the consumer. Therefore, it is very important that containers with trapezoid shapes should never be shipped. This filtering is carried out smoothly by cold or hot, visual or infrared inspection. On the other hand, to correspond to the production process, it is necessary to detect the defect and recognize it, preferably in the hot state. However, if this defect is detected due to the presence of a feature in the image, it turns out to be very polymorphic. It may be that the thread is missing, only one or both attachments are visible, or it may not have the typical arch shape. According to the invention, a classification based on two images in transmission and infrared mode makes it possible to recognize the trapezoid shape, since at least one transmission image (usually the two are made according to different observation angles) will precisely reveal a birdcage-shaped defect at the attachment, while the infrared radiation image will signal the presence of excess glass. In fact, it is preferable to use infrared light, which passes through the glass and is sensitive to thickness. It is thus possible to immediately inform an operator who acts on the basis of the manufacturing method and / or automatically, for example by modifying the temperature of the blank or the movement of the reverser. Thus, according to the invention, the trapezoid shape is part of the list of possible classes of objects.

[0069] Another example of a particular advantage of defect classification is that small inclusions, especially ceramic, are often confused with small air bubbles in only one of the observation modalities. This confusion is disadvantageous because these defects do not have the same severity and the same cause in the manufacturing process. As can be seen in the following Figures 12A-12B and 13A-13B, classification based on images in two modalities, transmission and infrared, allows to distinguish them and thus act correctly on the method, for example in the case of air or steam bubbles, to improve the refinement of the glass or to improve the lubrication of the scissors and the conduit called "delivery conduit".

[0070] The operations of analyzing the transmitted image It and the digital infrared radiation image Ir are known per se, in particular performing filtering and segmentation operations carried out in order to extract all regions that have objects. The method thus performs an operation SRt of segmenting and detecting candidate regions on the transmitted image It, and an operation SRr of segmenting and detecting candidate regions on the infrared radiation image Ir (figures 4, 6 and 7). According to the exemplary embodiment shown in figure 5, the operation SR of segmenting and detecting candidate regions is carried out on a composite image of the transmitted image It and the infrared radiation image Ir, as will be explained in the following description.

[0071] A segmentation operation conventionally consists in dividing an image into regions or zones, i.e. in assigning pixels belonging to regions. This segmentation operation aims at determining candidate regions in each image, through filtering, thresholding, contour tracing operations etc., but generally without necessarily measuring the parameters characterizing these regions. This image segmentation operation is performed according to a filtering method adapted to the modality, i.e. to the transmission image It and the infrared radiation image Ir.

[0072] These segmentation operations SRt, SRr, SR make it possible to detect candidate image areas defined by a contour limited to the object, these candidate image areas RTC, RRC and RCC being respectively transmission, infrared radiation or a composite image area of ​​transmission and infrared radiation. These segmentation operations SRt, SRr, SR also make it possible to detect candidate image areas defined by rectangles surrounding the objects RTE, RRE, RCE, these candidate image areas being respectively transmission and infrared radiation or these composite image areas being regions of transmission and infrared radiation. These segmentation operations SRt, SRr, SR also make it possible to detect candidate image areas defined by enlarged rectangles surrounding the objects RTL, RRL, RCL, taking into account the context of the object, these candidate image areas being respectively transmission, infrared radiation or a composite image area of ​​transmission and infrared radiation.

[0073] The method according to the invention performs an MC operation of matching candidate regions of the transmitted image or candidate regions of the infrared radiation image with corresponding regions of the infrared radiation image and the transmitted image, respectively, depending on their location on the container, which may also involve matching candidate regions of the transmitted image with candidate regions of the infrared radiation image.

[0074] This MC matching operation aims to check the matching of regions in the two images by comparing their respective positions on the container. In the most general case, a geometric transformation from one image to the other is determined, which makes it possible, starting from a region or pixel of the container image, to identify the region or pixel of the other image that corresponds to the same region or elementary part of the container. The geometric transformation can be of any type required, including, for example, translation / rotation, anamorphosis, change of proportions, etc.

