WO2019114372A1 - 一种基于人工智能的单板缺陷检测方法、系统及设备 - Google Patents

一种基于人工智能的单板缺陷检测方法、系统及设备 Download PDF

Info

Publication number
WO2019114372A1
WO2019114372A1 PCT/CN2018/108551 CN2018108551W WO2019114372A1 WO 2019114372 A1 WO2019114372 A1 WO 2019114372A1 CN 2018108551 W CN2018108551 W CN 2018108551W WO 2019114372 A1 WO2019114372 A1 WO 2019114372A1
Authority
WO
WIPO (PCT)
Prior art keywords
veneer
image
sample
illumination
board
Prior art date
Application number
PCT/CN2018/108551
Other languages
English (en)
French (fr)
Inventor
丁磊
张先耀
Original Assignee
北京木业邦科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 北京木业邦科技有限公司 filed Critical 北京木业邦科技有限公司
Publication of WO2019114372A1 publication Critical patent/WO2019114372A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
    • G01B11/00—Measuring arrangements characterised by the use of optical techniques
    • G01B11/02—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness
    • G01B11/06—Measuring arrangements characterised by the use of optical techniques for measuring length, width or thickness for measuring thickness ; e.g. of sheet material
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/17—Systems in which incident light is modified in accordance with the properties of the material investigated
    • G01N21/59—Transmissivity
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8806—Specially adapted optical and illumination features
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8854—Grading and classifying of flaws
    • G01N2021/888—Marking defects
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84—Systems specially adapted for particular applications
    • G01N21/88—Investigating the presence of flaws or contamination
    • G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
    • G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques

