RU2716465C1 - Optical inspection method and device using it - Google Patents

Abstract

FIELD: optical inspection.

SUBSTANCE: invention relates to optical inspection. Device for optical inspection of surface comprises: at least two radiation sources with different position and direction of emission; at least one of aperture and radiation-absorbing box, configured to block radiation, other than radiation scattered on surface; at least one chamber; and a processor comprising a trained classifier based on artificial intelligence and configured to: perform preprocessing of images, where preprocessing includes at least background removal, detect defect on at least one surface, where defect is detected on at least one surface, if at least one image from said set of images displays scattered radiation; determine the defect type, and whether the defect is false or true, by evaluating the detected defect by the trained classifier based on artificial intelligence. Determination whether the defect is false or true is performed based on a certain type of defect.

EFFECT: invention increases accuracy of determining defects.

26 cl, 6 dwg

Description

BACKGROUND OF THE INVENTION

FIELD OF THE INVENTION

[0001] The present invention relates, in general, to the field of optical verification and, in particular, to a method and apparatus for optical verification of a surface.

Description of the Related Art

[0002] Currently, a person checks the surface for defects in production is the most famous way to control product quality. Such verification includes not only visual detection of defects, distinguishing between true (non-removable) defects, for example, scratches, dents, etc., from false (removable) defects, such as dust, stains, fibers, etc., assessment the size of the defects, as well as the use of a cleaner to confirm the truth or falsehood of the defect. In addition, there are technical solutions that automatically carry out optical surface checks.

[0003] Such technical solutions are known, for example, from the following sources.

[0004] US Patent US7061598 B1, issued June 13, 2006, entitled "DARKFIELD INSPECTION SYSTEM HAVING PHOTODETECTOR ARRAY", provides a dark field surface inspection tool that includes a light source for illuminating an article and generating a light scattering pattern. The light scattering pattern is configured such that the positions of the light rays of the scattering pattern are uniquely related to the scattering angles of the light rays when they are scattered on the product. The tool also includes a photodetector array for detecting a light scattering pattern. The photodetector matrix generates an electrical signal, which is received by the electronics of the instrument signal processing unit and can be used to distinguish defects on the product. The invention also includes methods for checking a surface in a dark field. However, this technical solution does not provide a determination of the truth or falsehood of the defect, it is possible to check only a flat surface, requires sophisticated equipment in the form of a complex aspherical mirror.

[0005] US Patent US6169601 B1, issued January 2, 2001, entitled "METHOD AND APPARATUS FOR DISTINGUISHING PARTICLES FROM SUBSURFACE DEFECTS ON A SUBSTRATE USING POLARIZED LIGHT", provides a technical solution in which particles distinguish from subsurface defects on the surface of the substrate by defect with P – polarized light and S – polarized light and integrate the light to obtain a complete integral response and calculate the P – S ratio. Particles are distinguished from substrate defects by comparing the P – S ratio to a predetermined threshold value. However, this technical solution provides verification of only a flat surface and has a low sensitivity of 0.25 μm.

[0006] In the publication of patent application US2018232875 A1, published on 08.16.2018, entitled “COSMETIC DEFECT EVALUATION”, a method for assessing cosmetic defects of a device is proposed, which includes capturing multiple images of a device under a variety of lighting conditions with one or more cameras, processing images of a device with one or multiple cameras, identifying one or more device defects and calculating a defect index based on one or more defects. However, this technical solution does not provide a determination of the truth or falsehood of the defect and has a low sensitivity at which only large defects are recognized.

[0007] In the publication of patent application EP3088873, published on November 9, 2016, entitled "CONTAINER INSPECTION SYSTEM WITH INDIVIDUAL LIGHT CONTROL", a glass packaging verification system and methods for checking glass packaging are provided. The system includes a panel that includes many light sources for illuminating glass containers. The system includes a camera for receiving images of illuminated glass containers by each of the light sources. The system includes a controller for adjusting the amount of energy supplied to each of the light sources individually. The system includes a processor for evaluating the image of the illuminated glass containers for signs of defects in the container. However, this technical solution does not provide a determination of the truth or falsehood of the defect and has a low sensitivity at which only large defects are recognized.

[0008] US Patent US9182357 B2, November 10, 2015, entitled "SEMICONDUCTOR WAFER INSPECTION SYSTEM AND METHOD", provides a semiconductor wafer inspection system that includes a camera having a field of view, mounting an object to position at least a portion of surface 5 of object 3 in plane 15 of the object relative to the camera and within the field of view of the camera, and at least one surface portion 41 carrying a plurality of retroreflective reflectors 95 located at a greater distance from the camera than the plane of the object, and within the field of view of the camera. However, this technical solution does not provide a determination of the truth or falsehood of the defect, provides verification only of a flat surface and has a low sensitivity.

[0009] The present invention is made to eliminate at least one of the above problems and disadvantages, and to provide at least one of the advantages described below.