[0075] According to one embodiment variant, the two images or image regions of the container are matched pixel by pixel. To do this, a geometric transformation is determined for every pixel. It is also possible to calculate a transformed image for one of the two images or image parts that can be superimposed on the other image or image part. A geometric transformation and an interpolation of pixel values, for example a bilinear interpolation, is then applied to all pixels of the relevant region.

[0076] If the image acquisition system 10 matches the transmission image and the infrared radiation image pixel by pixel, the matching is performed directly. In this case, the acquisition device must be constructed very precisely so that the camera sensors have the same field of view, magnification, observation direction and resolution in pixels / mm. Thus, the matching is already performed, since each pixel of the transmission image and each pixel of the infrared radiation image correspond to the same exemplary surface part of the container. Of course, deviations from this ideal situation can be corrected by matching using appropriate geometric transformations.

[0077] According to another embodiment variant, areas are matched on the container whose centers or centroids are close, i.e. matched or adjacent to each other by geometric transformation, or areas where rectangles enclosing the objects RTE, RRE, RCE or enlarged rectangles enclosing the objects RTL, RRL, RCL intersect or overlap on a certain percentage of a given surface on the container.

[0078] Also, this MC matching can be done from candidate region to candidate region or from pixel to pixel as shown in the illustrated exemplary embodiment.

[0079] According to the exemplary embodiment shown in figures 5 and 8, a composite image IC is created by matching an image or image part It of transmission with an image or image part Ir of radiation. The method thus obtains the composite image IC by checking the combination of the transmission image and the infrared radiation image as a match. The MC matching operation can concern all the transmission and infrared radiation images or only a part of these images. This MC matching is performed pixel by pixel as described above. Each pixel pc(x,y) of coordinates x,y of the composite image is assigned a value that depends on the transmission obtained from the transmission image It and the infrared radiation obtained from the image of radiation Ir. This value is for example either a 16-bit scalar pc(x,y) with 8 transmission bits and 8 infrared radiation bits, or a vector vc(x,y) whose components {vt(x,y),vr(x,y)} are respectively a scalar of transmission and infrared radiation. The simplest method is to directly read the value of the pixel of the transmission image It and the value of the corresponding pixel of the image of radiation Ir. However, it is also possible to build each composite pixel from a combination of several adjacent pixels in the images, or from an interpolated value.

[0080] According to the exemplary embodiment shown in Figures 4, 6 and 7, matching is done from candidate region to candidate region. To match candidate regions on two different modalities of images, it is possible to perform a one-to-one image mapping. It is also possible to directly search for candidate regions located in the same area of ​​the container.

[0081] The method according to the invention aims at determining radiographic and transmission classification criteria, including transmission criteria taking into account n features ti of the transmission image, infrared criteria taking into account m features ri of the infrared radiation image and / or combination criteria taking into account q features ci of a logical or mathematical combination of the transmission and infrared radiation images, such as position, size, shape (concave, perimeter, surface, etc.) or photometric (mean level, contrast, variance, texture, etc.) features.

[0082] According to the exemplary embodiment shown in figures 4 and 5, the method according to the invention performs operations ECt, ECr, EC of extracting classification criteria. According to the exemplary embodiment shown in figure 4, the method performs an operation ECt of extracting classification criteria for candidate regions RTC, RTE, RTL of the transmission image, making it possible to define, for each candidate region, a vector Ct of dimension n representing n features ti obtained from the image of radiation It. Similarly, the method according to the invention performs operations ECr of extracting classification criteria for candidate regions RRC, RRE, RRL from the infrared radiation image, making it possible to define, for each candidate region, a vector Cr of dimension m representing m features ri obtained from the image of radiation Ir.