Definitions

  • the present application belongs to the field of artificial intelligence optical detection technology, and particularly relates to a method, system and device for detecting a single board defect based on artificial intelligence.
  • wood boards are divided into solid wood boards and wood-based boards.
  • the plywood and other glue layer substrates in the wood-based panel are bonded by a plurality of veneers.
  • high-quality veneers are used for panels of wood-based panels such as plywood, blockboard, stencils, veneers, etc.
  • lower-grade veneers are used as backboards and core plates.
  • the present application is an artificial intelligence-based automated veneer quality detection and classification method for veneer processing such as veneer.
  • the quality of the veneer shows a certain randomness due to the limitation of the veneer processing itself.
  • One of the most important features is the uneven thickness of the veneer after processing. This is due to the randomness of the growth of the wood itself, resulting in uneven hardness of the wood.
  • the thickness of the veneer is higher. The quality is higher.
  • the lower hardness portion produces a thinner veneer and is of lower quality.
  • the wood itself may also have some defects due to other factors, such as insect eyes, mineral lines, chromatic aberrations, etc. Such veneers are not suitable for use as panels.
  • the detection of veneer quality is an important part of veneer processing.
  • the traditional machine vision method can detect the color or texture defects of the veneer and cannot detect the thickness uniformity of the veneer.
  • the embodiment of the present application provides a technical solution for detecting a defect of a veneer based on artificial intelligence, which is suitable for automatic detection of veneer quality such as veneer and bamboo skin.
  • a method for detecting a single board defect based on artificial intelligence comprising:
  • the quality information of the to-be-detected board is obtained according to the result of the identification and matching.
  • the radiation source of the back-transmitted illumination is a light source.
  • defect detection model is obtained by using machine learning, and specifically includes the following steps:
  • the step of acquiring a back-smooth-irradiated single-plate sample image further includes:
  • the veneer samples are transmitted to the image acquisition zone by a transfer device that is in the same region as the illumination zone of the transmissive illumination system that produces the back-transmissive illumination.
  • the step of acquiring a back-smooth-irradiated single-plate sample image further includes:
  • the illuminating system emits light from the back side of the veneer sample through a light source.
  • the illumination intensity of the light source is controlled so that the light can penetrate the veneer sample and present an image in an image capture device placed on the front side of the veneer.
  • the illumination intensity is controlled by the controller so that the veneer sample can always penetrate through the predetermined thickness of the processed veneer;
  • the illumination intensity is automatically adjusted by the input or feedback of the image acquisition device, so that the illumination intensity can adapt to different veneer thicknesses; the image formed after the light penetrates the veneer can reflect the thickness distribution of the veneer sample.
  • the step of acquiring a back-smooth-irradiated single-plate sample image further includes:
  • the front panel illumination source is used to illuminate the veneer sample; the controller controls the illumination intensity of the front and back sides so that the light emitted by the transmitted light source can pass through the veneer to provide a clearer image.
  • annotation information includes one or more of sample image data, thickness level data of the sample, and back light intensity data.
  • the step of receiving the labeling information of the sample image of the board further includes:
  • the annotation information for the vulgar eye and/or mineral line defects presented in the image obtained by the back illumination system is received.
  • the output result of the quality information of the to-be-detected board includes: directly outputting a judgment of different quality levels; or
  • the output result further includes: identification of a defect type, and/or use classification of the board.
  • an artificial intelligence based veneer defect detection system comprising:
  • An image capturing device configured to acquire a transparent illumination image of the single board
  • a transmissive illumination device comprising a radiation source for generating radiation capable of penetrating the veneer, and enabling the transmissive irradiation to be acquired by the image acquisition device;
  • the quality detecting device is configured to identify the defect of the board by using the image acquired by the image collecting device, and output the recognition result.
  • the recognition result includes at least one of a determination of a quality level, an area of uneven thickness or thickness distribution, identification of a defect type, and use classification of a veneer.
  • the radiation source is a visible light source
  • the veneer is a veneer
  • the light source is a tunable light source, such that an image formed after the light penetrates the single board can reflect the thickness distribution of the single board sample.
  • the illumination intensity of the light source is controlled by the controller such that the light can always penetrate the single-plate sample at a predetermined thickness of the processed veneer;
  • the illumination intensity is automatically adjusted by input or feedback from the image capture device such that the illumination intensity can be adapted to different veneer sample thicknesses.
  • the transparent illumination device further includes a front illumination module, and the controller controls the front and back illumination intensity so that the light emitted by the transmission source can pass through the single board to present a clearer image.
  • the quality detecting device further includes an labeling module for marking a single board feature that can be presented in the image sample of the image capturing device by the transparent irradiation of the transparent illumination device.
  • the quality detecting device further includes: an automatic detecting module having an automatic detecting model, configured to input the image sample marked by the labeling module into the automatic detecting model.
  • the automatic detection model trains the neural network in combination with corresponding attributes, which are preset detection attributes or custom detected attributes.
  • the illumination intensity is input as a separate input to the neural network along with the image samples, normalizing the effect of the illumination intensity.
  • an artificial intelligence-based single-board defect detecting device including: a transmitting device for carrying a board through an image capturing area; and a communication module for connecting with the remote server; And a server connected to the detecting device;
  • the detecting device is capable of performing the artificial intelligence based single board defect detecting method described above;
  • the detecting device further includes the artificial intelligence based single board defect detecting system described above.
  • a computer readable medium wherein a plurality of instructions are stored, the instructions being adapted to be loaded by a processor and to perform an artificial intelligence based veneer defect detection as described above method.
  • a processor adapted to be loaded by a processor and to perform an artificial intelligence based veneer defect detection as described above method.
  • the quality information of the to-be-detected board is obtained according to the result of the identification and matching.
  • an artificial intelligence based single board defect detection system comprising:
  • a memory for storing instructions
  • an artificial intelligence based single board defect detection method such as:
  • the quality information of the to-be-detected board is obtained according to the result of the identification and matching.
  • the method, system and device provided by the embodiments of the present invention can also identify some types of defects, such as insects, mineral lines and the like in the veneer, on the premise of satisfying the defect detection of the thickness of the veneer.
  • some types of defects such as insects, mineral lines and the like in the veneer.
  • the technical solution provided by the present application is more intelligent, which greatly reduces the labor cost; meanwhile, the detection result is more accurate and faster.
  • 1 is an embodiment of the present application obtaining a veneer sample image through a back permeable illumination system
  • FIG. 2 is an image of a veneer sample obtained using a front projection light source in accordance with an embodiment of the present application
  • Figure 3 is a diagram showing an example of a veneer quality labeling in an embodiment of the present application.
  • FIG. 4 is a schematic diagram of a convolutional neural network in an embodiment of the present application.
  • FIG. 5 is a schematic structural diagram of a neural network for vene quality in an embodiment of the present application.
  • FIG. 6 is a structural block diagram of an example of a veneer automatic detecting device according to still another embodiment of the present application.
  • FIG. 7 is a structural block diagram of an example of a general-purpose computer device that implements and/or propagates the technical solution of the present application.
  • Thin wood commonly known as "wood veneer” is a wooden sheet-like thin veneer or veneer material with precious tree characteristics.
  • wood veneer There are many types of decorative thin wood (wood veneer).
  • a typical representative classification method is carried out in accordance with the manufacturing method, form, thickness, tree species, and the like of the thin wood.
  • the detection of veneer quality is an important part of veneer processing.
  • the traditional machine vision method can detect the color or texture defects of the veneer, but cannot detect the thickness uniformity of the veneer.
  • an artificial intelligence based device is proposed and is suitable for automated detection and classification of veneer quality.
  • an apparatus for deploying a detection plant comprising: a conveyor, an image sensing device, a transmissive illumination device, and a veneer quality detection device.
  • the image sensing device includes an image sensor for acquiring a sensory image, the transmissive illumination device including a radiation source for generating an irradiation capable of penetrating the veneer, and enabling the transmissive irradiation to be captured by the image sensor.
  • the veneer quality detecting device recognizes the thickness distribution of the veneer by the image information captured by the image sensing device, and outputs the recognition result.
  • FIG. 1 illustrates a typical application of the present application: an image of a veneer sample obtained by a back permeable illumination system.
  • the radiation source is preferably a visible light source, and further the source of radiation is preferably a intensity-tunable light source.
  • the specific working process is as follows: First, the veneer is transferred into an image acquisition area by the conveying device, and the image collecting area is in the same area as the transparent illumination system.
  • the transmissive illumination system projects light from the back side of the veneer sample through a light source, such as a planar light source consisting of a plurality of LED light sources.
  • the intensity of the light source is controlled so that the light can penetrate the veneer sample and the sensor placed on the front side of the veneer Present an image in .
  • the illumination intensity of the illumination system can be controlled by a controller that always penetrates the veneer sample at a given thickness of the processed veneer.
  • control system can automatically adjust the illumination intensity through input or feedback from the image acquisition device such that the illumination intensity can be adapted to different veneer thicknesses. Due to the effect of the thickness of the veneer on the light penetration, the image formed by the light penetrating the veneer can reflect the thickness distribution of the veneer sample.
  • the "back” and “front” of the veneer sample are relative concepts, not strictly defined; and the positions of the transparent light source and the image sensor are preferably interchangeable.
  • the solution can also include a communication module for connecting to the remote server and a server connected to the detection factory device.
  • the permeable illumination system further includes a frontal illumination system that controls the front and back illumination levels such that the light emitted by the transmitted light source passes through the veneer to provide a better image.
  • the controller can optimize and fix the illumination intensity in a pre-configured manner.
  • An adaptive method can also be used to adjust the light intensity on the back, or on the back and front sides.
  • a preferred method is to perform multi-intensity scanning.
  • the image capturing device is used to collect the sample image and input into the analyzer.
  • the analyzer can recognize whether the image sample carrying the thickness distribution information can be obtained under the illumination intensity, and if it can, stop changing the illumination. Intensity; if the condition is not met, continue to change the lighting conditions.
  • the transmissive illumination system in the present application is different from other illumination systems based on machine vision for supplemental light, and the technical solution of the present application converts the thickness detection as a length physical quantity measurement into an image by means of rear projection.
  • the method of identification Therefore, if the method of frontal illumination is used alone, only the texture features on the front side of the board can be identified, and any machine vision method cannot identify the thickness distribution of the veneer through the acquired image.
  • An image sample of the veneer under an example illumination system is shown in FIG.
  • an image sample of the veneer projected only by the frontal light source is shown in FIG.
  • the permeable illumination system It is one of the core inventions of this system.
  • the image sample carries the thickness distribution information of the veneer. This method is preferably applicable to the field of processing and/or inspection of veneers or similar veneers because the thickness of the veneer or similar veneer allows light to pass through, while other thicker products such as wood do not provide thickness information through the back illumination system.
  • an image sample of a veneer sample can be obtained.
  • the label is an image sample and the thickness distribution information of the sample.
  • One basic method of labeling is to mark the grade of the veneer thickness, where the grade of the thickness includes information on the thickness of the veneer and the uniformity of the thickness.
  • An advanced method of labeling is to divide the thickness level into multiple levels, corresponding to the quality level of the veneer for subsequent processing. For example, a label is as follows:
  • d is the data of one image sample, for example, may be original image data, or may be image sample data subjected to image processing.
  • Fn is the thickness-related grade.
  • the back illumination intensity S i can also be associated with an image sample as an information dimensioning dimension; wherein the letters n and i are natural numbers.
  • the thickness level can be associated with subsequent processing, so this level can also be considered as a quality level. For example, only v5 grade veneer can be used for the panel, while v5 below the veneer can only be used for the bottom plate or core board;
  • the f0 level can be used as a low quality grade with severe thickness unevenness making it unusable for any plywood processing.
  • the area and strength of the veneer thickness can be further labeled.
  • the area where the veneer thickness is below a predefined threshold or the area where the thickness distribution is not uniform is marked with an area and the quality level of the area is marked.
  • defects of the veneer may be marked, such as defects such as insect eyes, mineral lines, and the like.
  • This type of label not only labels the area and quality level, but also identifies the type of defect. That is to say, this annotation can record a more subdivided information about the quality of the veneer, which, like the thickness unevenness described above, can be presented in the image obtained by the back illumination system.
  • FIG. 3 A schematic diagram of the labeling method is given in Figure 3, where the area f1 is a dimension of the thickness level, meaning that it is an area of the f1 level.
  • f1 is not necessarily directly equal to the thickness, and may also contain information on the thickness distribution, and the f1 level is a lower level, meaning that the area is of poor quality.