SUMMARY OF THE INVENTION

[0010] An object of the present invention is to provide a method and apparatus for optically checking a surface that provides at least one of the following advantages:

- quick and accurate determination of the type of defect and whether the defect is false or true in automatic mode;

- verification of both flat and curved surfaces;

- simultaneous verification of two surfaces of the product, if the product is made of a material that transmits the radiation used for testing;

- high test sensitivity, in which it is possible to detect defects with dimensions parallel to the plane (or on the surface) of the product about 1 μm and perpendicular to the plane (or surface) of the product less than 30 nm;

- short test time, less than 2 seconds for each tested section of the plane (surface) of the product;

- a large area of each checked section of the product plane, more than 40˟35 mm2;

- the use of simple equipment.

[0011] One aspect of the present invention provides a method for optical surface verification, comprising the steps of: sequentially emitting radiation from at least two radiation sources to at least one surface of the article, wherein the positions of the at least two radiation sources relative to at least one surface differ from each other, and the directions of emission of the radiation of at least two radiation sources are different from each other; blocking at least one of the at least one aperture and the radiation absorbing duct against the at least one camera reflecting and / or direct radiation emitted by at least two radiation sources, in addition to radiation scattered on at least one surface; receive at least one camera a set of images of at least one surface, wherein each image from said set of images is obtained by emitting radiation from at least one radiation source from at least two radiation sources; pre-processing images from said set of images, wherein the pre-processing includes at least removing a background; detecting a defect on at least one surface, the defect being detected on at least one surface if scattered radiation display is detected on at least one image from said image set; determine the type of defect and whether the defect is false or true, by evaluating the detected defect by a trained classifier based on artificial intelligence, while determining whether the defect is false or true is performed based on the specific type of defect.

[0012] In a further aspect, the article may comprise two surfaces opposite to each other, wherein if said article is made of radiation-transmitting material, then a defect is detected on one or two surfaces.

[0013] In another further aspect, the at least two radiation sources can form at least two pairs of radiation sources, the radiation sources in each pair of radiation sources being opposite each other, the radiation sources in each pair of radiation sources emitting radiation at the same time, radiation sources emit radiation sequentially and the directions of emission of radiation of at least two pairs of radiation sources are different from each other.

[0014] In yet a further aspect, the preprocessing of said set of images may further comprise at least one of a contrast setting of said images, histogram equalization, contrast limited adaptive histogram equalization.

[0015] In yet a further aspect, detecting a defect on at least one surface may comprise at least one of binarizing said images, detecting contours on said images, spectrally analyzing said images, clustering said images, performing segmentation of said images, and processing said images using Markov random fields.

[0016] In yet another further aspect, a trained classifier based on artificial intelligence can be one of a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on a residual convolutional neural network, while determining the type of defect and whether the defect is false or true is performed by entering the said set of images into a trained classifier based on artificial intelligence, while the classifier is based on the arts nnogo intelligence pre-trained on a set of training sets of images of different defects, indicating types of defects.

[0017] In yet a further aspect, the method may further comprise, prior to the step of determining the type of defect and whether the defect is false or true, the step of extracting defect symptoms from said image set by means of a trained classifier based on artificial intelligence, wherein determining the type of defect and whether the defect is false or true is performed by a trained classifier based on artificial intelligence according to the extracted signs of the defect, while the classifier is based on artificial intelligence is pre-trained on a set of training sets of signs of various defects indicating types of defects, and the trained classifier based on artificial intelligence is one of the classifiers based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (ensemble of decision trees), the classifier based on a fully connected neural network, a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on new residual convolutional neural network, a descriptor-based classifier.

[0018] In yet a further aspect, the method may further comprise determining the suitability or defect of at least one surface based on predetermined conditions by analyzing the defect or set of defects.

[0019] Another aspect of the present invention provides an apparatus for optically checking a surface, comprising: at least two radiation sources configured to sequentially emit radiation on at least one surface of the article, wherein the positions of the at least two radiation sources are relative to at least one the surfaces are different from each other, and the directions of emission of the radiation of at least two radiation sources are different from each other; at least one of the at least one aperture and radiation absorbing ducts configured to block reflected and / or direct radiation emitted from the at least one camera emitted by at least two radiation sources other than the radiation scattered by at least one surface; at least one camera configured to receive a set of images of at least one surface, wherein each image from said set of images is obtained by emitting radiation from at least one radiation source from at least two radiation sources; and a processor comprising a trained classifier based on artificial intelligence and configured to: pre-process the images from said set of images, the pre-processing including at least removing the background, detecting a defect on at least one surface, the defect being detected on at least one surface, if at least one image from the said set of images detected display of scattered radiation; determine the type of defect and whether the defect is false or true, by evaluating the detected defect by a trained classifier based on artificial intelligence, while determining whether the defect is false or true is performed based on the specific type of defect.