[0083] It should be noted that according to the exemplary embodiment shown in FIG. 4, the matching operation MC between the candidate regions RTC, RTE, RTL of the transmitted image and the candidate regions RRC, RRE, RRL of the infrared radiation image results in a vector Cc of dimension n+m+q representing the n+m+q features ti, ri and ci obtained for each candidate region matched between the transmitted image and the infrared radiation image.

[0084] In an exemplary embodiment shown in figure 5, the method according to the invention performs an operation EC of extraction of transmission and emission classification criteria for a synthetic transmission and infrared emission candidate region RCC, RCE or RCL obtained after the division operation SR. This extraction operation results in a vector Cc of dimension n+m+q representing the n+m+q features ti, ri and ci obtained for each candidate region matched between the transmission image and the infrared emission image.

[0085] In the exemplary embodiment of figures 4 and 5, the classification criteria are determined by a preliminary analysis, i.e. business knowledge or statistical study. Image analysis algorithms determine the location, size, shape and photometric features ti, ri, ci. In the exemplary embodiment shown in figures 6 and 7, the classification criteria are determined by supervised learning embedded in trained neural networks CNN, CNN1, CNN2. The training set contains pairs of candidate regions (RTC, RRC), (RTE, RRE), (RTL, RRL) according to the two inspection modalities and as labels the classes of objects in a list of possible classes such as defect types.

[0086] Using pre-determined classification criteria of transmission and radiation, the method classifies the defects and, consequently, the containers having these defects. The classification operation allows the class of the candidate region or object Dk of the container to be determined among p possible classes D1, D2, ... Dp. If the candidate region is found in only one of the two images according to the first modality, an analysis of the defect is performed according to the criteria associated to the type of image used, but also taking into account the criteria associated to the other modality, i.e. the analysis is based on a combination of the criteria associated to the two types of images. The principle of the invention is based on taking into account the two inspection modalities to provide additional reliable information for making the decision to classify the object or container and for identifying the defect.

[0087] The classification decision assigns the container a class Dk of belonging among p possible classes, the p classes being mainly types of defects, in particular related to air bubbles, wrinkles, streaks, sagging marks, orange peel, cracks, fins, trapezoids, plugged necks, blisters, pebbles, mould joint marks, shadows, ornaments, coats of arms, strings, thin spots etc. If this defect has different shapes, for example trapezoid 1 and trapezoid 2, several classes can be provided for the same defect. According to one preferred exemplary embodiment, the aim of the invention is to classify each container according to at least one class of object in a list of possible classes comprising at least a type of defect, the list of possible classes comprising at least no defect, trapezoid, inclusions and air bubbles.

[0088] In the exemplary embodiment shown in Figures 4 and 5, the classifier C1 can be, for example, a support vector machine (SVM), a Bayesian classifier, or a neural network NN. The classifier C1 performs the classification using pre-determined classification criteria of transmission and emission. The classifier C1 is trained by a supervised learning method, which includes determining the parameters of the classifier from a set of objects or images of known classes, called a training set, and generally a test set. The training set includes pairs of feature vectors Ct, Cr associated with pairs of regions, preferably defect types, i.e. {Ct, Cr, Dk}. Typically, the supervised learning classifier is trained by a training database that includes a set of records, each of which includes a feature vector Ct, Cr associated with an exemplary vessel that it has observed. at least one radiographic image of the container, at least one radiographic image of the container, and at least one label for assigning at least one class of object to the exemplary container from a list of possible classes, such as a type of defect; Or, at least one radiographic image region of the exemplary container, at least one transmission image region of the exemplary container, and at least one label that assigns at least one class of membership in a list of object classes, such as a type of defect, to the corresponding region of the exemplary container; Includes.

[0089] According to an exemplary embodiment shown in FIG. 6, the classifier is a convolutional neural network CNN having as input data the pairs of candidate regions (RTC, RRC), (RTE, RRE) or (RTL, RRL) obtained after the matching operation MC.