  • the area is relatively bright, meaning that the thickness is thin and different patterns are distributed, meaning that the thickness distribution is not uniform.
  • the z1 area in the figure can be an annotation of a defect type, which can be represented as a mineral line or crack.
  • the labeled image samples are input into an automatic detection model, and the automatic detection model is combined with the corresponding attributes to train the neural network.
  • the neural network includes a neural network having a plurality of layers, each layer including a plurality of nodes, and a trainable weight between the two adjacent nodes.
  • FIG. 4 A schematic diagram of a convolutional neural network of an embodiment of the present application is shown in FIG. 4, which includes a plurality of convolutional and downsampled layers and a fully connected layer.
  • the convolutional layer is the core module of the convolutional neural network. By convolving with a filter, multiple nodes of the previous layer are connected to the nodes of the next layer. In general, each node of the convolutional layer is only connected to a part of the nodes of the previous layer.
  • the downsampling layer can use a max-pooling method to reduce a set of nodes into one node, preferably using a nonlinear maximum.
  • a fully connected layer After passing through multiple convolutional and downsampled layers, a fully connected layer is ultimately used to generate the detected output, and the fully connected layer connects all nodes of the previous layer to all nodes of the next layer, which is related to a traditional neural
  • the network is similar.
  • the training algorithm such as the gradient descent algorithm, changes the filter weight value in the neural network. This in turn minimizes the difference in detection between the output and the sample data.
  • the ever-changing network node values are constantly changing and improving, and the detection capability of the neural network is also improved.
  • a trained neural network is obtained, including the designed network architecture, such as the hierarchical design in Figure 4 and the connection method between the levels, as well as the trained filter weight values. These weight values are recorded and reused in later use.
  • the learning process can be done in a local inspection system or in the cloud.
  • the image acquisition device collects the image data of the veneer sample and the labeled data set is transmitted to the cloud server for model training, and the server transmits the trained model to the local processor and completes the deployment.
  • the cloud server can use training data from a variety of sources. For example, data collected and annotated from multiple local images, which in turn increases the amount of data obtained.
  • a conveyor belt carries a veneer sample through the image acquisition area, and the image acquisition area projects a light source on the back of the veneer sample through a permeable illumination system.
  • a preferred embodiment can control the illumination intensity of the back projection source by a controller, so that the image acquisition device can obtain sufficient transparent light.
  • the illumination intensity can be input to the trained neural network simultaneously with the image samples. Neural networks can perform more accurate analysis of images based on accurate light intensity. This is because the intensity of light can change the intensity of light transmitted through the veneer, which in turn affects the imaging effect of the image. Images produced by different illumination intensities may lead to misjudgment of the neural network.
  • the neural network determines that the veneer is thinner. Therefore, by using the illumination intensity as a separate input and inputting it to the neural network along with the image samples, the influence of the illumination intensity can be normalized, making the judgment of the neural network more accurate.
  • Figure 5 shows a corresponding schematic.
  • the results of the detection output can take many forms:
  • a detection output is a judgment that the neural network directly outputs different quality levels, for example, a rating information classified according to the thickness and its distribution;
  • Another detection output is that the neural network can not only give the grade judgment of the veneer quality, but also further mark the area where the thickness or thickness distribution is uneven; for example, the thickness of the veneer sample is not satisfied or the thickness distribution uniformity is indicated. In areas that do not meet the conditions, the neural network can identify and mark, and even identify and mark the quality rating and thickness information of the area;
  • Another detection output is that the neural network can not only give the grade judgment of the veneer quality and the labeling of the thickness or thickness distribution, but also identify some types of defects, such as the types of defects such as insect eyes and mineral lines in the veneer.
  • the neural network can give a classification of the use of the veneer.
  • the veneer can be adapted for any use of the later composite wood board, for example, for the surface or the back or intermediate layer of the wood board.
  • veneer defect detection including veneer.
  • random forests, support vector machines, deep belief networks, K-means, K-neighboring, etc. are also included in the scope of protection of the present application and are incorporated herein by reference.
  • an artificial intelligence based method is also proposed and applied to automatic detection and classification of veneer quality.
  • the method includes the following steps:
  • the radiation source is preferably a visible light source, and further the source of radiation is preferably a intensity-tunable light source.
  • the specific steps include:
  • the transmissive illumination system projects light from a back surface of the veneer sample through a light source, such as a planar light source composed of a plurality of LED light sources.
  • the intensity of the light source is controlled so that the light can penetrate the veneer sample and is placed on the front side of the veneer. An image is presented in the sensor.
  • This illumination intensity can be controlled by a controller that always penetrates the veneer sample at a given thickness of the processed veneer.
  • the illumination intensity can be automatically adjusted by input or feedback from the image capture device such that the illumination intensity can be adapted to different veneer thicknesses. Due to the effect of the thickness of the veneer on the light penetration, the image formed by the light penetrating the veneer can reflect the thickness distribution of the veneer sample.
  • the "back” and “front” of the veneer sample are relative concepts, not strictly defined; and the positions of the transparent light source and the image sensor are preferably interchangeable.
  • a front illumination step S13 is further included.
  • the front side is irradiated with a veneer sample.
  • the controller controls the positive and negative illumination intensity so that the light emitted by the transmitted light source can pass through the veneer to present a better image.
  • the controller can optimize and fix the illumination intensity in a pre-configured manner.
  • An adaptive method can also be used to adjust the light intensity on the back, or on the back and front sides.
  • a preferred method is to perform multi-intensity scanning.
  • the image capturing device is used to collect the sample image and input into the analyzer.
  • the analyzer can recognize whether the image sample carrying the thickness distribution information can be obtained under the illumination intensity, and if it can, stop changing the illumination. Intensity; if the condition is not met, continue to change the lighting conditions.
  • the annotation is an annotation of an image sample and thickness distribution information of the sample.
  • a label is as follows:
  • d is the data of one image sample, for example, may be original image data, or may be image sample data subjected to image processing.
  • Fn is the thickness-related grade.
  • S i is the light intensity back annotation Alternatively, in a preferred embodiment, S i may be denoted as an information dimension, be associated with the image sample.
  • the step S2 may further include the following steps:
  • This type of label not only labels the area and quality level, but also identifies the type of defect. That is to say, this annotation can record a more subdivided information about the quality of the veneer, which, like the thickness unevenness described above, can be presented in the image obtained by the back illumination system.
  • the features that can be marked must be veneer features that can be rendered in the image capture device through the back transmission illumination system, and other features that cannot be presented cannot be identified by subsequent veneer quality inspection systems/modules, even if they are labeled.
  • the labeled image sample is input into an automatic detection model (initial model), and the automatic detection model is combined with the corresponding attribute to train the neural network.
  • the neural network includes a neural network having a plurality of layers, each layer including a plurality of nodes, and a trainable weight between the two adjacent nodes.
  • the corresponding attribute is a preset detection attribute or a custom detection attribute.
  • the specific steps include:
  • the convolution layer connects a plurality of nodes of the previous layer with nodes of the next layer by a convolution operation with a filter.
  • each node of the convolutional layer is only connected to a part of the nodes of the previous layer.
  • the downsampling layer can use a max-pooling method to reduce a set of nodes into one node, preferably using a nonlinear maximum.
  • a fully connected layer is ultimately used to generate the detected output, and the fully connected layer connects all nodes of the previous layer to all nodes of the subsequent layer.
  • the learning process can be done in a local inspection system or in the cloud.
  • the image acquisition device collects the image data of the veneer sample and the labeled data set is transmitted to the cloud server for model training, and the server transmits the trained model to the local processor and completes the deployment.
  • the cloud server can use training data from a variety of sources. For example, data collected and annotated from multiple local images, which in turn increases the amount of data obtained.
  • the step S4 may further include the following steps:
  • the light intensity and the image sample are simultaneously input to the trained neural network.
  • the effect of the light intensity can be normalized, making the neural network more accurate.
  • the neural network performs a more accurate analysis of the image based on the exact illumination intensity. This is because the intensity of light can change the intensity of light transmitted through the veneer, which in turn affects the imaging effect of the image. Images produced by different illumination intensities may lead to misjudgment of the neural network.
  • the neural network outputs judgments of different quality levels. For example, a rating information based on thickness and its distribution;
  • S43 Marking an area of uneven thickness or thickness distribution in the image of the veneer sample. For example, in an image of a veneer sample, an area where the thickness does not satisfy the condition or the uniformity of the thickness distribution does not satisfy the condition is marked, and the neural network can identify and mark, and even identify and mark the quality rating and thickness information of the area;
  • Identify the defect type For example, the types of defects such as insect eyes and mineral lines in the veneer are identified and labeled.
  • the neural network gives a classification of the use of the veneer.
  • the veneer can be adapted for any use of the later composite wood board, for example, for the surface or the back or intermediate layer of the wood board.
  • the sequence number of each step does not mean the order of execution sequence, and the order of execution of each step should be determined by its function and internal logic, and should not be addressed.
  • the implementation process of the application specific embodiment constitutes any limitation; the "initial model” includes but is not limited to the untrained original model, and may also be a model trained by other batches or kinds of veneer data, but cannot be directly used for the current detection. , or any other detection model that can implement the corresponding function or effect of the present invention.
  • embodiments of the present application also provide a storage device, such as a computer readable medium, including computer readable instructions that, when executed, perform the operations of the steps of the method in the above embodiments.
  • the veneer automatic detecting device 100 may include:
  • a processor 110 a communication interface 120, a memory 130, and a communication bus 140. among them:
  • the processor 110, the communication interface 120, and the memory 130 complete communication with each other via the communication bus 140.
  • the communication interface 120 is configured to communicate with a network element such as a client.
  • the processor 110 is configured to execute the program 132, and specifically, the related steps in the foregoing method embodiments may be performed.
  • program 132 can include program code, the program code including computer operating instructions.
  • the processor 110 may be a central processing unit CPU, or an Application Specific Integrated Circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.
  • CPU central processing unit
  • ASIC Application Specific Integrated Circuit
  • the memory 130 is configured to store the program 132.
  • the memory 130 may include a high speed RAM memory and may also include a non-volatile memory such as at least one disk memory.
  • the program 132 may be specifically configured to cause the veneer automatic detecting device 100 to perform the following steps:
  • the quality information of the vulcan to be detected is obtained according to the result of the identification and matching.
  • program modules include routines, programs, components, data structures, and other types of structures that perform particular tasks or implement particular abstract data types.
  • program modules can be practiced using other computer system configurations, including hand-held devices, multi-processor systems, microprocessor-based or programmable consumer electronics, small computers, mainframe computers, and the like. It is also possible to use in a distributed computing environment where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in both local and remote memory storage devices.
  • the functions may be stored in a computer readable storage medium if implemented in the form of a software functional unit and sold or used as a standalone product.
  • the technical solution of the present application or the part contributing to the prior art, or the part of the technical solution, may be embodied in the form of a software product, which is stored in a storage medium, including
  • the instructions are used to cause a computer device (which may be a personal computer, server, or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present application.
  • a computer device which may be a personal computer, server, or network device, etc.
  • the technical solution of the present application can be implemented and/or propagated by at least one general purpose computer device 210 as shown in FIG. In FIG.
  • the general purpose computer device 210 includes: a computer system/server 212, an external device 214, and a display device 216; wherein the computer system/server 212 includes a processing unit 220, an I/O interface 222, and a network adaptation module.
  • the internal data transmission is usually implemented by a bus; further, the storage module 230 is generally composed of a plurality of storage devices, such as a RAM (Random Access Memory) 232, a cache 234, and a storage system (generally by a Or a plurality of large-capacity non-volatile storage media 236 or the like; a program 240 that implements some or all of the functions of the technical solutions of the present application is stored in the storage module 230, and generally exists in the form of a plurality of program modules 242.
  • a RAM Random Access Memory
  • the aforementioned computer readable storage medium includes physical volatility and non-volatile, removable and non-volatile, implemented in any manner or technology for storing information such as computer readable instructions, data structures, program modules or other data. Indong medium.
  • the computer readable storage medium includes, but is not limited to, a USB flash drive, a mobile hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), and an erasable programmable read only.
  • EPROM electrically erasable programmable read only memory
  • EEPROM electrically erasable programmable read only memory
  • flash memory or other solid state memory technology
  • CD-ROM compact disc
  • DVD digital versatile disc
  • HD-DVD high definition DVD
  • Blu-ray Blu-ray or other optical storage

