CN113362276B - Visual detection method and system for plates - Google Patents
Visual detection method and system for plates Download PDFInfo
- Publication number
- CN113362276B CN113362276B CN202110454981.7A CN202110454981A CN113362276B CN 113362276 B CN113362276 B CN 113362276B CN 202110454981 A CN202110454981 A CN 202110454981A CN 113362276 B CN113362276 B CN 113362276B
- Authority
- CN
- China
- Prior art keywords
- defect
- color image
- dimensional
- color
- model
- Prior art date
- Legal status (The legal status 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 status listed.)
- Active
Links
- 230000000007 visual effect Effects 0.000 title claims abstract description 18
- 238000001514 detection method Methods 0.000 title abstract description 40
- 230000007547 defect Effects 0.000 claims abstract description 216
- 238000000034 method Methods 0.000 claims abstract description 43
- 238000013145 classification model Methods 0.000 claims abstract description 36
- 238000012549 training Methods 0.000 claims description 47
- 230000003321 amplification Effects 0.000 claims description 22
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 22
- 238000012545 processing Methods 0.000 claims description 20
- 238000002372 labelling Methods 0.000 claims description 18
- 238000007781 pre-processing Methods 0.000 claims description 18
- 230000035929 gnawing Effects 0.000 claims description 17
- 239000000463 material Substances 0.000 claims description 17
- 238000011179 visual inspection Methods 0.000 claims description 17
- 238000001914 filtration Methods 0.000 claims description 16
- 238000000354 decomposition reaction Methods 0.000 claims description 12
- 238000009499 grossing Methods 0.000 claims description 10
- 239000011347 resin Substances 0.000 claims description 10
- 229920005989 resin Polymers 0.000 claims description 10
- 238000005520 cutting process Methods 0.000 claims description 9
- 238000010586 diagram Methods 0.000 claims description 8
- 238000009826 distribution Methods 0.000 claims description 8
- 230000008569 process Effects 0.000 claims description 8
- 230000010354 integration Effects 0.000 claims description 6
- 238000000605 extraction Methods 0.000 claims description 5
- 230000007246 mechanism Effects 0.000 claims description 4
- 239000013598 vector Substances 0.000 claims description 4
- 238000007689 inspection Methods 0.000 claims description 2
- 238000005452 bending Methods 0.000 description 21
- 238000005516 engineering process Methods 0.000 description 12
- 241000238631 Hexapoda Species 0.000 description 8
- 230000006872 improvement Effects 0.000 description 8
- 238000005259 measurement Methods 0.000 description 7
- 238000003384 imaging method Methods 0.000 description 5
- 238000004519 manufacturing process Methods 0.000 description 5
- 239000002023 wood Substances 0.000 description 5
- 238000004364 calculation method Methods 0.000 description 4
- 230000002950 deficient Effects 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000009643 growth defect Effects 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 238000011410 subtraction method Methods 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 3
- 239000007787 solid Substances 0.000 description 3
- 241000894006 Bacteria Species 0.000 description 2
- 230000009471 action Effects 0.000 description 2
- 230000007812 deficiency Effects 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 238000007726 management method Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005498 polishing Methods 0.000 description 2
- 238000003672 processing method Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008030 elimination Effects 0.000 description 1
- 238000003379 elimination reaction Methods 0.000 description 1
- 238000003860 storage Methods 0.000 description 1
- 238000010408 sweeping Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/30—Subject of image; Context of image processing
- G06T2207/30108—Industrial image inspection
- G06T2207/30161—Wood; Lumber
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Evolutionary Biology (AREA)
- Quality & Reliability (AREA)
- Investigating Materials By The Use Of Optical Means Adapted For Particular Applications (AREA)
- Length Measuring Devices By Optical Means (AREA)
Abstract
The invention discloses a visual detection method for a plate, which comprises the following steps: acquiring three-dimensional point cloud information of a target plate acquired by a 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the size defect of the target plate according to the three-dimensional model and a preset reference size; the method comprises the steps of acquiring two-dimensional color image information of a target plate acquired by a linear array camera, inputting the two-dimensional color image information into a pre-trained defect identification model to identify natural defects of the target plate, and inputting the two-dimensional color image information into a pre-trained color classification model to identify color categories of the target plate. The invention also discloses a plate visual detection system. By adopting the invention, the defects of the plate can be comprehensively and accurately detected.
Description
Technical Field
The invention relates to the technical field of solid wood detection, in particular to a plate visual detection method and a plate visual detection system.
Background
At present, the actual wood industry board picking process still adopts a manual detection mode, the whole production line needs to carry boards manually and detect the boards with naked eyes, the operation intensity of workers is generally high, and the board picking process needs to consume a large amount of human resources; as is well known, the judgment of the quality standard by workers is subjective, and the defects of randomness, large error and the like exist, so that the quality standard of the wood pick board cannot be ensured; in addition, in the whole solid wood production process, a plurality of working procedures are needed for manual detection, the detection efficiency of the whole mass of the produced solid wood is low, and the cost is high; in addition, the plate detection has higher requirements on the experience of workers, the talents with related skilled experiences can rapidly detect the defect points of the plate, the plate is optimized and processed, and the culture period of the skilled workers is too long. Therefore, manual detection is already the most original detection mode for imminent elimination.
Accordingly, manufacturers are currently beginning to strive on automated measurement equipment, image preprocessing, algorithms to try to solve the problems of floor measurement and quality inspection. At present, some enterprises begin to adopt image measurement modes to identify defect detection, but the problems of high time delay, poor performance and the like generally exist.
Disclosure of Invention
The invention aims to solve the technical problem of providing a plate visual detection method and a plate visual detection system, which can be used for comprehensively and accurately detecting defects of plates.
In order to solve the technical problems, the invention provides a plate visual inspection method, which comprises the following steps: acquiring three-dimensional point cloud information of a target plate acquired by a 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the size defect of the target plate according to the three-dimensional model and a preset reference size; the method comprises the steps of acquiring two-dimensional color image information of a target plate acquired by a linear array camera, inputting the two-dimensional color image information into a pre-trained defect identification model to identify natural defects of the target plate, and inputting the two-dimensional color image information into a pre-trained color classification model to identify color categories of the target plate.