[0020] In an additional aspect, the product may contain two surfaces that are opposite to each other, while if the product is made of radiation-transmitting material, the processor detects a defect on one or two surfaces.

[0021] In another further aspect, the device may comprise at least two pairs of radiation sources, wherein the radiation sources in each pair of radiation sources are opposite each other, the radiation sources in each pair of radiation sources emit radiation simultaneously, while the pairs of radiation sources emit radiation in series and directions for emitting radiation of at least two pairs of radiation sources are different from each other.

[0022] In yet a further aspect, in the pre-processing of said set of images, the processor may further be configured to perform at least one of adjusting the contrast of said images, equalizing the histogram, and contrast-limited adaptive equalization of the histogram.

[0023] In yet a further aspect, in the process of detecting a defect on at least one surface, the processor may be configured to perform at least one of binarization of said images, detecting contours on said images, spectral analysis of said images, clustering said images, performing segmenting said images; and processing said images using Markov random fields.

[0024] In another additional aspect, a trained classifier based on artificial intelligence can be one of a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on a residual convolutional neural network, while the processor can determine the type of defect and whether the defect is false or true by entering the said set of images into a trained classifier based on artificial intelligence, and the classifier on the basis of artificial intelligence is pre-trained on a variety of training sets of images of different defects indicating the types of defects.

[0025] In yet a further aspect, the processor may further be configured, prior to determining the type of defect and whether the defect is false or true, to extract signs of the defect from said set of images by means of a trained classifier based on artificial intelligence, the processor determining the type of defect and whether the defect is a false or true trained classifier based on artificial intelligence according to the extracted signs of the defect, while the classifier is based on art of intellectual intelligence is pre-trained on a set of training sets of signs of different defects indicating the types of defects, and the trained classifier based on artificial intelligence is one of the classifiers based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (ensemble of decision trees), classifier based on a fully connected neural network, classifier based on a convolutional neural network, classifier based on a deep convolutional neural network, classification Ora based on residual convolution neural network classifier based on the descriptor.

[0026] In yet a further aspect, the processor may be further configured to determine the suitability or marriage of at least one surface based on predetermined conditions by analyzing a defect or a set of defects.

[0027] In yet a further aspect, each of the at least two radiation sources may be one of a visible light source, an ultraviolet radiation source, an infrared radiation source, the radiation may be polarized or non-polarized, monochromatic, or with a predetermined spectrum.

[0028] In yet a further aspect, at least one camera may be selected in accordance with the emission spectrum of the radiation sources, each camera from at least one camera may be one of a color camera, a monochrome camera, an ultraviolet camera, an infrared camera each camera from at least one camera may contain spectral and / or polarizing filters.

[0029] In yet a further aspect, said two surfaces may be flat or curved, or one surface of said two surfaces may be flat, and the other surface of said two surfaces may be curved.

[0030] In yet a further aspect, at least two radiation sources and at least one camera may be located on the side of one surface of said two surfaces.

[0031] In yet a further aspect, at least two radiation sources may be located on the side of one surface of said two surfaces, and at least one camera may be on the side of another surface of said two surfaces.

[0032] In another additional aspect, at least two radiation sources can be located both on the side of one surface of the two surfaces, and on the other surface of the two surfaces, and at least one camera can be on the side of one surface of said two surfaces or on the side of another surface of said two surfaces.

[0033] In yet a further aspect, the device may comprise one aperture located between at least one camera and one surface of said two surfaces.

[0034] In yet a further aspect, the device may comprise one aperture located between at least two radiation sources and one surface of said two surfaces.

[0035] In yet a further aspect, the device may comprise two apertures, wherein one aperture is between at least two radiation sources from one side of said two surfaces and said one surface, and the other aperture is between at least two radiation sources from the side of another surface of said two surfaces and said other surface.

[0036] In yet a further aspect, the device may comprise a radiation absorbing duct located on the side of said one surface opposite to said other surface, on the side of which there is at least one chamber.

BRIEF DESCRIPTION OF THE DRAWINGS

[0037] The above and other aspects, features and advantages of the present invention will be more apparent from the following detailed description, taken in conjunction with the accompanying drawings, in which:

[0038] FIG. 1 is a flowchart of an optical surface verification method.

[0039] FIG. 2 is a block diagram illustrating an embodiment of an apparatus for optically checking a surface.

[0040] FIG. 3 is a block diagram illustrating another embodiment of an apparatus for optically checking a surface.

[0041] FIG. 4 is a block diagram illustrating another embodiment of an apparatus for optically checking a surface.

[0042] FIG. 5 is a block diagram illustrating another embodiment of an apparatus for optically checking a surface.

[0043] FIG. 6 is an exemplary arrangement of radiation sources with respect to the product being tested.

[0044] In the following description, unless otherwise indicated, the same reference numbers are used for the same elements when they are depicted in different drawings, and their parallel description is not given.