[0090] According to the exemplary embodiment shown in Fig. 7, the classifier comprises a first convolutional neural network CNN1 having as input data the candidate regions of transmission (RTC, RTE, RTL) obtained after an operation SRt of segmenting and detecting candidate regions on the transmission image It. The classifier also comprises a second convolutional neural network CNN2 having as input data the candidate regions of infrared radiation (RRC, RRE, RRL) obtained after an operation SRr of segmenting and detecting candidate regions on the infrared radiation image Ir. The first convolutional neural network CNN1 and the second convolutional neural network CNN2 operate in parallel on the candidate regions of transmission (RTC, RTE, RTL) and on the candidate regions of infrared radiation (RRC, RRE, RRL), respectively. These two regions are associated by a matching operation MC according to the technique described above, according to two modalities.

[0091] The output of the first convolutional neural network CNN1 and the second convolutional neural network CNN2 is the input data of a classifier, for example of SVM, random forest, Bayesian type, preferably neural network NN, and can be classified according to two modes. The output of the first convolutional neural network CNN1 and the second convolutional neural network CNN2 is, for example, a hypothesis of the class of belonging. However, they can be more complex data with vectors of dimensions larger than the number p of classes. The learning set of the neural network includes pairs of candidate regions (RTC, RRC), (RTE, RRE), (RTL, RRL) according to the two inspection modes, and includes as labels the class of the object in a list of possible classes, such as defect type.

[0092] According to a second embodiment, which is carried out in the embodiment of figures 8 and 9, the operation of analysing the image does not aim to extract candidate regions from the image, but to take into account all or part of the transmission image It and the infrared radiation image Ir, without carrying out a prior segmentation operation. According to these two embodiments, the analysis operation is based on the execution of a neural network. If only parts of the image are analysed, these parts preferably correspond to one or more areas of interest of the container, such as the finish, the neck, the shoulder, the body, the rim or the right or left half side, or the area where the imprint is present.

[0093] In the embodiment shown in Fig. 8, an operation MC of matching the transmission image It and the radiation image Ir is performed to obtain a composite image IC, which is obtained by combining at least one radiation image with at least one transmission image of the container or by combining an area of ​​at least one radiation image with a corresponding area of ​​at least one transmission image.

[0094] The transmission image It and the radiographic image Ir are taken during the acquisition operations Act, Acr performed by the image acquisition system 10 as explained in the above description. This pixel-to-pixel image matching MC is performed as explained in the exemplary embodiment of FIG. 5. This composite image IC is used as input data for a convolutional neural network CNN, which, through learning, can take into account the location, size, shape and photometric features important for the intended classification. The learning set comprises the composite images IC or synthetic image regions that have as labels the classes of objects in a list of possible classes, such as types of defects. The transmission and radiometric classification criteria are taken into account in the weights obtained from the learning and definition of the convolutional neural network CNN. It should be noted that, unlike the examples of FIGS. 5 and 6, in this variant no segmentation operation is necessary. This is because the convolutional neural network procedure, through learning, can classify the images according to their content without prior segmentation and determine positionally the location, size, shape and photometric features important for the classification. However, the segmentation operation is possible, for example, by replacing in FIG. 5 the extraction of features EC and the classifier CL by a classifier based on a convolutional neural network CNN.

[0095] In the exemplary embodiment shown in Fig. 9, the transmission image It (in whole or in part) taken by the image acquisition system 10 is used as input data for a first convolutional neural network CNN1 capable of determining, through learning, the location, size, shape and photometric features important for the provided classification. The infrared radiation image Ir (in whole or in part) taken by the image acquisition system 10 is used as input data for a second convolutional neural network CNN2 capable of determining, through learning, the location, size, shape and photometric features important for the intended classification.

[0096] The first convolutional neural network CNN1 and the second convolutional neural network CNN2 operate in parallel on two candidate images of each modality, respectively. The outputs of the first convolutional neural network CNN1 and the second convolutional neural network CNN2 are input data for a classifier, for example SVM, Random Forest, Bayesian, preferably neural network NN, which allows the container to be classified according to both modalities. It should be noted that the two candidate images of each modality on which the first convolutional neural network CNN1 and the second convolutional neural network CNN2 operate are associated with a matching operation.