As an improvement of the above scheme, the board visual inspection method further comprises defect identification model training, specifically comprising: acquiring an original color image; intercepting and combining the original color image to form a basic color image; marking defect positions and defect categories in the basic color image to form a marked color image, wherein the defect categories comprise dead joints, resin bags, skin clamping and decay; performing generalization amplification treatment on the marked color image to form a sample color image; integrating all sample color images into a sample color image set; training the defect recognition model according to the sample color image set.
As an improvement of the scheme, the defect identification model adopts a DNN model to extract multi-scale features of an input sample color image, combines semantic information and position information to fuse the multi-scale features, and combines an attention mechanism and a spatial pyramid structure to extract features so as to identify and output defect types, defect positions and confidence information.
As an improvement of the above scheme, the board visual detection method further comprises training of a color classification model, and specifically comprises the following steps: acquiring an original color image; extracting a target foreground region of the original color image through a visual algorithm to form an RO I color image; scaling the target foreground region of the RO I color image and cutting the target foreground region into a plurality of partial images; marking color categories in the partial image to form a marked classified image, the color categories including blue, dark, medium, and light; performing generalization amplification treatment on the labeling classified images to form sample classified images; integrating all the sample classification images into a sample classification image set; training the color classification model according to the sample classification image set.
As an improvement of the scheme, the color classification model adopts a DNN model to extract multi-scale features of an input sample classification image, classifies high-dimensional feature vectors, and then recognizes semantic classification information of the image to output color category and confidence information.
As an improvement of the above-mentioned scheme, before inputting the two-dimensional color image information into a defect recognition model and a color classification model trained in advance, preprocessing the two-dimensional color image information, the preprocessing step includes: SVD decomposition is carried out on the two-dimensional color image information; gabor filtering is carried out on the two-dimensional color image information after SVD decomposition; and performing difference operation on the two-dimensional color image information subjected to Gabor filtering by adopting an iterative difference method.
As an improvement of the scheme, the size defect comprises any one or more of a length-width-height defect, a bending defect, a torsion defect, a two-end straight angle defect, a broken surface gnawing head defect, a tile-shaped defect and a insect growth defect; when the length, width and height defects are identified, the length of the target plate is the length of the three-dimensional model, the width of the target plate is the width of the three-dimensional model, and the thickness of the target plate is the thickness of the three-dimensional model; when the curvature defect is identified, extracting height distribution along the length direction of the three-dimensional model, selecting a maximum height value as chord height, and dividing the chord height by the length to obtain curvature in the length direction; when the torsion degree defect is identified, calculating the distance from any one corner to the other three corner planes on one surface in the three-dimensional model, and taking the maximum distance value as the torsion degree; when the two-end straight angle defects are identified, calculating an included angle between planes in the three-dimensional model, and taking the included angle as the two-end straight angle; when the broken surface gnawing head defect is identified, a surface smoothing algorithm is adopted to process the three-dimensional model, the smoothed data and the original data are subtracted to obtain a broken surface gnawing head region, or an upper surface equation is fitted through upper surface three-dimensional point cloud information, a lower surface equation is fitted through lower surface three-dimensional point cloud information, the distance from each point to the upper surface equation and the lower surface equation is calculated, and a gray level diagram is drawn to obtain the broken surface gnawing head region; when the tile-shaped defect is identified, fitting a circular equation from the inner side and the outer side of the three-dimensional model, calculating the radius of a circle, and determining the tile-shaped defect according to the radius; and when the insect-living defect is identified, processing the three-dimensional model by adopting a surface smoothing algorithm, and subtracting the smoothed data from the original data to obtain an insect-living region.
Correspondingly, the invention also provides a plate visual detection system, which comprises: the dimensional defect identification module is used for acquiring three-dimensional point cloud information of the target plate acquired by the 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the dimensional defect of the target plate according to the three-dimensional model and a preset reference size; the natural defect identification module is used for acquiring the two-dimensional color image information of the target plate acquired by the linear array camera, and inputting the two-dimensional color image information into a defect identification model trained in advance to identify the natural defect of the target plate; the color classification and identification module is used for acquiring the two-dimensional color image information of the target plate acquired by the linear array camera, and inputting the two-dimensional color image information into a color classification model trained in advance to identify the color class of the target plate.
As an improvement of the above solution, the board visual inspection system further includes a defect recognition training module, and the defect recognition training module includes: a first acquisition unit configured to acquire an original color image; the merging unit is used for intercepting and merging the original color images to form a basic color image; a first marking unit for marking defect positions and defect categories in the basic color image to form a marked color image, wherein the defect categories comprise dead joints, resin bags, skin clamping and decay; the first amplification unit is used for performing generalization amplification treatment on the marked color image so as to form a sample color image; a first integration unit for integrating all the sample color images into a sample color image set; and the first training unit is used for training the defect recognition model according to the sample color image set.
As an improvement of the above scheme, the board visual detection system further comprises a color classification training module, and the color classification training module comprises: a second acquisition unit configured to acquire an original color image; an extracting unit for extracting a target foreground region of the original color image through a visual algorithm to form an ROI color image; a cutting unit for performing scaling processing on the target foreground region of the RO I color image and cutting the target foreground region into a plurality of partial images; the second labeling unit is used for labeling color categories in the local image to form a labeling classified image, wherein the color categories comprise blue, dark, medium and light; the second amplification unit is used for carrying out generalization amplification treatment on the labeling classified images so as to form sample classified images; the second integration unit is used for integrating all the sample classification images into a sample classification image set; and the second training unit is used for training the color classification model according to the sample classification image set.
The implementation of the invention has the following beneficial effects:
According to the invention, a three-dimensional laser acquisition technology is adopted, and the 3D profiler is used for acquiring the three-dimensional point cloud information of the plate, so that the accuracy is high; meanwhile, the invention also adopts a three-dimensional imaging technology, the acquired three-dimensional point cloud information is integrated and displayed to form a three-dimensional model, and the dimensional defect of the target plate is identified by analyzing the three-dimensional model; the invention also adopts a defect detection technology based on deep learning, and the granularity and defect definition of defect detection are researched, and the method is applied to training of an identification model, so that the model is continuously optimized to improve the identification accuracy;
furthermore, the invention also adopts a visual image acquisition preprocessing technology to preprocess the images, and obtains all the plate surface images as clearly as possible.