DETAILED DESCRIPTION OF EMBODIMENTS OF THE PRESENT INVENTION

[0045] The following description with reference to the accompanying drawings is provided in order to facilitate a thorough understanding of various embodiments of the present invention defined by the claims and its equivalents. The description includes various specific details to facilitate such an understanding, but these details should be considered only exemplary. Accordingly, those skilled in the art will find that various changes and modifications to the various embodiments described herein can be developed without departing from the scope of the present invention. In addition, descriptions of well-known functions and constructions may be omitted for clarity and conciseness.

[0046] The terms and wording used in the following description and claims are not limited to bibliographic meanings, but simply used by the creator of the present invention to provide a clear and consistent understanding of the present invention. Accordingly, it should be apparent to those skilled in the art that the following description of various embodiments of the present invention is provided for illustration only.

[0047] It should be understood that the singular forms include the plural, unless the context clearly indicates otherwise.

[0048] Additionally, it should be understood that the terms “comprises”, “comprising”, “includes” and / or “including” when used in this application, mean the presence of the stated features, meanings, operations, elements and / or components, but do not exclude the presence or addition of one or more other features, values, operations, elements, components and / or groups thereof.

[0049] Hereinafter, various embodiments of the present invention are described in more detail with reference to the accompanying drawings.

[0050] The present invention is intended to optically verify the surface of an article. The present invention provides a quick and accurate determination of the type of defect and whether the defect is false or true in automatic mode with high sensitivity and short check time. In addition, the present invention provides verification of both flat and curved surfaces of the product. The present invention provides for simultaneous verification of two surfaces of an article if the article is made of a material that transmits the radiation used for verification.

[0051] The present invention may have, for example, the following applications:

- to check the quality of the surface of the screen of a TV, smartphone, tablet computer, etc.,

- to check the quality of jewelry,

- in medicine to detect age-related facial wrinkles, flaw detection of the surface of the cornea of the eye, etc.,

- quality control of precise optical elements,

- in the manufacture of vehicles for the detection of defects in paintwork,

- in the production of airplanes, spaceships, sea ships for flaw detection of the surface of parts and components,

- for checking thin films, coatings, etc.,

- in microelectronics for checking the surface of semiconductor wafers.

[0052] The proposed method 100 for optical surface verification, which can be implemented by the device 200 for optical surface verification, is described with reference to FIG. 1.

[0053] In step 101, the radiation of at least two radiation sources 201 is sequentially emitted to at least one surface of the article, the positions of the at least two radiation sources relative to the at least one surface being different from each other, and the directions of emission of the radiation of at least two radiation sources 201 are different from each other.

[0054] In step 102, at least one of the at least one aperture 202 and radiation-absorbing duct 203 are blocked from entering at least one camera 204 of reflected and / or direct radiation emitted by at least two radiation sources 201 other than radiation, scattered on at least one surface of the product.

[0055] In step 103, at least one camera 204 obtains a set of images of at least one surface of the product, each image from said set of images being obtained by emitting radiation from at least one radiation source 201 from at least two radiation sources 201.

[0056] At step 104, the processor 205 performs preprocessing of images from said set of images, wherein the preprocessing includes at least background removal.

[0057] At step 105, the processor 205 detects a defect on at least one surface of the article, the defect being detected on at least one surface if scattered radiation is detected on at least one image from said image set.

[0058] In step 106, the type of defect is determined and whether the defect is false or true by evaluating the detected defect with a trained classifier based on artificial intelligence contained in the processor 205, and determining whether the defect is false or true is performed based on the specific type of defect .

[0059] If the product contains two surfaces opposite to each other, while said product is made of radiation-transmitting material, then a defect is detected on one or two surfaces.

[0060] In a further embodiment, at least two radiation sources 201 form at least two pairs of radiation sources 201, the radiation sources 201 in each pair of radiation sources 201 being opposite each other, the radiation sources 201 in each pair of radiation sources 201 at the same time, the pairs of radiation sources 201 emit radiation sequentially and the emission directions of the at least two pairs of radiation sources 201 are different from each other.

[0061] In step 104 of the image preprocessing, the method may further comprise at least one of adjusting the contrast of said images, equalizing the histogram, contrast-limited adaptive equalization of the histogram. However, image pre-processing in the present invention is not limited to the above operations. Image preprocessing can be performed using any known operations that enable the detection of a defect in images. Since such operations are known, their detailed description is not required.

[0062] In a defect detection step 105, the method may further comprise at least one of binarizing said images, detecting contours on said images, spectrally analyzing said images, clustering said images, performing segmentation of said images, and processing said images using Markov random fields. However, defect detection in the present invention is not limited to the above operations. Defect detection can be performed using any known operation that is most suitable for the particular application of the present invention. Since such operations are known, their detailed description is not required.