Landscapes

  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Chemical & Material Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • General Health & Medical Sciences (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Signal Processing (AREA)
  • Engineering & Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)

Abstract

本申请公开了一种基于人工智能的单板缺陷检测方法、系统及设备。通过获取经背面透过性照射的待检测单板图像，根据经过机器学习的缺陷检测模型对所述待检测单板图像进行识别和匹配，并根据所述识别和匹配的结果得到所述待检测单板的质量信息。解决了传统的机器视觉方法只能检测木皮的颜色或纹理缺陷，无法检测木皮的厚薄均匀度等问题。

Description

一种基于人工智能的单板缺陷检测方法、系统及设备 技术领域
本申请属于人工智能光学检测技术领域，具体涉及一种基于人工智能的单板缺陷检测方法、系统及设备。
背景技术
在木材加工领域，木板分成实木木板以及人造板。其中人造板中的胶合板和其它胶合层基材是通过多张木皮黏合而成的。一般优质单板用于胶合板、细木工板、模板、贴面板等人造板的面板，等级较低的单板用作背板和芯板。
本申请是针对木皮等单板加工的基于人工智能的自动化木皮质量检测和分类方法。
以木皮为例，由于木皮加工本身的限制，使得木皮的质量呈现出一定的随机性。其中最重要的一个特征在于加工后的木皮的厚薄不均匀，这是由于木材本身生长的随机性导致木材的硬度不均匀，在通过切刀的时候，硬度较高的部分产生的木皮厚度较大，质量较高。然而硬度较低的部分产生的木皮厚度较薄，质量较差。当一整块木皮的厚薄度严重不均匀时，该木皮就需要经过修补后才可以进入下一步工序，甚至整张变成等外材不适合加工。除此之外，木材本身也可能由于其他因素出现一些缺陷，例如虫眼、矿物线、色差等缺陷，这样的木皮不适合作为面板使用。
因此，对木皮质量的检测是木皮加工中重要的一个环节。然而，传统的机器视觉方法可以检测木皮的颜色或纹理缺陷，无法检测木皮的厚薄均匀度的问题。
发明内容
本申请实施例提供一种基于人工智能的单板缺陷检测技术方案，适用于木皮、竹皮等单板质量的自动化检测。
在一种可能的实施方式中，提供了
一种基于人工智能的单板缺陷检测方法，所述方法包括：
获取经背面透过性照射的待检测单板图像；
根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
根据所述识别和匹配的结果得到所述待检测单板的质量信息。
进一步地，所述背面透过性照射的辐射源为光源。
进一步地，所述缺陷检测模型利用机器学习获得，具体包括如下步骤：
获取经背面透过性照射的单板样本图像；
接收对所述单板样本图像的标注信息；
将标注后的图像样本输入到需进行机器学习的初始模型中；
根据所述单板样本图像和对应的所述标注信息进行训练，获得经过机器学习的缺陷检测模型。
进一步地，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
通过传送装置将单板样本传输到图像采集区，所述图像采集区与产生背面透过性照射的透过性照射系统的照射区处于同一区域。
进一步地，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
透过性照射系统通过光源，从单板样本的背面投射光线，该光源的光照强度通过控制，使得光线可以穿透单板样本，在置于单板正面的图像采集装置中呈现图像。
进一步地，该光照强度通过控制器来控制，使得在加工的单板的既定厚度下总是能够穿透单板样本；或者，
通过图像采集装置的输入或反馈自动调节光照强度，使得光照强度能够自适应不同的单板厚度；使得光线穿透单板之后所形成的图像能够反映该单板样本的厚度分布。
进一步地，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
利用正面照射光源正面照射单板样本；控制器控制正、反面的光照强度，使得透射光源发射的光线透过木皮后能够呈现更清晰的图像。
进一步地，所述标注信息包括样本图像数据、样本的厚度等级数据和背面光照强度数据中的一个或多个。
进一步地，所述接收对所述单板样本图像的标注信息的步骤进一步包括：
接收对单板厚度存在问题的区域和强度的标注信息；和/或，
接收对通过背面光照系统得到的图像中呈现的单板虫眼和/或矿物线缺陷的标注信息。
进一步地，所述待检测单板的质量信息的输出结果包括：直接输出不同质量等级的判断；或者，
输出不同质量等级的判断、并标注出厚度或厚度分布不均匀的区域。
进一步地，所述输出结果进一步包括：缺陷类型的识别，和/或单板的用途分类。
在另一种可能的实施方式中，提供了一种基于人工智能的单板缺陷检测系统，所述系统包括：
图像采集装置，用于获取单板的透过性照射图像；
透过性照射装置，所述透过性照射装置包括用于产生能够穿透单板的辐照的辐射源，并使得透过性辐照能够被图像采集装置获取；
以及，
质量检测装置，用于通过图像采集装置获取的图像，对单板的缺陷进行识别，并输出识别结果。
进一步地，所述识别结果包括：质量等级的判断、厚度或厚度分布不均匀的区域、缺陷类型的识别、单板的用途分类中的至少一个。
进一步地，所述辐射源为可见光光源，所述单板为木皮。
进一步地，所述光源为可调光源，使得光线穿透单板之后所形成的图像能够反映该单板样本的厚度分布。
进一步地，所述光源的光照强度通过控制器来控制，使得在加工单板的既定厚度下光线总是能够穿透单板样本；
或者，
所述光照强度通过图像采集装置的输入或反馈自动调节，使得光照强度能 够自适应不同的单板样本厚度。
进一步地，所述透过性照射装置还包括一个正面的光照模块，控制器控制正、反面光照强度，使得透射光源发射的光线透过单板后能够呈现更清晰的图像。
进一步地，所述质量检测装置还包括：标注模块，用于标注能够通过透过性照射装置的透过性辐照呈现在图像采集装置的图像样本中的单板特征。
进一步地，所述质量检测装置还包括：具有自动检测模型的自动检测模块，用于将标注模块标注后的图像样本输入到自动检测模型中。
进一步地，自动检测模型结合相应的属性对神经网络进行训练，所述相应的属性为预设的检测属性或自定义检测的属性。
进一步地，将光照强度作为一个单独的输入，与图像样本一起输入至所述神经网络，归一化该光照强度的影响。
在又一种可能的实施方式中，提供了一种基于人工智能的单板缺陷检测设备，包括：传送装置，用于携带单板通过图像采集区域；用于与远端服务器连接的通信模块；以及与检测设备连接的服务器；其特征在于，
所述检测设备能够执行上文所述的基于人工智能的单板缺陷检测方法；
或者，
所述检测设备还包括上文所述的基于人工智能的单板缺陷检测系统。
在又一种可能的实施方式中，提供了一种计算机可读介质，其中存储有多条指令，所述指令适用于由处理器加载并执行如上文所述的基于人工智能的单板缺陷检测方法。比如：
获取经背面透过性照射的待检测单板图像；
根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
根据所述识别和匹配的结果得到所述待检测单板的质量信息。
在又一种可能的实施方式中，提供了一种基于人工智能的单板缺陷检测系 统，其特征在于，所述系统包括：
存储器，用于存放指令；
处理器，用于执行所述存储器存储的指令，所述指令使得所述处理器执行如上文所述的基于人工智能的单板缺陷检测方法。比如：
获取经背面透过性照射的待检测单板图像；
根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
根据所述识别和匹配的结果得到所述待检测单板的质量信息。
本申请实施例提供的方法、系统及设备在满足单板厚度相关缺陷检测的前提下，还能够对一些缺陷类型进行识别，例如木皮中虫眼、矿物线等缺陷类型。相对于传统的缺陷检测方式，本申请提供的技术方案更加智能，大大降低了人力成本；同时，检测结果更加准确快速。
附图说明
为了更清楚地说明本申请实施例或现有技术中的技术方案，下面将对实施例或现有技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例。
图1为本申请实施例通过背部透过性照射系统得到木皮样本图像；
图2为依照本申请实施例使用正面投射光源得到的木皮样本图像；
图3为本申请一个实施例中带有木皮质量标注的示例图；
图4为本申请一个实施例中卷积神经网络示意图；
图5为本申请一个实施例中用于木皮质量的神经网络结构示意图；
图6为本申请又一实施例的木皮自动检测装置的示例的结构框图；
图7为实现和/或传播本申请技术方案的通用型计算机设备的一种示例的结构框图。
具体实施方式
为使得本申请的发明目的、特征、优点能够更加的明显和易懂，下面将结合本申请实施例中的附图，对本申请实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本申请一部分实施例，而非全部实施例。基于本申请中的实施例，本领域技术人员在没有做出创造性劳动前提下所获得的 所有其他实施例，都属于本申请保护的范围。
本领域技术人员可以理解，本申请中的“第一”、“第二”等术语仅用于区别不同设备、模块或参数等，既不代表任何特定技术含义，也不表示它们之间的必然逻辑顺序。
薄木，俗称“木皮”，是一种具有珍贵树种特色的木质片状薄型饰面或贴面材料。装饰薄木(木皮)的种类较多，目前，国内外还没有统一的分类方法。一般具有代表性的分类方法是按薄木的制造方法、形态、厚度及树种等来进行的。木皮质量的检测是木皮加工中重要的一个环节。传统的机器视觉方法可以检测木皮的颜色或纹理缺陷，但无法检测木皮的厚薄均匀度的问题。