Drawings
FIG. 1 is a flow chart of an embodiment of a visual inspection method for a sheet material of the present invention;
FIG. 2 is a schematic view of the curvature of a sheet material in accordance with the present invention;
FIG. 3 is a schematic illustration of the degree of torsion of the sheet material of the present invention;
FIG. 4 is a schematic view of the two ends of the sheet material of the present invention;
FIG. 5 is a schematic tile-shaped view of the sheet of the present invention;
FIG. 6 is a flow chart of an embodiment of a method of training a defect recognition model in accordance with the present invention;
FIG. 7 is a schematic diagram of a training method of a defect recognition model in the present invention;
FIG. 8 is a schematic diagram of a defect recognition model according to the present invention;
FIG. 9 is a flow chart of an embodiment of a training method of a color classification model in the present invention;
FIG. 10 is a schematic diagram of a training method of a color classification model in the present invention;
FIG. 11 is a schematic view of the structure of the color classification model according to the present invention;
FIG. 12 is a schematic diagram of the visual inspection system for sheet material of the present invention;
FIG. 13 is a schematic diagram of a defect recognition training module according to the present invention;
FIG. 14 is a schematic diagram of a color class training module according to the present invention;
fig. 15 is a schematic view showing the structure of a pretreatment module in the present invention.
Detailed Description
The present invention will be described in further detail with reference to the accompanying drawings, for the purpose of making the objects, technical solutions and advantages of the present invention more apparent.
Referring to fig. 1, fig. 1 shows a flowchart of an embodiment of a visual inspection method of a sheet material of the present invention, which includes:
s101, acquiring three-dimensional point cloud information of a target plate acquired by a 3D profiler.
In the invention, four 3D profilers are adopted to measure the heights of the plate in the upper, lower, left and right directions, so when the plate to be detected enters a detection area formed by the four 3D profilers from a motion device, four laser lines of the four 3D profilers scan four planes of the plate respectively, and three-dimensional point cloud information of the plate is obtained.
S102, constructing a three-dimensional model according to the three-dimensional point cloud information.
The acquired three-dimensional point cloud information can be integrated and displayed through a three-dimensional imaging technology to form a three-dimensional model.
And S103, identifying the size defect of the target plate according to the three-dimensional model and the preset reference size.
The actual size of the plate can be obtained through the three-dimensional model, and then the actual size is compared with the reference size, so that the size defect of the plate can be identified.
The working procedure of the 3D profiler is known from the comprehensive steps S101-S103 as follows:
(1) And (3) starting acquisition: opening the 3D profiler when the sheet material is in the field of view of the 3D profiler;
(2) Sweeping the plate: the 3D profiler scans the rapidly moving plate, and simultaneously sends scanned three-dimensional point cloud information to a computer;
(3) Stopping acquisition: stopping the 3D profiler when the plate passes through the field of view of the 3D profiler, and splicing and reconstructing the acquired three-dimensional point cloud information of the whole plate;
(4) Whether or not it is qualified: and (3) comparing the calculation structure with a judgment standard according to the detection procedure of each parameter, and identifying the specific defect of the related parameter.
Specifically, the size defect includes any one or more of a length-width-height defect, a curvature defect, a torsion defect, a two-end straight angle defect, a face collapse and head gnawing defect, a tile defect, a insect growth defect and a bending measurement defect, but is not limited thereto.
The following describes various dimensional defects in detail in connection with specific identification methods:
(1) When the length, width and height defects are identified, the length of the target plate is the length of the three-dimensional model, the width of the target plate is the width of the three-dimensional model, and the thickness of the target plate is the thickness of the three-dimensional model.
The length of the sheet can be obtained by the length of the three-dimensional model. For example, the length of the test board is 910mm, and the transmission speed is 1m/s, and the accuracy can be within 0.5mm by adopting a 2kHz 3D profiler.
The width of the plate is measured by the X-axis direction laser line width of a 3D profiler arranged up and down, and the width precision is within 0.051-0.097 mm.
The thickness of the plate is measured by the X-axis direction laser linewidth of the 3D profiler arranged left and right, and the precision and the width precision are the same.
(2) When the curvature defect is identified, extracting height distribution along the length direction of the three-dimensional model, selecting the maximum height value as chord height, and dividing the chord height by the length to obtain the curvature in the length direction.
As shown in fig. 2, the plate is in the shape of a slender cuboid, and according to the definition of curvature: the chord height of the bending per meter length is the bending per meter, and the ratio of the total chord height of the total length bending to the total length is the total bending. Therefore, the height distribution in the length direction in the three-dimensional model can be converted, the chord height of the maximum value is selected, and the value is divided by the total length to obtain the curvature in the length direction; the maximum of the two surfaces is the curvature due to the upper and lower surfaces.
(3) When the torsion degree defect is identified, the distance from any one corner to the other three corner planes on one surface in the three-dimensional model is calculated, and the maximum distance value is taken as the torsion degree.
As shown in fig. 3, the plate is in the shape of a slender cuboid, and according to the definition of torsion resistance: one corner (corner A) of a plane arch is upwards, and the other three corners (corner B, C, D) are arranged on one plane, wherein the distance between the suspended corner and the other three-point plane is the torsion degree. Therefore, the method can be converted into calculation of the distance from any angle of one surface to the plane of the other triangle, and the maximum distance is taken as the curvature; because of the upper and lower sides, the maximum value is taken.
(4) When the defects of the straight angles at the two ends are identified, calculating the included angle between planes in the three-dimensional model, and taking the included angle as the straight angles at the two ends.
As shown in fig. 4, the straight angles at two ends can be considered as the angle measurement between two ends and the side, and can be converted into the included angle (angle E) between planes according to the three-dimensional model to be solved.
(5) When the broken surface gnawing defects are identified, a three-dimensional model is processed by adopting a surface smoothing algorithm, the broken surface gnawing head area can be obtained by subtracting the smoothed data from the original data, or the broken surface gnawing head area can be obtained by fitting an upper surface equation through upper surface three-dimensional point cloud information and fitting a lower surface equation through lower surface three-dimensional point cloud information, the distance between each point and the upper surface equation and the lower surface equation is calculated, and a gray level graph is drawn.
The surface collapse is mainly caused by that a cutter does not fall into or the height of the plate surface is slightly lower, and is characterized in that obvious burrs are formed on the surface of the plate, so that an obvious dark area is formed at the waist of the plate; therefore, the area of the broken surface can be obtained by processing through a surface smoothing algorithm and subtracting the smoothed area from the original data, and the lower surface is also processed. For a few areas with insignificant burrs on the surfaces, calculating three-dimensional point cloud information of the upper surface and the lower surface, namely, firstly, fitting the three-dimensional point cloud information of the upper surface and the lower surface to obtain two square equations, calculating the distance from each point of the surface to a plane equation, drawing a high gray level graph, and obtaining the area of the relevant area, wherein the lower surface can be treated as the same.