[0063] A trained classifier based on artificial intelligence can be one of a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on a residual convolutional neural network, while determining the type of defect and whether the defect is false or true is performed by inputting said set of images into a trained classifier based on artificial intelligence, while the classifier based on artificial intelligence is previously about uchayut on a set of training sets of images of different defects, indicating types of defects. However, the present invention is not limited to the above classifiers, any known classifiers for recognizing image contents that are most suitable for a particular application of the present invention can be applied.

[0064] The training sets of images of various defects contain images marked with various scratches, chips, scuffs, stains, specks, fibers, etc. various shapes and sizes indicating the type of defect and whether the defect is true or false, that is, not removable or removable, respectively. The training of such classifiers using a plurality of training sets of images and the processing of sets of images by such classifiers for recognition of image contents is known in the art and a detailed description thereof is not required.

[0065] The method may further comprise, until step 106 of determining the type of defect and whether the defect is false or true, the step of extracting signs of the defect from said image set by means of a trained classifier based on artificial intelligence, wherein determining the type of defect and whether the defect is false or true, they are performed by a trained classifier based on artificial intelligence according to the extracted signs of a defect, and at the same time, a classifier based on artificial intelligence is pre-trained t on the set of training sets of signs of different defects indicating the types of defects, and the trained classifier based on artificial intelligence is one of the classifiers based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (ensemble of decision trees), the classifier based on fully connected neural network, classifier based on convolutional neural network, classifier based on deep convolutional neural network, classifier based on residual convolutional neural hydrochloric network classifier based on the descriptor, such as, but not limited to, SIFT (Scale-invariant transformation characteristics), VGG16 (predobuchennoy feature vector of the neural network), SURF (stable acceleration Signs). However, the present invention is not limited to the above classifiers, any known classifiers can be applied to recognize the contents of the images according to the extracted features, most suitable for the specific application of the present invention.

[0066] The symptoms may be the geometry (shape and size) of different defects, the modulation of radiation scattered by different defects, the texture of different defects, the spectrum of radiation scattered by different defects, the structure of different defects, the results of clustering images of different defects, etc. The training of such classifiers using a variety of training sets of features and the processing of features by such classifiers to recognize image content is known in the art and a detailed description thereof is not required.

[0067] The method may further comprise determining by the processor 205 the suitability or defect of at least one surface based on predetermined conditions by analyzing a defect or a set of defects. Specified conditions for determining the suitability or marriage of a product can be, for example, the type of defect, an indication of whether the defect is true or false, the size of the defect, the number of defects on the surface / product, the distance between defects, etc. depending on the quality requirements that the product must meet.

[0068] Embodiments of an optical surface verification apparatus 200 are described in further detail below with reference to FIG. 2-6. The device 200 is capable of operating according to embodiments of a method 100 for optical surface verification.

[0069] The device 200 for optical surface verification comprises at least two radiation sources 201, at least one of at least one aperture and radiation-absorbing duct, at least one camera, and a processor 205 comprising a trained classifier based on artificial intelligence . It should be understood that embodiments of the device 200 described with reference to FIG. 2-6 are exemplary and non-limiting, and device 200 may have other configurations for arranging its components, such as radiation sources 201, aperture (s) 202, radiation-absorbing box 203, and camera (s) 204. In addition, device 200 may comprise different devices I / O to monitor the operation of the device 200 and display the results of the optical check.

[0070] At least two radiation sources 201 are arranged to sequentially emit radiation on at least one surface of the article, wherein the positions of the at least two radiation sources 201 relative to at least one surface are different from each other, and the directions of emission of radiation are at least at least two radiation sources 201 are different from each other.

[0071] At least one of the at least one aperture 202 and radiation absorbing duct 203 are configured to block reflected and / or direct radiation from at least two radiation sources 201 other than radiation from entering at least one camera 204 scattered on at least one surface.

[0072] At least one camera 204 is configured to acquire a set of images of at least one surface, wherein each image from said set of images is obtained by emitting radiation from at least one radiation source 201 from at least two radiation sources 201.

[0073] The processor 205 comprises a trained classifier based on artificial intelligence and is configured to pre-process images from said image set, the pre-processing including at least removing a background, detecting a defect on at least one surface, the defect being detected on at least at least one surface, if at least one image from the said set of images detected display of scattered radiation; determine the type of defect and whether the defect is false or true, by evaluating the detected defect by a trained classifier based on artificial intelligence, while determining whether the defect is false or true is performed based on the specific type of defect.

[0074] If the product contains two surfaces opposite each other and is made of radiation-transmitting material, then processor 205 is capable of detecting a defect on one or two surfaces. The two said surfaces may be flat or curved, or one surface of said two surfaces is flat, and the other surface of said two surfaces is curved

[0075] The device 200 may comprise at least two pairs of radiation sources 201, wherein the radiation sources in each pair of radiation sources 201 are opposite each other, the radiation sources 201 in each pair of radiation sources 201 emit radiation simultaneously, while the pairs of radiation sources 201 emit the radiation in series and the emission directions of the at least two pairs of radiation sources 201 are different from each other.