在本申请的实施例中，提出了基于人工智能的设备并适用于木皮质量的自动化检测和分类。在一种实施方式中，提供了一种部署于检测工厂的设备，所述设备包括：传送装置、图像传感装置、透过性照射装置和木皮质量检测装置。图像传感装置包括用于获取传感图像的图像传感器，透过性照射装置包括用于产生能够穿透木皮的辐照的辐射源，并使得透过性辐照能够被图像传感器所捕捉。木皮质量检测装置通过图像传感装置捕捉到的图像信息，对木皮的厚度分布进行识别，并输出识别结果。
图1展示了本申请一个典型的应用方式：一个通过背部透过性照射系统得到木皮样本图像。所述辐射源优选为可见光光源，进一步地，所述辐射源优选为强度可调光源。其具体工作过程如下：首先，木皮被传送装置传递进入一个图像采集区域，图像采集区与透过性照射系统处于同一区域。透过性照射系统通过一个光源，例如由多个LED光源组成的平面光源，从木皮样本的背面投射光线，该光源的强度通过控制，使得光线可以穿透木皮样本，在置于木皮正面的传感器中呈现一个图像。该光照系统的光照强度可以通过一个控制器来控制，使得在加工木皮的既定厚度下总是能够穿透木皮样本。
在一种优选实施方式中，该控制系统可以通过图像采集设备的输入或反馈自动调节光照强度，使得光照强度能够自适应不同的木皮厚度。由于木皮厚度对光线穿透性的影响，使得光线穿透木皮之后所形成的图像能够反映该木皮样本的厚度分布。
其中，所述木皮样本的“背面”与“正面”是相对的概念，并非严格的方位限定；并且，透过性光源与图像传感器的位置优选可以互换。
进一步，该方案还可以包含一个用于与远端服务器连接的通信模块以及一 个与部署于检测工厂设备连接的服务器。
在一种优选的实施方式中，透过性照射系统还包括一个正面的光照系统，控制器控制正反面光照强度，使得透射光源发射的光线透过木皮后能够呈现一个较好的图像。
进一步，所述控制器可以使用一个预先配置的方式将光照强度优化并固定。同时也可以使用一个自适应的方式来调节背面、或背面和正面两侧光照强度。一种优选的方式是进行多强度扫描，同时，使用图像采集设备采集样本图像并输入分析器，分析器能够识别该光照强度下是否能够得到携带厚度分布信息的图像样本，如果能够则停止改变光照强度；如果不能满足条件，则继续改变光照条件。
注意，本申请中的透过性照射系统与其他基于机器视觉的用于补光的光照系统不同，本申请技术方案通过背面投射的方式，使得作为一个长度物理量测量的厚度检测，转化成一个图像识别的方法。因此，如果单纯使用正面光照的方法，仅能识别木板正面的纹理特征，任何机器视觉的方法均无法通过获取的图像识别木皮的厚度分布。图1中给出了一个示例光照系统下的木皮的图像样本。作为对比，图2中给出了一个只有正面光源投射下的该木皮的图像样本。从图中可以清晰见到，只有透过性照射系统下的图像能够反映木皮厚度分布信息，而正面投射光源无论光源强度多高，只能得到木皮的纹理和颜色特征，因此透过性照射系统是本系统的核心发明点之一。通过透过性照射系统，同样作为一个图像样本，使得该图像样本携带了木皮的厚度分布信息。此方法优选适用于木皮或类似单板的加工和/或检测领域，因为木皮或类似单板的厚度能够允许光线穿透，而其他较厚产品如木板则无法通过背面光照系统得到厚度信息。
在此，本领域技术人员应该能够理解，上述基于可见光源的透射检测方式仅为举例，现有或今后出现的其他情况，例如，基于红外光、紫外光、太赫兹辐射等方面的透射成像方式也应该包含在本申请的保护范围内，并以引用的形式包含于此。
基于透过性照射系统和图像采集设备，可以获得一个木皮样本的图像样本。为了使用以下描述的机器学习的方法，需要对样本进行标注。此处，标注是对一个图像样本以及该样本的厚度分布信息进行标注。一种基础的标注方法为，将该木皮厚度的等级进行标记，此处厚度的等级包含木皮的厚度以及厚度均匀度的信息。一种进阶的标注方式为，将厚度等级划分为多个等级，分别对应于 木皮用于后续加工的质量等级。例如一个标注如下：
示例一
[d][fn][S i]
其中，d为一个图像样本的数据，例如可以是原始图像数据，也可以是经过图像处理的图像样本数据。fn为厚度相关的等级。在一种优选的实施方式中，背部光照强度S i也可以作为一个信息标注维度，与图像样本进行关联；其中，字母n和i为自然数。厚度等级可以与后续加工相关联，因此该等级也可以被视作是一种质量等级，例如只有f5级的木皮才可以用于面板，而f5以下的木皮只能用于底板或芯板；再例如，可以将f0级别作为一种低质量的等级，厚度严重不均匀使得其无法用于任何胶合板的加工。
在一种优选的实施方式中，则可以对木皮厚度存在问题的区域和强度进行进一步标注。例如，对木皮厚度低于一个预定义阈值的区域或厚度分布不均匀的区域进行区域标注以及该区域的质量等级进行标注。
在另一种优选的实施方式中，则可以对木皮其他的缺陷进行标注，例如虫眼、矿物线等缺陷。这种标注不仅标注区域和质量等级，还可以标注缺陷的类型。也就是说这种标注可以记录更细分的有关木皮质量的信息，这些信息与上述厚度不均匀一样，可以通过背部光照系统得到的图像中得以呈现。
根据以上标注方式可以看出，通过对木皮图像进行更细致的标注，可以使得样本中携带的有关木皮质量的信息得以被标记。注意，其中能够被标记的特征必须是能够通过背部透射光照系统得以呈现在图像采集装置中的木皮特征，其他无法呈现的特征即使被标注也无法应用后续的木皮质量检测系统/模块所识别。图3中给了一种所述标注方法的示意图，其中f1区域是一个厚度等级的标注，意味着此处为一个f1等级的区域。此处f1不一定直接等于厚度，也可以包含厚度分布的信息，f1等级是一个较低的等级，意味着该区域质量较差。从图3中可以看到，该区域较为明亮，意味着厚度较薄，并且分布着不同的花纹，意味着厚度分布不均匀。图中z1区域可以是一个缺陷类型的标注，可以代表为一个矿物线或裂痕。
进一步，将上述标注后的图像样本输入到自动检测模型中，自动检测模型结合相应的属性对神经网络进行训练。
其中上述神经网络包括含有多个层、每个层包含多个节点、相邻两层多个节点之间存在可训练权重的神经网络。
图4中给出了本申请实施例的一个卷积神经网络的示意图，其中包括了多个卷积层和降采样层以及全连接层。卷积层是卷积神经网络的核心模块，通过与一个滤波器(filter)的卷积操作，将前一层的多个节点与下一层的节点相连。一般来说，卷积层的每一个节点只与前一层的部分节点相连。通过训练过程，其中使用初始值的滤波器可以根据训练数据不断改变自身的权重，进而生成最终的滤波器取值。降采样层可以使用最大池化(max-pooling)的方法将一组节点降维成一个节点，优选使用非线性取最大值的方法。在经过多个卷积层和降采样层后，一个全连接层最终用于产生检测的输出，全连接层将前一层的所有节点与后一层的所有节点相连，这与一个传统的神经网络类似。
在学习训练过程中，我们将木皮的样本数据作为输入，将其所在的自定义检测等属性作为输出，通过训练算法，例如梯度下降(gradient descent)算法使得神经网络中的滤波器权重值改变，进而使得输出与样本数据中的检测差异最小。随着使用的训练数据量的不断增大，不断改变的网络节点值不断改变并提高，神经网络的检测能力也就得到了提升。当训练结束后，得到一个训练好的神经网络，包括所设计的网络架构，例如图4中的层级设计以及层级之间连接方法，以及经过训练而改变的滤波器权重值。这些权重值被记录下来，并在后期的使用中被重复利用。
学习过程可以在本地检测系统中完成，也可以在云端完成。在一种实施方式中，图像采集装置采集木皮样本的图像数据以及标注后的数据集传送到云端服务器进行模型训练，服务器将训练后的模型传输到本地的处理器并完成部署。
在一种实施方式中，云端服务器可以使用多种来源的训练数据。例如来自多个本地图像采集并标注的数据，进而使得获得的数据量增大。
在检测状态，一个传送带携带一个木皮样本通过图像采集区域，图像采集区域通过透过性照射系统在木皮样本背部投射光源。一种优选的实施方式可以通过一个控制器对背部投射光源光照强度进行控制，使得图像采集装置能够获取到足够的透过性光线。将采集到的图像输入至训练后的神经网络中，可以得到一个用于判断木皮质量的输出。在另一种优选的实施方式中，光照强度可以与图像样本同时输入至训练后的神经网络。神经网络可以根据准确的光照强度对图像进行更加精准的分析。这是由于，光照强度能够改变透过木皮的光线的强度，进而影响图像的成像效果，不同的光照强度产生的图像可能导致神经网络的误判。例如增强光照强度后神经网络判断该木皮厚度较薄。因此，将光照 强度作为一个单独的输入，与图像样本一起输入至神经网络，则可以归一化该光照强度的影响，使得神经网络的判断更为精准。图5给出了一个对应的示意图。
关于检测输出的结果，可以有多种形式：
一种检测输出是，神经网络直接输出不同质量等级的判断，例如根据厚度及其分布情况分类得到的一个评级信息；
另外一种检测输出是，神经网络不仅可以给出木皮质量的等级判断，还可以进一步标注出厚度或厚度分布不均匀的区域；例如在木皮样本的图像中标出厚度不满足条件或厚度分布均匀度不满足条件的区域，神经网络可以进行识别并标记，甚至对该区域的质量评级以及厚度信息进行识别并标记；
又一种检测输出是，神经网络不仅可以给出木皮质量的等级判断以及厚度或厚度分布的标注，还可以对一些缺陷类型进行识别，例如木皮中虫眼、矿物线等缺陷类型进行标注。
再又一种检测输出是，神经网络可以给出木皮的用途分类。例如，该木皮可以适用于后期复合木板的哪种用途，例如可以用于木板的表面或背面或中间层。
在此，本领域技术人员应该能够理解，除了上述的神经网络，其他的机器学习方法也同样适用于包括木皮在内的单板缺陷检测。例如，随机森林、支持向量机、深度置信网络、K-means、K-neighboring等方法也应该包含在本申请的保护范围内，并以引用的形式包含于此。
在本申请的实施例中，还提出了一种基于人工智能的方法并适用于木皮质量的自动化检测和分类。
所述方法包括如下步骤：
S1、通过背部透过性照射获取木皮样本图像。所述辐射源优选为可见光光源，进一步地，所述辐射源优选为强度可调光源。
具体步骤包括：
S11、通过传送装置将木皮传输到图像采集区域，所述图像采集区与透过性照射系统处于同一区域；
S12、透过性照射系统通过光源，例如由多个LED光源组成的平面光源，从木皮样本的背面投射光线，该光源的强度通过控制，使得光线可以穿透木皮样本，在置于木皮正面的传感器中呈现一个图像。该光照强度可以通过一个控 制器来控制，使得在加工木皮的既定厚度下总是能够穿透木皮样本。