(6) When the tile-shaped defect is identified, fitting a circular equation from the inner side and the outer side of the three-dimensional model, calculating the radius of the circle, and determining the tile-shaped defect according to the radius.
As shown in fig. 5, the tile bending degree is one of defects of the sheet material, and the tile shape height is required to be equal to or less than 0.5 mm, and the thickness after removing the tile shape is equal to or more than +1 mm of the thickness of the formed element sheet. The tile shape of the sheet material can be classified into a longitudinal tile shape and a transverse tile shape according to the specifications provided. The tile shape is similar to a section of circular arc, taking the tile shape with longitudinal cross section as an example, the inner side and the outer side of the tile shape can be fitted with a circular equation, and the radius R and the center of a circle are obtained. The more serious the tile shape is, the smaller the radius of the circle is; the lighter the tile, the greater or infinite the radius of the resulting circle. Therefore, by setting the fitted circle radius threshold value, whether the tile shape is formed or not can be judged. The tile-shaped height can be obtained by calculating the extreme value of the three-dimensional point cloud information of the upper surface and the lower surface; or the arc height is calculated by planar geometry.
(7) When the insect defect is identified, a three-dimensional model is processed by adopting a surface smoothing algorithm, and the smoothed data and the original data are subtracted to obtain an insect region.
(8) When identifying the bending defect, removing the side bending part, wherein the width is wide enough; if the width is enough, one side is straight, the plate surface is good, and the other side is not straight.
The plate is deformed under load or action, the sizes of each point along the length direction of the member are different, and the deformed plate presents a curve shape; the degree of deflection in the middle of the regular curve is the greatest, and this value is called lateral bending.
Correspondingly, the error reporting method of the size defect is that through the setting function of an open defect threshold, when a certain size defect exceeds the threshold, unqualified information is prompted or unqualified information is sent.
S104, acquiring two-dimensional color image information of the target plate acquired by the linear array camera.
The invention adopts four linear array cameras to sweep the upper, lower, left and right surfaces of the plate at the same time, and then carries out color imaging on the surfaces to obtain two-dimensional color image information of the respective surfaces.
S105, inputting the two-dimensional color image information into a defect recognition model trained in advance to recognize natural defects of the target plate.
Accordingly, the natural deficiencies include, but are not limited to, dead joints, resin sacs, jackets, and decay.
In addition, when identifying the defects of the crack defects (surface cracks, end cracks and fracture cracks), the three-dimensional model and the defect identification model are combined for processing. Specifically, the surface cracks can be identified through the color imaging of a linear array camera and single-angle polishing, and the depth of the cracks can be analyzed through the combination of the defect identification model and the three-dimensional model, so that the cracks can be further determined.
S106, inputting the two-dimensional color image information into a color classification model trained in advance to identify the color class of the target plate.
Accordingly, the color categories include, but are not limited to, blue, dark, medium, light, and variegated. The plate is changed in blue when being invaded by the color-changing bacteria, and the distribution range of the blue-changing defect area is large, so that the plate is characterized by blue, light blue and the like, and can be identified by adopting a color classification model.
In addition, when the wormhole defect is identified, the three-dimensional model and the color classification model are combined for processing. The judging basis of the wormholes mainly comprises two points: 1. the wormholes are darker than the surrounding plate and are approximately circular; 2. the wormholes can generate deep pits in the plate, and the wormholes can form height differences after being scanned by a 3D profiler. Therefore, wormholes can be identified through the color classification model, and the heights of the wormholes can be analyzed by combining a 3D profiler.
Therefore, the defect of the plate can be comprehensively and accurately detected through the invention, wherein: the length, width and height defects, curvature defects, torsion defects, straight angle defects at two ends, face collapse and head gnawing defects, tile defects, insect growth defects and bending detection defects can be identified through a three-dimensional model constructed through three-dimensional point cloud information, dead joints, resin bags, skin clamping and decay defects can be identified through a defect identification model, blue-change defects and color classification can be identified through a color classification model, crack defects need to be identified through a three-dimensional model and a defect identification model, and wormhole defects need to be identified through a three-dimensional model and a color classification model.
Furthermore, the two-dimensional color image information needs to be preprocessed before being input into a defect recognition model and a color classification model which are trained in advance. The specific pretreatment steps comprise:
(1) SVD decomposition is carried out on the two-dimensional color image information;
(2) Gabor filtering is carried out on the two-dimensional color image information after SVD decomposition;
(3) And performing difference operation on the two-dimensional color image information subjected to Gabor filtering by adopting an iterative difference method.
After median filtering and image enhancement processing are performed by the iterative subtraction method, multiple pixel template extraction and subtraction operation are performed on the image, so that a defect map with higher contrast can be obtained, the recognition interference of the background on the defect is effectively inhibited, the defect characteristics are highlighted, and effective conditions are provided for the next machine learning and recognition.
Therefore, the processing method combining SVD decomposition, gabor filtering and iterative subtraction method is adopted, so that the missing of the processing capacity of the SVD in the non-horizontal vertical direction is avoided, the calculated amount in the Gabor filtering is reduced, and the processing effect and the processing efficiency are both considered.
Referring to fig. 6 and 7, a flowchart of a training method of a defect recognition model according to an embodiment of the present invention includes:
s201, acquiring an original color image;
S202, intercepting and combining the original color image to form a basic color image;
Since the aspect ratio of the original image size is too large after the image is acquired by the linear array camera, the image is not suitable for model learning, and therefore, the image is required to be intercepted and combined according to prior knowledge to form image reconstruction.
S203, marking defect positions and defect categories in the basic color image to form a marked color image;
when the frame is marked, the defect position and defect category can be marked through manual experience. Specifically, defect categories include dead joints, resin sacs, jackets, and decay, but are not so limited.
S204, performing generalization amplification treatment on the marked color image to form a sample color image;
the number of actually collected defective plates is often limited, so that generalization and amplification can be performed through a special algorithm.
S205, integrating all sample color images into a sample color image set;
s206, training a defect recognition model according to the sample color image set.
As shown in fig. 8, the defect recognition model adopts a DNN model to extract multi-scale features of an input sample color image, combines semantic information and position information to fuse the multi-scale features, and combines a attention mechanism and a spatial pyramid structure to extract features so as to recognize and output defect types, defect positions and confidence information. Accordingly, the sample color image is a 640 x 640 reconstructed image.