[0097] The outputs of the first convolutional neural network CNN1 and the second convolutional neural network CNN2 are, for example, hypotheses of class belonging. However, they can be more complex data with vectors of dimensions larger than the number p of classes. The learning set of the neural network contains pairs of images according to the two inspection modalities and, as labels, the class of the object in the list of possible classes, such as the type of defect. The classification criteria of transmission and radiation are taken into account in the weights obtained from the learning and definition of the neural networks CNN1, CNN2 and NN. It should be noted that, unlike the example of FIG. 7, in this cooperation example no segmentation operation is necessary. This is because the convolutional neural network procedure allows, through learning, to classify images according to their content without prior segmentation and to determine positionally the location, size, shape and photometric features that are important for classification.

[0098] The object of the present invention is to be advantageously used within the framework of a manufacturing facility to allow better detection and classification of defects present in containers formed at high temperatures, some defects being more easily identified, detected and classified according to one of two modes or by a combination of the two modes.

[0099] Figure 10A is a simulated image of the defect seen in transmission, at the level of the hole in the container, while Figures 10B and 10C are simulated images of the same defect seen in infrared when the cavity is filled or when the glass is stuck, respectively. The defect gives the impression of a trapezoidal shape contained in a cavity with low absorption. It is as if an air bubble has formed between the hole and its top, leaving only the glass thread. It could also be another piece of glass stuck to the wall of the container or caught in the hole inside the container. The infrared image could confirm the presence of the glass thread, which emits more light, and the air bubble, which emits less light.

[0100] If the contrast in the transmission mode changes significantly and the contrast in the infrared radiation mode changes significantly, we can conclude that the material contained in the bubble has accumulated at this position, which corresponds to the trapezoid shape (Figure 10B). If the contrast in the transmission mode changes significantly and the contrast in the infrared radiation mode changes weakly, two pieces of glass are stuck at the height of the depression (Figure 10C).

[0101] FIG. 11A is an image of a defect seen in transmission, while FIGS. 11B and 11C are simulated images of the same defect seen in infrared when it is a piece of glass or an oil stain, respectively. This could be a serious defect, i.e. a trapezoid, i.e. a piece of glass contained within the container. But it could also be a trace of oil visible on the container wall, i.e. a superficial defect. In the case of a thick piece of glass contained within the container, its infrared radiation will add to that of the container wall, increasing in intensity. Conversely, in the case of a simple oil stain, there should not be a significant difference between the infrared radiation of the container wall and that of the superficial defect.

[0102] A large change in the transmission contrast and a large change in the infrared emission contrast indicate a material accumulation at this location corresponding to the trapezoid (Figure 11B). A large change in the transmission contrast and a weak change in the infrared emission contrast indicate a defect absorbing visible light, probably a grease stain at the surface level of the vessel (Figure 11C).

[0103] Figures 12A, 13A and 14A are real images obtained with an infrared camera in InGaAs technology, and Figures 12B, 13B and 14B are real images of the same defect seen in transmission, the acquisition of the container images being made at the output of a manufacturing machine in a manufacturing plant in which a camera is installed using an inspection device according to the present invention.

[0104] Figure 12A is an image of a defect seen in infrared and Figure 12B is an image of the same defect seen in transmission. According to the prior art, it is unclear whether the defect observed in the transmission image of Figure 12B belongs to the class of pebble or blister. According to the present invention, the infrared radiation image shows the defect with infrared corresponding to the presence of glass, eliminating the confusion and concluding that a pebble is present.

[0105] Figure 13A is an image of a defect seen in infrared and Figure 13B is an image of the same defect seen in transmission. The defect in the transmission image suggests a pebble or blister defect. However, in the infrared emission image, it can be seen that the defect is barely emitting and has less contrast than in Figure 12A. Thus, the defect is a blister.