在一种优选实施方式中，可以通过图像采集设备的输入或反馈自动调节光照强度，使得光照强度能够自适应不同的木皮厚度。由于木皮厚度对光线穿透性的影响，使得光线穿透木皮之后所形成的图像能够反映该木皮样本的厚度分布。
其中，所述木皮样本的“背面”与“正面”是相对的概念，并非严格的方位限定；并且，透过性光源与图像传感器的位置优选可以互换。
在一种优选的实施方式中，还包括一个正面的光照步骤S13，
S13、正面照射木皮样本。控制器控制正、反面光照强度，使得透射光源发射的光线透过木皮后能够呈现一个较好的图像。
进一步，所述控制器可以使用一个预先配置的方式将光照强度优化并固定。同时也可以使用一个自适应的方式来调节背面、或背面和正面两侧光照强度。一种优选的方式是进行多强度扫描，同时，使用图像采集设备采集样本图像并输入分析器，分析器能够识别该光照强度下是否能够得到携带厚度分布信息的图像样本，如果能够则停止改变光照强度；如果不能满足条件，则继续改变光照条件。
S2、对获得的木皮样本图像进行标注。所述标注是对一个图像样本以及该样本的厚度分布信息进行标注。例如一个标注如下：
[d][fn][S i]
其中，d为一个图像样本的数据，例如可以是原始图像数据，也可以是经过图像处理的图像样本数据。fn为厚度相关的等级。背部光照强度S i为可选标注，在一种优选的实施方式中，S i也可以作为一个信息标注维度，与图像样本进行关联。
所述步骤S2还可以进一步包括如下步骤：
S21、对木皮厚度存在问题的区域和强度进行进一步标注。例如，对木皮厚度低于一个预定义阈值的区域或厚度分布不均匀的区域进行区域标注以及该区域的质量等级进行标注。
S22、对木皮其他的缺陷进行标注，例如虫眼、矿物线等缺陷。这种标注不仅标注区域和质量等级，还可以标注缺陷的类型。也就是说这种标注可以记录更细分的有关木皮质量的信息，这些信息与上述厚度不均匀一样，可以通过背部光照系统得到的图像中得以呈现。其中能够被标记的特征必须是能够通过 背部透射光照系统得以呈现在图像采集装置中的木皮特征，其他无法呈现的特征即使被标注也无法应用后续的木皮质量检测系统/模块所识别。
S3、将标注后的图像样本输入到自动检测模型(初始模型)中，自动检测模型结合相应的属性对神经网络进行训练。其中上述神经网络包括含有多个层、每个层包含多个节点、相邻两层多个节点之间存在可训练权重的神经网络。其中，相应的属性为预设的检测属性或自定义检测的属性。
具体步骤包括：
S31、生成最终的滤波器取值。卷积层通过与一个滤波器(filter)的卷积操作，将前一层的多个节点与下一层的节点相连。一般来说，卷积层的每一个节点只与前一层的部分节点相连。通过训练过程，其中使用初始值的滤波器可以根据训练数据不断改变自身的权重，进而生成最终的滤波器取值。
S32、节点降维。降采样层可以使用最大池化(max-pooling)的方法将一组节点降维成一个节点，优选使用非线性取最大值的方法。在经过多个卷积层和降采样层后，一个全连接层最终用于产生检测的输出，全连接层将前一层的所有节点与后一层的所有节点相连。
学习过程可以在本地检测系统中完成，也可以在云端完成。在一种实施方式中，图像采集装置采集木皮样本的图像数据以及标注后的数据集传送到云端服务器进行模型训练，服务器将训练后的模型传输到本地的处理器并完成部署。
在一种实施方式中，云端服务器可以使用多种来源的训练数据。例如来自多个本地图像采集并标注的数据，进而使得获得的数据量增大。
S4、将采集到的图像输入至训练后的自动检测模型的神经网络中，得到一个用于判断木皮质量的输出。
所述步骤S4还可以进一步包括如下步骤：
S41、光照强度与图像样本同时输入至训练后的神经网络。将光照强度作为一个单独的输入，与图像样本一起输入至神经网络，则可以归一化该光照强度的影响，使得神经网络的判断更为精准。神经网络根据准确的光照强度对图像进行更加精准的分析。这是由于，光照强度能够改变透过木皮的光线的强度，进而影响图像的成像效果，不同的光照强度产生的图像可能导致神经网络的误判。
S42、神经网络输出不同质量等级的判断。例如根据厚度及其分布情况分 类得到的一个评级信息；
S43、在木皮样本的图像中标注出厚度或厚度分布不均匀的区域。例如在木皮样本的图像中标出厚度不满足条件或厚度分布均匀度不满足条件的区域，神经网络可以进行识别并标记，甚至对该区域的质量评级以及厚度信息进行识别并标记；
S44、对缺陷类型进行识别。例如木皮中虫眼、矿物线等缺陷类型进行识别标注。
S45、神经网络给出木皮的用途分类。例如，该木皮可以适用于后期复合木板的哪种用途，例如可以用于木板的表面或背面或中间层。
本领域技术人员可以理解，在本申请具体实施方式的上述方法中，各步骤的序号大小并不意味着执行顺序的先后，各步骤的执行顺序应以其功能和内在逻辑确定，而不应对本申请具体实施方式的实施过程构成任何限定；“初始模型”包括但不限于未经训练的原始模型，也可以是其他批次或种类的单板数据训练的、但不能直接用于当前检测的模型，或其他任何可实现本发明对应功能或效果的检测模型。
此外，本申请实施例还提供了一种存储设备，例如，计算机可读介质，包括在被执行时进行以下操作的计算机可读指令：执行上述实施方式中的方法的各步骤的操作。
本申请实施例的木皮自动检测装置的又一种示例的结构，本申请具体实施例并不对木皮自动检测装置的具体实现做限定。如图6所示，该木皮自动检测装置100可以包括：
处理器(processor)110、通信接口(Communications Interface)120、存储器(memory)130、以及通信总线140。其中：
处理器110、通信接口120、以及存储器130通过通信总线140完成相互间的通信。
通信接口120，用于与比如客户端等的网元通信。
处理器110，用于执行程序132，具体可以执行上述方法实施例中的相关步骤。
具体地，程序132可以包括程序代码，所述程序代码包括计算机操作指令。
处理器110可能是一个中央处理器CPU，或者是特定集成电路ASIC(Application Specific Integrated Circuit)，或者是被配置成实施本申请实施例的 一个或多个集成电路。
存储器130，用于存放程序132。存储器130可能包含高速RAM存储器，也可能还包括非易失性存储器(non-volatile memory)，例如至少一个磁盘存储器。程序132具体可以用于使得所述木皮自动检测装置100执行以下步骤：
获取经背面透过性照射的木皮样本图像；
接收对所述木皮样本图像的标注信息；
将标注后的图像样本输入到需进行机器学习的初始模型中；根据所述木皮样本图像和对应的所述标注信息进行训练，获得经过机器学习的缺陷检测模型
获取经背面透过性照射的待检测木皮图像；
根据经过机器学习的缺陷检测模型对所述待检测木皮图像进行识别和匹配；
根据所述识别和匹配的结果得到所述待检测木皮的质量信息。
程序132中各步骤的具体实现可以参见上述实施例中的相应步骤和单元中对应的描述，在此不赘述。所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的设备和模块的具体工作过程，可以参考前述方法实施例中的对应过程描述，在此不再赘述。
尽管此处所述的主题是在结合操作系统和应用程序在计算机系统上的执行而执行的一般上下文中提供的，但本领域技术人员可以认识到，还可结合其他类型的程序模块来执行其他实现。一般而言，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、组件、数据结构和其他类型的结构。本领域技术人员可以理解，此处所述的本主题可以使用其他计算机系统配置来实践，包括手持式设备、多处理器系统、基于微处理器或可编程消费电子产品、小型计算机、大型计算机等，也可使用在其中任务由通过通信网络连接的远程处理设备执行的分布式计算环境中。在分布式计算环境中，程序模块可位于本地和远程存储器存储设备的两者中。
本领域普通技术人员可以意识到，结合本文中所公开的实施例描述的各示例的单元及方法步骤，能够以电子硬件、或者计算机软件和电子硬件的结合来实现。这些功能究竟以硬件还是软件方式来执行，取决于技术方案的特定应用和设计约束条件。专业技术人员可以对每个特定的应用来使用不同方法来实现所描述的功能，但是这种实现不应认为超出本申请的范围。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方 案本质上或者说对原有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例所述方法的全部或部分步骤。比如，典型地，本申请的技术方案可通过至少一个如图7所示的通用型计算机设备210来实现和/或传播。在图7中，通用型计算机设备210包括：计算机系统/服务器212、外接设备214和显示设备216；其中，所述计算机系统/服务器212包括处理单元220、I/O接口222、网络适配模块224和存储模块230，内部通常通过总线实现数据传输；进一步地，存储模块230通常由多种存储设备组成，比如，RAM(Random Access Memory，随机存储器)232、缓存234和存储系统(一般由一个或多个大容量非易失性存储介质组成)236等；实现本申请技术方案的部分或全部功能的程序240保存在存储模块230中，通常以多个程序模块242的形式存在。
而前述的计算机可读取存储介质包括以存储如计算机可读指令、数据结构、程序模块或其他数据等信息的任何方式或技术来实现的物理易失性和非易失性、可移动和不可因东介质。计算机可读取存储介质具体包括，但不限于，U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、可擦除可编程只读存储器(EPROM)、电可擦可编程只读存储器(EEPROM)、闪存或其他固态存储器技术、CD-ROM、数字多功能盘(DVD)、HD-DVD、蓝光(Blue-Ray)或其他光存储设备、磁带、磁盘存储或其他磁性存储设备、或能用于存储所需信息且可以由计算机访问的任何其他介质。
以上实施方式仅用于说明本发明，而并非对本发明的限制，有关技术领域的普通技术人员，在不脱离本发明的精神和范围的情况下，还可以做出各种变化和变型，因此所有等同的技术方案也属于本发明的范畴，本发明的专利保护范围应由权利要求限定。