Referring to fig. 9 and 10, a flowchart of an embodiment of a training method of a color classification model according to the present invention includes:
S301, acquiring an original color image;
S302, extracting a target foreground region of an original color image through a visual algorithm to form an ROI color image;
s303, scaling the target foreground region of the ROI color image and cutting the target foreground region into a plurality of partial images;
S304, marking color categories in the local images to form marked classified images;
When the frame is marked, the color category can be marked through manual experience; specifically, the color classes include, but are not limited to, blue, dark, medium, light, and mottled.
S305, performing generalization amplification treatment on the labeling classified images to form sample classified images;
the number of actually collected defective plates is often limited, so that generalization and amplification can be performed through a special algorithm.
S306, integrating all the sample classification images into a sample classification image set;
S307, training a color classification model according to the sample classification image set.
As shown in fig. 11, the color classification model performs multi-scale feature extraction on an input sample classification image by using a DNN model, classifies high-dimensional feature vectors, and then identifies image semantic classification information to output color category and confidence information. Accordingly, the sample classification image is a 320×320 tile image.
Therefore, the original color image is obtained through the linear array camera, the defect characteristics are highlighted by using the image preprocessing technology, and the identification is carried out through the trained model algorithm, so that the aim of replacing manual detection is fulfilled, and the method has the advantages of high efficiency, low cost and high flexibility.
Referring to fig. 12, fig. 12 shows a specific structure of a visual inspection system 100 for a sheet material according to the present invention, which includes a size defect recognition module 1, a natural defect recognition module 2, and a color classification recognition module 3.
The following describes the above three modules in detail:
1. Size defect recognition module 1
The dimension defect identification module 1 is used for acquiring three-dimensional point cloud information of the target plate acquired by the 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the dimension defect of the target plate according to the three-dimensional model and a preset reference dimension.
In the invention, four 3D profilers are adopted to measure the heights of the plate in the up, down, left and right directions. When a plate to be detected enters a detection area consisting of four 3D profilers from a motion device, opening the four 3D profilers; then, four laser lines of the four 3D profilers respectively scan four planes of the rapidly-moving plate to obtain three-dimensional point cloud information of the plate, and simultaneously, the scanned three-dimensional point cloud information is sent to the size defect identification module 1; stopping the 3D profiler when the plate passes through the field of view of the 3D profiler, and splicing and reconstructing the acquired three-dimensional point cloud information of the whole plate; and finally, comparing the calculation structure with a judgment standard according to the detection procedure of each parameter, and identifying the specific size defect of the related parameter.
Accordingly, the size defect includes any one or more of length, width and height defects, bending defects, torsion defects, straight angle defects at two ends, face collapse and head gnawing defects, tile defects, insect growth defects and bending detection defects, but is not limited thereto. Specifically:
when the length, width and height defects are identified, the length of the target plate is the length of the three-dimensional model, the width of the target plate is the width of the three-dimensional model, and the thickness of the target plate is the thickness of the three-dimensional model. That is, the length of the sheet material can be obtained by the length of the three-dimensional model; the width of the plate is measured by the X-axis direction laser linewidth of a 3D profiler arranged up and down; the thickness of the plate is measured by the X-axis direction laser linewidth of the 3D profiler arranged left and right.
When the curvature defect is identified, extracting height distribution along the length direction of the three-dimensional model, selecting the maximum height value as chord height, and dividing the chord height by the length to obtain the curvature in the length direction. As shown in fig. 2, the plate is in the shape of a slender cuboid, and according to the definition of curvature: the chord height of the bending per meter length is the bending per meter, and the ratio of the total chord height of the total length bending to the total length is the total bending. Therefore, the height distribution in the length direction in the three-dimensional model can be converted, the chord height of the maximum value is selected, and the value is divided by the total length to obtain the curvature in the length direction; the maximum of the two surfaces is the curvature due to the upper and lower surfaces.
When the torsion degree defect is identified, the distance from any one corner to the other three corner planes on one surface in the three-dimensional model is calculated, and the maximum distance value is taken as the torsion degree. As shown in fig. 3, the plate is in the shape of a slender cuboid, and according to the definition of torsion resistance: one corner of a certain plane is arched upwards, and the other three corners are placed on one plane, wherein the distance between the suspended corner and the other three-point plane is torsion. Therefore, the method can be converted into calculation of the distance from any angle of one surface to the plane of the other triangle, and the maximum distance is taken as the curvature; because of the upper and lower sides, the maximum value is taken.
When the defects of the straight angles at the two ends are identified, calculating the included angle between planes in the three-dimensional model, and taking the included angle as the straight angles at the two ends. As shown in fig. 4, the straight angles at two ends can be regarded as the angle measurement between two ends and the side edge, and the angle measurement can be converted into a plane-to-plane angle solution according to a three-dimensional model.
When the broken surface gnawing defects are identified, a three-dimensional model is processed by adopting a surface smoothing algorithm, the broken surface gnawing head area can be obtained by subtracting the smoothed data from the original data, or the broken surface gnawing head area can be obtained by fitting an upper surface equation through upper surface three-dimensional point cloud information and fitting a lower surface equation through lower surface three-dimensional point cloud information, the distance between each point and the upper surface equation and the lower surface equation is calculated, and a gray level graph is drawn. The surface collapse is mainly caused by that the cutter does not fall into or the height of the plate surface is slightly lower, and is characterized in that obvious burrs are formed on the surface of the plate, so that an obvious dark area is formed at the waist of the plate; therefore, the area of the broken surface can be obtained by processing through a surface smoothing algorithm and subtracting the smoothed area from the original data, and the lower surface is also processed. For a few areas with insignificant burrs on the surfaces, calculating three-dimensional point cloud information of the upper surface and the lower surface, namely, firstly, fitting the three-dimensional point cloud information of the upper surface and the lower surface to obtain two square equations, calculating the distance from each point of the surface to a plane equation, drawing a high gray level graph, and obtaining the area of the relevant area, wherein the lower surface can be treated as the same.