[0076] An exemplary arrangement of radiation sources 201 is shown in FIG. 6, wherein the radiation sources 201 are distributed in pairs around the surface of the article, and the pairs of radiation sources 201 emit radiation in series. However, it should be understood that this is not a limiting example of the placement of radiation sources 201. The number of radiation sources 201 can be different, and they can emit light in pairs. The basic requirement for radiation sources 201 is that radiation sources 201 must be two or more, they must emit radiation in series, the positions of two or more radiation sources 201 relative to at least one surface must be different from each other, and the directions of emission of radiation of two or more radiation sources 201 must be different from each other.

[0077] In the process of pre-processing the image set, the processor 205 may additionally perform at least one of adjusting the contrast of said images, equalizing the histogram, and contrast-limited adaptive equalization of the histogram. However, as indicated above with respect to the optical surface verification method 100, image pre-processing in the present invention is not limited to the above operations.

[0078] In the process of detecting a defect on at least one surface, the processor 205 may perform at least one of binarization of said images, detecting contours on said images, spectral analysis of said images, clustering said images, performing segmentation of said images, and processing said images from using Markov random fields. However, as described above with respect to the optical surface inspection method 100, defect detection in the present invention is not limited to the above operations.

[0079] In one embodiment of the device 200, a trained classifier based on artificial intelligence may be one of a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on a residual convolutional neural network, while the processor 205 performs type determination the defect and whether the defect is false or true by entering the said set of images into a trained classifier based on artificial intelligence, and at the same time the classifier The artificial intelligence-based ikator is pre-trained on a variety of training sets of images of various defects indicating the types of defects. However, the present invention is not limited to the above classifiers, any known classifiers for recognizing image contents that are most suitable for a particular application of the present invention can be applied.

[0080] In another embodiment of the device 200, the processor 205 may further, prior to determining the type of defect and whether the defect is false or true, extract signs of the defect from said set of images by means of a trained classifier based on artificial intelligence, while the processor determines the type of defect and whether the defect is a false or true trained classifier based on artificial intelligence according to the extracted signs of the defect, while the classifier is based on artificial intelligence The teller is pre-trained on a set of training sets of signs of different defects indicating the types of defects, and the trained classifier based on artificial intelligence is one of the classifiers based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (an ensemble of decision trees), the classifier based on a fully connected neural network, a classifier based on a convolutional neural network, a classifier based on a deep convolutional neural network, a classifier based on e residual convolutional neural network, a descriptor-based classifier such as, but not limited to, SIFT (scale-invariant transformation of attributes), VGG16 (feature vector of a pre-trained neural network), SURF (Persistent Accelerated Attributes). However, the present invention is not limited to the above classifiers.

[0081] In another embodiment of the device 200, the processor 205 may further determine the suitability or marriage of at least one surface based on predetermined conditions by analyzing a defect or a set of defects. This operation and predetermined conditions have already been described with respect to the optical surface verification method 100, and therefore, such a description is omitted here.

[0082] Optionally, the radiation sources 201 may be one of a visible light source, an ultraviolet radiation source, an infrared radiation source, the radiation may be polarized or unpolarized, monochromatic, or with a predetermined spectrum.

[0083] At least one camera 204 can be selected in accordance with the emission spectrum of the radiation sources, each camera from at least one camera being one of a color camera, a monochrome camera, an ultraviolet camera, an infrared camera, and each camera from at least one camera may contain spectral and / or polarizing filters.

[0084] The device 200 may include one or more cameras 204. The arrangement of the device 200 with multiple cameras 204 can provide an increase in the field of view, with one camera 204 seeing a portion of the surface / surfaces to be checked, and several cameras 204 covering the entire area of the surface / surfaces to be checked. The arrangement of the device 200 with multiple cameras 204 can also provide images of the entire area of the surface / surfaces or parts of the surface / surfaces to be examined from different angles for different orientations of the cameras 204.

[0085] Each camera 204 may be coaxial or misaligned with the axis of symmetry of the lighting system of the device 200 containing at least two radiation sources 201.

[0086] FIG. 2 shows an arrangement of a device 200 in which at least two radiation sources 201 and at least one camera 204 are located on one side of two product surfaces.

[0087] FIG. 3 and 4 illustrate the arrangement of the device 200, in which at least two radiation sources 201 are located on the side of one surface of two surfaces of the product, and at least one camera 204 is located on the side of the other surface of two surfaces of the product.

[0088] FIG. 5 shows an arrangement of a device 200 in which at least two radiation sources 201 are located both on the side of one surface of two surfaces of the product and on the other side of two surfaces of the product, and at least one camera 204 is located on the side of one surface from two surfaces of the product or from the side of another surface of two surfaces of the product.

[0089] FIG. 2 and 3 illustrate the arrangement of the device 200, in which the device 200 comprises one aperture 202 located between at least one camera 204 and one surface of the product.