Claims (24)

  1. 一种基于人工智能的单板缺陷检测方法，其特征在于，所述方法包括：
    获取经背面透过性照射的待检测单板图像；
    根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
    根据所述识别和匹配的结果得到所述待检测单板的质量信息。
  2. 如权利要求1所述的方法，其特征在于，所述背面透过性照射的辐射源为光源。
  3. 如权利要求1或2所述的方法，其特征在于，所述缺陷检测模型利用机器学习获得，具体包括如下步骤：
    获取经背面透过性照射的单板样本图像；
    接收对所述单板样本图像的标注信息；
    将标注后的图像样本输入到需进行机器学习的初始模型中；
    根据所述单板样本图像和对应的所述标注信息进行训练，获得经过机器学习的缺陷检测模型。
  4. 如权利要求3所述的方法，其特征在于，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
    通过传送装置将单板样本传输到图像采集区，所述图像采集区与产生背面透过性照射的透过性照射系统的照射区处于同一区域。
  5. 如权利要求4所述的方法，其特征在于，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
    透过性照射系统通过光源，从单板样本的背面投射光线，该光源的光照强度通过控制，使得光线可以穿透单板样本，在置于单板正面的图像采集装置中呈现图像。
  6. 如权利要求5所述的方法，其特征在于，该光照强度通过控制器来控制，使得在加工的单板的既定厚度下总是能够穿透单板样本；或者，
    通过图像采集装置的输入或反馈自动调节光照强度，使得光照强度能够自适应不同的单板厚度；使得光线穿透单板之后所形成的图像能够反映该单板样本的厚度分布。
  7. 如权利要求6所述的方法，其特征在于，所述获取经背面透过性照射的单板样本图像的步骤进一步包括：
    利用正面照射光源正面照射单板样本；控制器控制正、反面的光照强度，使得透射光源发射的光线透过木皮后能够呈现更清晰的图像。
  8. 如权利要求3所述的方法，其特征在于，所述标注信息包括样本图像数据、样本的厚度等级数据、样本的质量等级数据和背面光照强度数据中的一个或多个。
  9. 如权利要求8所述的方法，其特征在于，所述接收对所述单板样本图像的标注信息的步骤进一步包括：
    接收对单板厚度存在问题的区域和强度的标注信息；和/或，
    接收对通过背面光照系统得到的图像中呈现的单板虫眼和/或矿物线缺陷的标注信息。
  10. 如权利要求1-9任一项所述的方法，其特征在于，所述待检测单板的质量信息的输出结果包括：直接输出不同质量等级的判断；或者，
    输出不同质量等级的判断、并标注出厚度或厚度分布不均匀的区域。
  11. 如权利要求10所述的方法，其特征在于，所述输出结果进一步包括：缺陷类型的识别，和/或单板的用途分类。
  12. 一种基于人工智能的单板缺陷检测系统，其特征在于，所述系统包括：
    图像采集装置，用于获取单板的透过性照射图像；
    透过性照射装置，所述透过性照射装置包括用于产生能够穿透单板的辐照的辐射源，并使得透过性辐照能够被图像采集装置获取；
    以及，
    质量检测装置，用于通过图像采集装置获取的图像，对单板的缺陷进行识别，并输出识别结果。
  13. 如权利要求12所述的检测系统，其特征在于，所述识别结果包括：质量等级的判断、厚度或厚度分布不均匀的区域、缺陷类型的识别、单板的用途分类中的至少一个。
  14. 如权利要求12所述的检测系统，其特征在于，所述辐射源为可见光光 源。
  15. 如权利要求14所述的检测系统，其特征在于，所述光源为可调光源，使得光线穿透单板之后所形成的图像能够反映该单板样本的厚度分布。
  16. 如权利要求14或15所述的检测系统，其特征在于，所述光源的光照强度通过控制器来控制，使得在加工单板的既定厚度下光线总是能够穿透单板样本；
    或者，
    所述光照强度通过图像采集装置的输入或反馈自动调节，使得光照强度能够自适应不同的单板样本厚度。
  17. 如权利要求14或15所述的检测系统，其特征在于，所述透过性照射装置还包括一个正面的光照模块，控制器控制正、反面光照强度，使得透射光源发射的光线透过单板后能够呈现更清晰的图像。
  18. 如权利要求12-15任一项所述的检测系统，其特征在于，所述质量检测装置还包括：标注模块，用于标注能够通过透过性照射装置的透过性辐照呈现在图像采集装置的图像样本中的单板特征。
  19. 如权利要求18所述的检测系统，其特征在于，所述质量检测装置还包括：具有自动检测模型的缺陷检测模块，用于将标注模块标注后的图像样本输入到自动检测模型中。
  20. 如权利要求19所述的检测系统，其特征在于，自动检测模型结合相应的属性对神经网络进行训练，所述相应的属性为预设的检测属性或自定义检测的属性。
  21. 如权利要求20所述的检测系统，其特征在于，将光照强度作为一个单独的输入，与图像样本一起输入至所述神经网络，归一化该光照强度的影响。
  22. 一种基于人工智能的单板缺陷检测设备，包括：传送装置，用于携带单板通过图像采集区域；用于与远端服务器连接的通信模块；以及与检测设备连接的服务器；其特征在于，
    所述检测设备能够执行权利要求1-12任一项所述的基于人工智能的单板缺陷检测方法；
    或者，
    所述检测设备还包括权利要求13-21任一项所述的基于人工智能的单板缺陷检测系统。
  23. 一种计算机可读介质，其中存储有多条指令，所述指令适用于由处理器加载并执行：
    获取经背面透过性照射的待检测单板图像；
    根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
    根据所述识别和匹配的结果得到所述待检测单板的质量信息。
  24. 一种基于人工智能的单板缺陷检测系统，其特征在于，所述系统包括：
    存储器，用于存放指令；
    处理器，用于执行所述存储器存储的指令，所述指令使得所述处理器执行以下步骤：
    获取经背面透过性照射的待检测单板图像；
    根据缺陷检测模型对所述待检测单板图像进行识别和匹配；
    根据所述识别和匹配的结果得到所述待检测单板的质量信息。
PCT/CN2018/108551 2017-12-14 2018-09-29 一种基于人工智能的单板缺陷检测方法、系统及设备 WO2019114372A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201711335726.0A CN108362702A (zh) 2017-12-14 2017-12-14 一种基于人工智能的单板缺陷检测方法、系统及设备
CN201711335726.0 2017-12-14