When the tile-shaped defect is identified, fitting a circular equation from the inner side and the outer side of the three-dimensional model, calculating the radius of the circle, and determining the tile-shaped defect according to the radius. As shown in fig. 5, the tile bending is one of the defects of the sheet material, and the tile shape height is required to be equal to or less than 0.5 mm, and the thickness after removing the tile shape is equal to or more than +1 mm of the thickness of the formed element sheet. The tile shape of the sheet material can be classified into a longitudinal tile shape and a transverse tile shape according to the specifications provided. The tile shape is similar to a section of circular arc, taking the tile shape with longitudinal cross section as an example, the inner side and the outer side of the tile shape can be fitted with a circular equation, and the radius and the center of a circle are obtained. The more serious the tile shape is, the smaller the radius of the circle is; the lighter the tile, the greater or infinite the radius of the resulting circle. Therefore, by setting the fitted circle radius threshold value, whether the tile shape is formed or not can be judged. The tile-shaped height can be obtained by calculating the extreme value of the three-dimensional point cloud information of the upper surface and the lower surface; or the arc height is calculated by planar geometry.
When the insect defect is identified, a three-dimensional model is processed by adopting a surface smoothing algorithm, and the smoothed data and the original data are subtracted to obtain an insect region.
When identifying the bending defect, removing the side bending part, wherein the width is wide enough; if the width is enough, one side is straight, the plate surface is good, and the other side is not straight. The plate is deformed under load or action, the sizes of each point along the length direction of the member are different, and the deformed plate is in a curve shape; the degree of deflection in the middle of the regular curve is the greatest, and this value is called lateral bending.
Correspondingly, the error reporting method of the size defect is that through the setting function of an open defect threshold, when a certain size defect exceeds the threshold, unqualified information is prompted or unqualified information is sent.
2. Natural defect recognition module 2
The natural defect recognition module 2 is used for acquiring the two-dimensional color image information of the target plate acquired by the linear array camera, and inputting the two-dimensional color image information into a defect recognition model trained in advance to recognize the natural defect of the target plate. Accordingly, the natural deficiencies include, but are not limited to, dead joints, resin sacs, jackets, and decay.
The invention uses four linear cameras to sweep the upper, lower, left and right surfaces of the board, and then to color image the surfaces to obtain two-dimensional color image information of each surface.
Further, when identifying the crack defects (surface cracks, end cracks and fracture cracks), the three-dimensional model and the defect identification model are combined for processing. Specifically, the surface cracks can be identified through the color imaging of a linear array camera and single-angle polishing, and the depth of the cracks can be analyzed through the combination of the defect identification model and the three-dimensional model, so that the cracks can be further determined.
3. Color classification recognition module 3
The color classification and identification module 3 is used for acquiring the two-dimensional color image information of the target plate acquired by the line camera, and inputting the two-dimensional color image information into a color classification model trained in advance to identify the color class of the target plate.
Accordingly, the color categories include, but are not limited to, blue, dark, medium, light, and variegated. The plate is changed in blue when being invaded by the color-changing bacteria, and the distribution range of the blue-changing defect area is large, so that the plate is characterized by blue, light blue and the like, and can be identified by adopting a color classification model.
Further, when identifying the wormhole defect, the three-dimensional model and the color classification model are combined for processing. The judging basis of the wormholes mainly comprises two points: 1. the wormholes are darker than the surrounding plate and are approximately circular; 2. the wormholes can generate deep pits in the plate, and the wormholes can form height differences after being scanned by a 3D profiler. Therefore, wormholes can be identified through the color classification model, and the heights of the wormholes can be analyzed by combining a 3D profiler.
Therefore, the invention can carry out comprehensive and accurate detection on the defects of the plate, can solve the problems of fatigue error, unqualified detection rate and the like of the current manual detection, and has the advantages of good flexibility, strong performance and high detection rate.
As shown in fig. 13, the sheet visual inspection system 100 further includes a defect recognition training module 4; specifically, the defect recognition training module 4 includes:
A first acquisition unit 41 for acquiring an original color image;
a merging unit 42 for performing a clipping and merging process on the original color image to form a basic color image; since the aspect ratio of the original image size is too large after the image is acquired by the linear array camera, the image is not suitable for model learning, and therefore, the image is required to be intercepted and combined according to prior knowledge to form image reconstruction.
A first labeling unit 43 for labeling defect positions and defect categories in the basic color image to form a labeled color image, wherein the defect categories comprise dead joints, resin bags, jackets and decay; when the frame is marked, the defect position and defect category can be marked through manual experience. Specifically, defect categories include dead joints, resin sacs, jackets, and decay, but are not so limited.
A first amplification unit 44 for performing a generalization amplification process on the labeled color image to form a sample color image; the number of actually collected defective plates is often limited, so that generalization and amplification can be performed through a special algorithm.
A first integration unit 45 for integrating all the sample color images into a sample color image set;
A first training unit 46 for training the defect recognition model on the basis of the set of sample color images.
As shown in fig. 8, the defect recognition model adopts a DNN model to extract multi-scale features of an input sample color image, combines semantic information and position information to fuse the multi-scale features, and combines a attention mechanism and a spatial pyramid structure to extract features so as to recognize and output defect types, defect positions and confidence information. Accordingly, the sample color image is a 640 x 640 reconstructed image.
As shown in fig. 14, the board visual inspection system 100 further includes a color classification training module 5; specifically, the color classification training module 5 includes:
a second acquisition unit 51 for acquiring an original color image;
An extracting unit 52 for extracting a target foreground region of the original color image by a visual algorithm to form an ROI color image;
A cutting unit 53 for performing a scaling process on a target foreground region of the ROI color image and cutting into a plurality of partial images;
a second labeling unit 54 for labeling color classes in the partial images to form labeled classified images, the color classes including blue, dark, medium, and light; when the frame is marked, the color category can be marked through manual experience; specifically, the color classes include, but are not limited to, blue, dark, medium, light, and mottled.
A second amplification unit 55, configured to perform generalized amplification processing on the labeling classification image to form a sample classification image; the number of actually collected defective plates is often limited, so that generalization and amplification can be performed through a special algorithm.
A second integration unit 56 for integrating all the sample classification images into a sample classification image set;
A second training unit 57 for training a color classification model based on the set of sample classification images.
As shown in fig. 11, the color classification model performs multi-scale feature extraction on an input sample classification image by using a DNN model, classifies high-dimensional feature vectors, and then identifies image semantic classification information to output color category and confidence information. Accordingly, the sample classification image is a 320×320 tile image.
Therefore, the invention can obtain the original color image through the linear array camera, then uses the image preprocessing technology to highlight the defect characteristics, and recognizes through the trained model algorithm, thereby achieving the purpose of replacing manual detection and having the advantages of high efficiency, low cost and strong flexibility.