[0090] FIG. 2 and 4 illustrate the arrangement of the device 200, in which the device 200 comprises one aperture 202 located between at least two radiation sources 201 and one surface of two product surfaces.

[0091] FIG. 5 shows an arrangement of a device 200 in which the device comprises two apertures 202, wherein one aperture 202 is located between at least two radiation sources 201 on the side of one surface of two product surfaces and said one surface of the product, and the other aperture 202 is between at least two sources 201 of radiation from the side of the other surface of the two surfaces of the product and said other surface of the product.

[0092] FIG. 2 to 5, there are shown the arrangement of the device 200, in which the device 200 comprises a radiation absorbing box 203 located on the side of one surface of the product, opposite the other surface of the product, on the side of which at least one chamber 204 is located.

[0093] The above descriptions of embodiments of the invention are illustrative, and modifications to the configuration and implementation are not beyond the scope of the present description. For example, although embodiments of the invention are described generally in connection with figures 1-6, the descriptions given are exemplary. Although the subject matter of the invention is described in a language characteristic of structural features or methodological operations, it is understood that the subject matter of the invention is not necessarily limited to the specific features or operations described above. Moreover, the specific features and operations described above are disclosed as exemplary forms of implementing the claims. The invention is not limited to the shown order of the steps of the method, the order can be modified by a specialist without innovative innovations. Some or all of the steps of the method may be performed sequentially or in parallel.

[0094] Accordingly, it is contemplated that the scope of an embodiment of the invention is limited only by the following claims.

Claims (39)