Publications (1)

Publication Number Publication Date
WO2019114372A1 true WO2019114372A1 (zh) 2019-06-20

Family

ID=63009982

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/108551 WO2019114372A1 (zh) 2017-12-14 2018-09-29 一种基于人工智能的单板缺陷检测方法、系统及设备

Country Status (2)

Country Link
CN (1) CN108362702A (zh)
WO (1) WO2019114372A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114002225A (zh) * 2021-10-19 2022-02-01 业成科技（成都）有限公司 光学检测系统与方法

Families Citing this family (9)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108362702A (zh) * 2017-12-14 2018-08-03 北京木业邦科技有限公司 一种基于人工智能的单板缺陷检测方法、系统及设备
CN109523505B (zh) * 2018-09-18 2021-06-11 深圳市智信精密仪器有限公司 一种基于机器视觉的陶瓷砖表面花纹缺陷检测方法
CN109557104A (zh) * 2018-11-19 2019-04-02 临沂众为智能科技有限公司 一种基于深度学习的木料优选锯智能检测方法
KR102613277B1 (ko) * 2018-11-30 2023-12-12 제이에프이 스틸 가부시키가이샤 표면 결함 검출 방법, 표면 결함 검출 장치, 강재의 제조 방법, 강재의 품질 관리 방법, 강재의 제조 설비, 표면 결함 판정 모델의 생성 방법 및, 표면 결함 판정 모델
CN110059765B (zh) * 2019-04-26 2021-02-02 中国地质大学（北京） 一种矿物智能识别分类系统与方法
CN110084812B (zh) * 2019-05-10 2023-02-10 广东工业大学 一种太赫兹图像缺陷检测方法、装置、系统及存储介质
CN110458814A (zh) * 2019-07-29 2019-11-15 山东艾雷维特智能科技有限公司 一种木材单板缺陷的检测方法、装置及系统
US20220161454A1 (en) * 2020-11-20 2022-05-26 Boise Cascade Company Method and system for layered wood product production
CN115619787B (zh) * 2022-12-19 2023-03-31 苏州赫芯科技有限公司 一种uv胶缺陷检测方法、系统、设备及介质

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2514788A1 (en) * 2004-08-06 2006-02-06 John Laurent System and method for the detection of bluestain and rot on wood
CN101669023A (zh) * 2007-04-20 2010-03-10 株式会社名南制作所 木材的检测方法及装置及程序
JP2014215233A (ja) * 2013-04-26 2014-11-17 橋本電機工業株式会社 単板の選別堆積装置及び単板選別方法
CN108362702A (zh) * 2017-12-14 2018-08-03 北京木业邦科技有限公司 一种基于人工智能的单板缺陷检测方法、系统及设备

Family Cites Families (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE9502611D0 (sv) * 1995-07-14 1995-07-14 Casco Nobel Ab Prediction of the properties of board
US5960104A (en) * 1996-08-16 1999-09-28 Virginia Polytechnic & State University Defect detection system for lumber
CN101778203B (zh) * 2002-12-27 2012-06-06 株式会社尼康 图像处理装置
CN101224590B (zh) * 2008-02-01 2012-11-21 中国林业科学研究院木材工业研究所 一种人造板单板单元及其制备方法
JP5520006B2 (ja) * 2009-10-29 2014-06-11 パナソニック株式会社 木材欠陥検出装置およびその方法
JP5825278B2 (ja) * 2013-02-21 2015-12-02 オムロン株式会社 欠陥検査装置および欠陥検査方法
EP3063495A4 (en) * 2013-10-31 2017-08-09 3M Innovative Properties Company Multiscale uniformity analysis of a material
CN104034735A (zh) * 2014-02-28 2014-09-10 金东纸业（江苏）股份有限公司 一种多层物料的检测设备
CN205538740U (zh) * 2016-01-29 2016-08-31 广州番禺职业技术学院 一种智能木材表面缺陷检测系统
CN106017340A (zh) * 2016-07-06 2016-10-12 北京大恒图像视觉有限公司 一种基于机器视觉的透光容器壁厚检测装置及方法
CN106310955B (zh) * 2016-09-28 2019-05-17 天津理工大学 一种滤膜厚度的检测方法及其检测装置
CN106683093B (zh) * 2017-01-12 2019-11-29 国家林业和草原局北京林业机械研究所 板材外观质量综合量化评价方法
CN106875381B (zh) * 2017-01-17 2020-04-28 同济大学 一种基于深度学习的手机外壳缺陷检测方法
CN107437094B (zh) * 2017-07-12 2021-04-20 北京木业邦科技有限公司 基于机器学习的木板分拣方法及系统
CN107392896B (zh) * 2017-07-14 2019-11-08 佛山市南海区广工大数控装备协同创新研究院 一种基于深度学习的木材缺陷检测方法和系统

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CA2514788A1 (en) * 2004-08-06 2006-02-06 John Laurent System and method for the detection of bluestain and rot on wood
CN101669023A (zh) * 2007-04-20 2010-03-10 株式会社名南制作所 木材的检测方法及装置及程序
JP2014215233A (ja) * 2013-04-26 2014-11-17 橋本電機工業株式会社 単板の選別堆積装置及び単板選別方法
CN108362702A (zh) * 2017-12-14 2018-08-03 北京木业邦科技有限公司 一种基于人工智能的单板缺陷检测方法、系统及设备

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
CHENG, WEI ET AL.: "Design of the Veneer Automatic Classification System based on computer vision", CHINA WOOD INDUSTRY, vol. 21, no. 3, 30 May 2007 (2007-05-30), pages 25 *

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114002225A (zh) * 2021-10-19 2022-02-01 业成科技（成都）有限公司 光学检测系统与方法
CN114002225B (zh) * 2021-10-19 2023-05-12 业成科技（成都）有限公司 光学检测系统与方法

Also Published As

Publication number Publication date
CN108362702A (zh) 2018-08-03

Similar Documents

Publication Publication Date Title
WO2019114372A1 (zh) 一种基于人工智能的单板缺陷检测方法、系统及设备
US10942933B2 (en) Computerized technical authentication and grading system for collectible objects
CN108247764B (zh) 基于机器学习的木板切割方法、装置、电子设备和介质
AU2011257540B2 (en) Device for testing the authenticity of valuable documents
CN109540831B (zh) 基于高光谱成像技术的枸杞子品种识别方法
CN207636508U (zh) 一种单板缺陷检测系统及设备
JP2006322774A (ja) 木材の節探査方法及び装置及びプログラム
CN110705655A (zh) 一种基于光谱和机器视觉耦合的烟叶分类方法
CN106663201A (zh) 一种手掌检测方法和设备
CN105241400A (zh) 一种玻璃平整度检测的方法及其装置
Bhardwaj et al. Diabetic retinopathy severity grading employing quadrant‐based Inception‐V3 convolution neural network architecture
JP2007147442A (ja) 木材の検査方法及び装置及びプログラム
CN112634237A (zh) 一种基于YOLOv3改进网络的长竹条表面缺陷检测方法及系统
CN108362703A (zh) 一种基于人工智能的单板检测方法和检测设备
WO2016102854A1 (fr) Procédé et système d'acquisition et d'analyse d'empreintes avec détection de fraude
JP6776672B2 (ja) 品質評価方法及び品質評価装置
CN1426016A (zh) 证件真伪识别装置及真伪识别方法
CN108388858B (zh) 虹膜防伪方法与装置
EP3726423B1 (fr) Dispositif et procédé d'acquisition et de traitement biométrique
JPH0534281A (ja) メロンの外観評価装置
CN205156882U (zh) 一种玻璃平整度检测的装置
WO2023098187A1 (zh) 一种处理方法及处理装置、处理系统
US8987676B2 (en) System and method for the detection of soiling in bank notes
US9047723B2 (en) Defect categorization
CN113887543B (zh) 一种基于高光谱特征的箱包鉴伪方法与光谱采集装置

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 18888505

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205N DATED 04/08/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18888505

Country of ref document: EP

Kind code of ref document: A1