[0106] Figure 14A is an image of the defect as seen in infrared, and Figure 14B is an image of the same defect as seen in transmission. The defect in the transmission image has a clear transition in the middle of the black stain, suggesting a blister defect. However, in the infrared radiometric image, the entire defect is radial, revealing the defect to be a pebble.

[0107] These different examples show the advantage of using two inspection modalities to improve both defect detection and classification for the purposes of the present invention. The second inspection modality can confirm the classification of the defect performed using the first modality or invalidate it by allowing the defect to be classified in another class. For defects that are barely visible in one modality, images from the other modality provide additional information to correctly classify the defect.

[0108] Depending on the two modalities considered, defects may have weaker signals. The conversion to a composite image may reveal objects that were too weak in contrast in the two modalities. Performing the merge makes it easier to find those objects.

[0109] The invention applies to any method of manufacturing glass containers, including bottles, pots, vials, syringes, light bulbs, table glasses, jars, plates, etc. In fact, in all these manufacturing methods there is a long step of cooling the glass object after forming, and it is useful to inspect and recognize defects as early as possible.

Claims

1. 1. A method for inspecting hot glass containers (2) exiting a manufacturing facility (3) to identify types of defects in the containers, comprising the steps of: For each container, obtaining at least one transmission image (It) of the container illuminated by a light source (14) of light passing through the container, and at least one infrared radiation image (Ir) of the container; analyzing at least one of the transmission images and at least one of the infrared radiation images; determining a match between at least a portion of the transmission image and at least a portion of the infrared radiation image; classifying the container from the matched at least one transmission image and at least one infrared radiation image to identify at least one of the defect types for the container; 4. A method for testing comprising:

2. 2. The method of claim 1, wherein the container is illuminated by the light source (14) having an emission spectrum in the wavelength range below 0.8 μm and the infrared radiation image of the container is obtained in the wavelength range above 0.8 μm.

3. 3. The method of claim 1 or 2, wherein the infrared radiation image of the container is acquired with the light source (14) turned off.

4. 3. The method of claim 1 or 2, wherein the infrared radiation image is acquired at a viewing direction such that the light emitted from the light source (14) is not captured by the infrared radiation of the container.

5. To confirm a match between the transmitted image and the infrared radiation image, the method includes detecting candidate regions in the transmitted image and the infrared radiation image; For each container, - matching the candidate regions (RTC, RTE, RTL) of the transmission image or the candidate regions (RRC, RRE, RRL) of the infrared radiation image with corresponding regions of the infrared radiation image and the transmission image, respectively, according to their location on the container; or matching the candidate region of the transmission image with the candidate region of the infrared radiation image; The inspection method according to claim 1 , further comprising:

6. said matching identifying a combination of said transmission image and said infrared radiation image to obtain a composite image (IC); The method comprises: extracting classification features from the composite image to indicate transmission and radiation classification criteria; and classifying said container using said classification criteria of transmission and radiation. The inspection method according to claim 5, further comprising:

7. said matching identifying a combination of said transmission image and said infrared radiation image to obtain a composite image; The method comprises: Segmenting the composite image to detect candidate regions for synthesis; extracting classification features from the synthesis candidate regions to indicate transmission and radiation classification criteria; classifying the container using the classification criteria of transmission and radiation of the synthesis candidate region; The inspection method according to claim 1 , further comprising:

8. A classification criterion for the transmission is extracted from the transmission image; Radiation classification criteria are extracted from the infrared radiation image; The container is classified using the classification criteria for transmission and radiation. The inspection method according to any one of claims 1 to 7.

9. 9. The inspection method of claim 8, wherein a radiation classification criterion is selected for the infrared radiation image, a transmission classification criterion is selected for the transmission image, and / or a combination criterion is selected taking into account properties of a logical or mathematical combination of the transmission image and the infrared radiation image, and these radiation and transmission classification criteria are position, size, shape, or photometric criteria.