As shown in fig. 15, the board visual inspection system 100 further includes a preprocessing module 6, and the preprocessing module 6 is further required to preprocess the two-dimensional color image information before inputting the two-dimensional color image information into the defect recognition model and the color classification model trained in advance. Specifically, the preprocessing module 6 includes:
An SVD unit 61 for performing SVD decomposition on the two-dimensional color image information;
a Gabor unit 62 for Gabor filtering the two-dimensional color image information after SVD decomposition;
And a difference image unit 63, configured to perform a difference image operation on the two-dimensional color image information after Gabor filtering by using an iterative difference image method. After median filtering and image enhancement processing are performed by the iterative subtraction method, multiple pixel template extraction and subtraction operation are performed on the image, so that a defect map with higher contrast can be obtained, the recognition interference of the background on the defect is effectively inhibited, the defect characteristics are highlighted, and effective conditions are provided for the next machine learning and recognition.
Therefore, the preprocessing module 6 adopts a processing method combining SVD decomposition, gabor filtering and iterative subtraction method, which not only avoids the lack of processing capacity of SVD in the non-horizontal vertical direction, but also reduces the calculated amount in Gabor filtering, and gives consideration to the processing effect and the processing efficiency.
In addition, the invention adopts the edge computing technology to process the data in the size defect recognition module 1, the natural defect recognition module 2, the color classification recognition module 3, the defect recognition training module 4, the color classification training module 5 and the preprocessing module 6.
It should be noted that, the image data obtained by the image acquisition end of the traditional machine vision detection needs to be uploaded to the cloud end, or the image data is transmitted to the image preprocessing center and the discrimination server for processing and recognition through the local area network, and the recognition result is returned to the client. Because of the large photo data volume, if the number of production lines is large, the centralized image preprocessing and judging service center will be subjected to serious performance challenges, the time delay is large, and the detection efficiency is reduced. According to the invention, an edge computing technology is adopted, an edge cloud cooperation concept is used, and a public cloud or private cloud machine learning platform is used for image storage, model training, tuning, publishing, management, evaluation, labeling system and other functions, so that a trained model can be issued to edge equipment, two-dimensional color image information acquired by a linear array camera can be directly and nearby identified and an identification result can be returned, the characteristics of time delay and distributed computing are greatly reduced, the problem of high load and low performance of a centralized judging center and an image processing center is solved, the method can effectively adapt to application scenes with multiple production lines, and the detection efficiency is greatly improved.
Correspondingly, the invention also introduces a middle stage concept, and the common general capabilities of a machine learning frame, a deep learning frame, model training, model tuning, model evaluation, model release, model life cycle management, manual standard, semi-automatic labeling, defect statistics and the like are abstractly integrated to form the field capabilities of a model center, a standard center, an intelligent learning center, a statistical center and the like, and the upper layer application can realize the corresponding functions only by calling an interface, so that repeated development and repeated construction are avoided, and other fields of machine vision detection can be based on the middle stage of the project AI to rapidly hatch corresponding vision detection products.
From the above, the invention combines the image preprocessing technology, the edge computing technology and the deep learning technology to create the defect detection equipment with the combination of softness and hardness and Bian Yun cooperation, has the advantages of good flexibility, high detection rate and low time delay, effectively controls the shipment quality, reduces the after-sale cost, and has far-reaching significance for pushing the automation of the production line and the industry of 4.0.
While the foregoing is directed to the preferred embodiments of the present invention, it will be appreciated by those skilled in the art that changes and modifications may be made without departing from the principles of the invention, such changes and modifications are also intended to be within the scope of the invention.
Claims (9)
1. The plate visual inspection method is characterized by comprising the following steps of:
acquiring three-dimensional point cloud information of a target plate acquired by a 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the size defect of the target plate according to the three-dimensional model and a preset reference size;
Acquiring two-dimensional color image information of a target plate acquired by a linear array camera, inputting the two-dimensional color image information into a pre-trained defect recognition model to recognize natural defects of the target plate, and inputting the two-dimensional color image information into a pre-trained color classification model to recognize color categories of the target plate;
The preprocessing of the two-dimensional color image information before the two-dimensional color image information is input into a defect recognition model and a color classification model which are trained in advance, and the preprocessing comprises the following steps: SVD decomposition is carried out on the two-dimensional color image information; gabor filtering is carried out on the two-dimensional color image information after SVD decomposition; and performing difference operation on the two-dimensional color image information subjected to Gabor filtering by adopting an iterative difference method.
2. The method for visual inspection of sheet material according to claim 1, further comprising training a defect identification model, comprising:
acquiring an original color image;
intercepting and combining the original color image to form a basic color image;
marking defect positions and defect categories in the basic color image to form a marked color image, wherein the defect categories comprise dead joints, resin bags, skin clamping and decay;
Performing generalization amplification treatment on the marked color image to form a sample color image;
integrating all sample color images into a sample color image set;
training the defect recognition model according to the sample color image set.
3. The visual inspection method of sheet material according to claim 2, wherein the defect recognition model adopts a DNN model to extract multi-scale features of the input sample color image, combines semantic information and position information to fuse the multi-scale features, and combines a attention mechanism and a spatial pyramid structure to extract features to recognize and output defect types, defect positions and confidence information.
4. The method for visual inspection of sheet material according to claim 1, further comprising training a color classification model, comprising:
acquiring an original color image;
Extracting a target foreground region of the original color image through a visual algorithm to form an ROI color image;
scaling the target foreground region of the RO I color image and cutting the target foreground region into a plurality of partial images;
Marking color categories in the partial image to form a marked classified image, the color categories including blue, dark, medium, and light;
Performing generalization amplification treatment on the labeling classified images to form sample classified images;
integrating all the sample classification images into a sample classification image set;
Training the color classification model according to the sample classification image set.
5. The method of claim 4, wherein the color classification model uses DNN model to extract multi-scale features of the input sample classification image, classifies the high-dimensional feature vectors, and identifies the semantic classification information of the image to output color class and confidence information.