1. A method for optical surface verification, comprising the steps of: sequentially emit radiation from at least two radiation sources on at least one surface of the product, the positions of at least two radiation sources relative to at least one surface are different from each other, and the directions of emission of radiation of at least two radiation sources are different from each other ; blocking at least one of the at least one aperture and the radiation absorbing duct against the at least one camera reflecting and / or direct radiation emitted by at least two radiation sources, in addition to radiation scattered on at least one surface; receive at least one camera a set of images of at least one surface, wherein each image from said set of images is obtained by emitting radiation from at least one radiation source from at least two radiation sources; pre-processing images from said set of images, wherein the pre-processing includes at least removing a background; detecting a defect on at least one surface, the defect being detected on at least one surface if scattered radiation display is detected on at least one image from said image set; determine the type of defect and whether the defect is false or true by evaluating the detected defect by a trained classifier based on artificial intelligence, while determining whether the defect is false or true is performed based on the specific type of defect. 2. The method according to p. 1, in which the product contains two surfaces opposite to each other, while if the said product is made of radiation-transmitting material, then the defect is detected on one or two surfaces. 3. The method according to p. 1, in which at least two radiation sources form at least two pairs of radiation sources, the radiation sources in each pair of radiation sources being located opposite each other, the radiation sources in each pair of radiation sources emit radiation at the same time, this pair of radiation sources emit radiation sequentially and the direction of emission of radiation of at least two pairs of radiation sources are different from each other. 4. The method of claim 1, wherein pre-processing said image set further comprises at least one of a contrast setting of said images, histogram equalization, contrast-limited adaptive histogram equalization. 5. The method of claim 1, wherein detecting the defect on at least one surface comprises at least one of binarizing said images, detecting contours on said images, spectrally analyzing said images, clustering said images, performing segmentation of said images, and processing said images using Markov random fields. 6. The method according to claim 1, in which the trained classifier based on artificial intelligence is one of the classifier based on the convolutional neural network, the classifier based on the deep convolutional neural network, the classifier based on the residual convolutional neural network, while determining the type of defect and whether a defect false or true is performed by entering the aforementioned set of images into a trained classifier based on artificial intelligence, while the classifier is based on artificial intelligence and pre-trained on a set of training sets of images of different defects, indicating types of defects. 7. The method of claim 1, further comprising, prior to the step of determining the type of defect and whether the defect is false or true, the step of extracting the signs of the defect from said set of images by means of a trained classifier based on artificial intelligence, and determining the type of defect is whether the defect is false or true is performed by a trained classifier based on artificial intelligence according to the extracted attributes of the defect, while the classifier based on artificial intelligence is pre-trained and a lot of training sets of signs of different defects indicating types of defects, and the trained classifier based on artificial intelligence is one of the classifier based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (ensemble of decision trees), the classifier based on fully connected neural network, classifier based on convolutional neural network, classifier based on deep convolutional neural network, classifier based on residual convolutional neural th network classifier based on the descriptor. 8. The method according to claim 1, further comprising the step of determining the suitability or marriage of at least one surface based on predetermined conditions by analyzing a defect or a set of defects. 9. A device for optical surface verification, comprising: at least two radiation sources configured to sequentially emit radiation on at least one surface of the product, wherein the positions of the at least two radiation sources relative to at least one surface are different from each other and the directions of emission of radiation of at least two radiation sources are different apart at least one of the at least one aperture and radiation absorbing ducts configured to block reflected and / or direct radiation emitted from the at least one camera emitted by at least two radiation sources other than the radiation scattered by at least one surface; at least one camera configured to receive a set of images of at least one surface, wherein each image from said set of images is obtained by emitting radiation from at least one radiation source from at least two radiation sources; and a processor containing a trained classifier based on artificial intelligence and configured to: perform image preprocessing from said image set, wherein preprocessing includes at least background removal, detect a defect on at least one surface, wherein a defect is detected on at least one surface if scattered radiation display is detected on at least one image from said set of images; determine the type of defect and whether the defect is false or true by evaluating the detected defect by a trained classifier based on artificial intelligence, while determining whether the defect is false or true is performed based on the specific type of defect. 10. The device according to p. 9, in which the product contains two surfaces opposite to each other, while if the said product is made of radiation-transmitting material, the processor detects a defect on one or two surfaces. 11. The device according to claim 9, in which said device contains at least two pairs of radiation sources, the radiation sources in each pair of radiation sources being opposite each other, the radiation sources in each pair of radiation sources emit radiation at the same time, while the pairs of radiation sources emit radiation sequentially and the directions of emission of radiation of at least two pairs of radiation sources are different from each other. 12. The device according to p. 9, in which in the process of preprocessing said set of images, the processor is further configured to perform at least one of adjusting the contrast of said images, equalizing the histogram, contrast-limited adaptive equalization of the histogram. 13. The device according to claim 9, in which, in the process of detecting a defect on at least one surface, the processor is configured to perform at least one of binarization of said images, detecting contours on said images, spectral analysis of said images, clustering said images, performing segmenting said images; and processing said images using Markov random fields. 14. The device according to claim 9, in which the trained classifier based on artificial intelligence is one of the classifier based on the convolutional neural network, the classifier based on the deep convolutional neural network, the classifier based on the residual convolutional neural network, while the processor determines the type of defect and whether the defect is false or true by entering the said set of images into a trained classifier based on artificial intelligence, while the classifier is based on art intelligence was pre-trained on a variety of training sets of images of various defects indicating the types of defects. 15. The device according to p. 9, in which the processor is additionally configured to, before determining the type of defect and whether the defect is false or true, to extract the signs of the defect from said set of images by means of a trained classifier based on artificial intelligence, while the processor performs the determination of the type of defect and whether the defect is a false or true trained classifier based on artificial intelligence according to the extracted signs of the defect, while the classifier is based on artificial intelligence and pre-trained on a variety of training sets of signs of different defects indicating types of defects, and the trained classifier based on artificial intelligence is one of the classifiers based on XGBoost (extreme gradient improvement), the classifier based on Random Forest (ensemble of decision trees), the classifier based on a fully connected neural network, classifier based on convolutional neural network, classifier based on deep convolutional neural network, classifier based on residual ary convolution neural network classifier based on the descriptor. 16. The device according to claim 9, in which the processor is additionally configured to determine the suitability or marriage of at least one surface based on predetermined conditions by analyzing a defect or a set of defects. 17. The device according to claim 9, in which each of the at least two sources of radiation is one of a source of visible light, a source of ultraviolet radiation, a source of infrared radiation, wherein the radiation can be polarized or non-polarized, monochromatic, or with a given spectrum. 18. The device according to claim 9, in which at least one camera is selected in accordance with the radiation spectrum of the radiation sources, wherein each camera from at least one camera is one of a color camera, a monochrome camera, an ultraviolet camera, an infrared camera, each camera from at least one camera may contain spectral and / or polarizing filters. 19. The device of claim 10, wherein said two surfaces are flat or curved or one surface of said two surfaces is flat and the other surface of said two surfaces is curved. 20. The device according to p. 19, in which at least two radiation sources and at least one camera are located on the side of one surface of the two surfaces. 21. The device according to p. 19, in which at least two radiation sources are located on the side of one surface of said two surfaces, and at least one camera is on the side of another surface of said two surfaces. 22. The device according to p. 19, in which at least two radiation sources are located both on the side of one surface of said two surfaces and on the side of another surface of said two surfaces, and at least one camera is located on the side of one surface of said two surfaces or on the side of another surface of said two surfaces. 23. The device according to any one of paragraphs. 19-22, in which said device comprises one aperture located between at least one camera and one surface of said two surfaces. 24. The device according to any one of paragraphs. 19-21, in which said device comprises one aperture located between at least two radiation sources and one surface of said two surfaces. 25. The device of claim 19, wherein said device comprises two apertures, wherein one aperture is located between at least two radiation sources from one side of said two surfaces and said one surface, and the other aperture is between at least two radiation sources from the side of another surface of said two surfaces and said other surface. 26. The device according to any one of paragraphs. 19-25, in which said device comprises a radiation absorbing duct located on the side of said one surface opposite to said other surface, on the side of which there is at least one chamber.

Images