10. classifying the container with a supervised learning classifier; The input data for the classifier is The classification criteria for radiation and transmission, or the radiation image and the transmission image, or a part of the radiation image and a part of the transmission image, The inspection method according to claim 1 ,

11. 10. The inspection method according to claim 1, further comprising classifying the container by a supervised learning classifier using as input data at least one composite image (IC) obtained by combining at least one of the radiographic images with at least one of the transmission images of the container, or by combining a region of at least one of the radiographic images with a corresponding region of at least one of the transmission images.

12. classifying the container with a supervised learning classifier trained with a training database including a set of records; Each of the sets of records includes, for an exemplary vessel observed: at least one label that assigns at least one classification of an object to the exemplary container from a list of possible classifications, such as a type of defect, based on at least one of the radiographic images of the container, at least one of the transmission images of the container, and at least one label that assigns at least one classification of an object to the exemplary container from a list of possible classifications, such as a type of defect; Or, at least one radiographic image region of the exemplary container, at least one transmission image region of the exemplary container, and at least one label that assigns at least one classification of an object from a list of possible classifications, such as a type of defect, to a corresponding region of the exemplary container; The inspection method according to claim 10 or 11, comprising:

13. 13. The method of claim 1, wherein each container is classified according to at least one classification of object from a list of possible classifications including at least a type of defect, the list of possible classifications including at least no defect, trapezoidal, inclusions, and air bubbles.

14. 14. The inspection method according to claim 1, wherein a step is performed of taking into account at least one type of detected defects in order to estimate adjustment information for at least one control parameter of the manufacturing facility.

15. 1. An apparatus for inspecting hot glass containers (2) exiting a manufacturing facility (3) to identify types of defects in the containers, comprising: a system (10) for acquiring a transmission image of the container and an infrared radiation image of the container; an information processing unit (11) connected to the image acquisition system (10); Including, The information processing unit includes: a system for analyzing at least one of the transmission images and at least one of the infrared radiation images of the container; a system for matching at least one region of the transmission image of the container with at least one region of at least one of the infrared radiation images; a container classifier for identifying at least one type of defect for a container based on at least one region of at least one of the transmission images and at least one region of the at least one of the infrared radiation images being matched to one another; An apparatus configured to include:

16. The apparatus of claim 15, wherein the system (10) for acquiring the transmission image of the container and the infrared radiation image of the container comprises, on the one hand, a camera (13) sensitive to infrared radiation emitted from the container and provided with a lens (13a), and, on the other hand, a light source (14) of light passing through the container and a camera (15) sensitive to light transmitted through the container and provided with a lens (15a).

17. 16. The apparatus of claim 15, wherein the image acquisition system (10) includes a system (19) for selecting light emitted from the light source (14), the system being arranged to remove light emitted from the light source from radiation captured by the infrared sensitive camera.

18. The system (10) for acquiring the transmission image of the container and the infrared radiation image of the container comprises: A light source (14) for illuminating said container; a sensor (13b) sensitive to infrared radiation emitted from said container; a sensor (15b) sensitive to light emitted from the light source (14) and transmitted through the container; a common optical lens (18) for collecting infrared radiation emitted from the container and light transmitted through the container from the radiation received by the infrared-sensitive sensor, the optical lens (18) being associated with an optical separation and filtering system for removing light emitted from the light source; 16. The apparatus of claim 15, comprising:

19. The information processing unit (11) an ejection device for controlling the ejection of the containers identified as defective; and / or a display device for presenting the identified defects, the transmission image of the container, and the infrared radiation image to an operator; 19. The device according to any one of claims 15 to 18, wherein the device is connected to

20. The information processing unit (11) is connected to a production ECU (7) that manages the manufacturing equipment (3), receiving from the production ECU time information enabling the containers, their images, and their detected defects to be associated with a mold number or molding cavity; transmitting the identified defects and results of the performed measurements to the production ECU, enabling the production ECU to automatically estimate adjustment information for at least one control parameter of the manufacturing equipment; 20. Apparatus according to any one of claims 15 to 19.

Images