6. The visual inspection method of the sheet material according to claim 1, wherein the dimensional defect comprises any one or more of a length-width-height defect, a curvature defect, a torsion defect, a two-end straight angle defect, a broken surface gnawing head defect, a tile-shaped defect and a insect-growth defect;
When the length, width and height defects are identified, the length of the target plate is the length of the three-dimensional model, the width of the target plate is the width of the three-dimensional model, and the thickness of the target plate is the thickness of the three-dimensional model;
When the curvature defect is identified, extracting height distribution along the length direction of the three-dimensional model, selecting a maximum height value as chord height, and dividing the chord height by the length to obtain curvature in the length direction;
When the torsion degree defect is identified, calculating the distance from any one corner to the other three corner planes on one surface in the three-dimensional model, and taking the maximum distance value as the torsion degree;
when the two-end straight angle defects are identified, calculating an included angle between planes in the three-dimensional model, and taking the included angle as the two-end straight angle;
when the broken surface gnawing head defect is identified, a surface smoothing algorithm is adopted to process the three-dimensional model, the smoothed data and the original data are subtracted to obtain a broken surface gnawing head region, or an upper surface equation is fitted through upper surface three-dimensional point cloud information, a lower surface equation is fitted through lower surface three-dimensional point cloud information, the distance from each point to the upper surface equation and the lower surface equation is calculated, and a gray level diagram is drawn to obtain the broken surface gnawing head region;
When the tile-shaped defect is identified, fitting a circular equation from the inner side and the outer side of the three-dimensional model, calculating the radius of a circle, and determining the tile-shaped defect according to the radius;
And when the insect-living defect is identified, processing the three-dimensional model by adopting a surface smoothing algorithm, and subtracting the smoothed data from the original data to obtain an insect-living region.
7. A panel vision inspection system, comprising:
the dimensional defect identification module is used for acquiring three-dimensional point cloud information of the target plate acquired by the 3D profiler, constructing a three-dimensional model according to the three-dimensional point cloud information, and identifying the dimensional defect of the target plate according to the three-dimensional model and a preset reference size;
the natural defect identification module is used for acquiring the two-dimensional color image information of the target plate acquired by the linear array camera, and inputting the two-dimensional color image information into a defect identification model trained in advance to identify the natural defect of the target plate;
the color classification and identification module is used for acquiring two-dimensional color image information of a target plate acquired by the linear array camera, and inputting the two-dimensional color image information into a color classification model trained in advance to identify the color class of the target plate;
The preprocessing module is used for preprocessing the two-dimensional color image information before inputting the two-dimensional color image information into the defect recognition model and the color classification model which are trained in advance; wherein, the preprocessing module includes: the SVD unit is used for carrying out SVD decomposition on the two-dimensional color image information; the Gabor unit is used for carrying out Gabor filtering on the two-dimensional color image information after SVD decomposition; and the difference image unit is used for carrying out difference image operation on the two-dimensional color image information subjected to Gabor filtering by adopting an iterative difference image method.
8. The sheet visual inspection system of claim 7, further comprising a defect identification training module, the defect identification training module comprising:
a first acquisition unit configured to acquire an original color image;
The merging unit is used for intercepting and merging the original color images to form a basic color image;
a first marking unit for marking defect positions and defect categories in the basic color image to form a marked color image, wherein the defect categories comprise dead joints, resin bags, skin clamping and decay;
the first amplification unit is used for performing generalization amplification treatment on the marked color image so as to form a sample color image;
a first integration unit for integrating all the sample color images into a sample color image set;
and the first training unit is used for training the defect recognition model according to the sample color image set.
9. The sheet visual inspection system of claim 7 further comprising a color classification training module, the color classification training module comprising:
a second acquisition unit configured to acquire an original color image;
An extraction unit for extracting a target foreground region of the original color image by a visual algorithm to form an RO I color image;
a cutting unit for performing scaling processing on the target foreground region of the RO I color image and cutting the target foreground region into a plurality of partial images;
The second labeling unit is used for labeling color categories in the local image to form a labeling classified image, wherein the color categories comprise blue, dark, medium and light;
The second amplification unit is used for carrying out generalization amplification treatment on the labeling classified images so as to form sample classified images;
the second integration unit is used for integrating all the sample classification images into a sample classification image set;
and the second training unit is used for training the color classification model according to the sample classification image set.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110454981.7A CN113362276B (en) | 2021-04-26 | 2021-04-26 | Visual detection method and system for plates |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202110454981.7A CN113362276B (en) | 2021-04-26 | 2021-04-26 | Visual detection method and system for plates |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN113362276A CN113362276A (en) | 2021-09-07 |
| CN113362276B true CN113362276B (en) | 2024-05-10 |
Family
ID=77525533
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110454981.7A Active CN113362276B (en) | 2021-04-26 | 2021-04-26 | Visual detection method and system for plates |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN113362276B (en) |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114199892B (en) * | 2021-12-10 | 2022-11-18 | 江苏雷默智能科技有限公司 | Plate measuring method and system based on machine vision |
| CN114812440A (en) * | 2022-04-22 | 2022-07-29 | 国网湖北省电力有限公司十堰供电公司 | High-rise building fireproof plugging module defect detection device and method based on line structured light |
| CN114820761B (en) * | 2022-05-07 | 2024-05-10 | 北京毅能博科技有限公司 | XY direction included angle measurement and motion compensation method based on image microscopic scanning platform |
Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107052086A (en) * | 2017-06-01 | 2017-08-18 | 扬州苏星机器人科技有限公司 | Stamping parts surface defect detection apparatus and detection method based on 3D vision |
| WO2018000731A1 (en) * | 2016-06-28 | 2018-01-04 | 华南理工大学 | Method for automatically detecting curved surface defect and device thereof |
| CN108010036A (en) * | 2017-11-21 | 2018-05-08 | 江南大学 | A kind of object symmetry axis detection method based on RGB-D cameras |
| WO2018136262A1 (en) * | 2017-01-20 | 2018-07-26 | Aquifi, Inc. | Systems and methods for defect detection |
| EP3373251A1 (en) * | 2017-03-07 | 2018-09-12 | Trimble AB | Scan colorization with an uncalibrated camera |
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN110308151A (en) * | 2019-07-22 | 2019-10-08 | 重庆大学 | A kind of surface of workpiece defect identification method and device based on machine vision |
| CN110533642A (en) * | 2019-08-21 | 2019-12-03 | 深圳新视达视讯工程有限公司 | A kind of detection method of insulator damage |
Family Cites Families (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US10046459B2 (en) * | 2015-11-16 | 2018-08-14 | Abb Schweiz Ag | Three-dimensional visual servoing for robot positioning |
| US20190096135A1 (en) * | 2017-09-26 | 2019-03-28 | Aquifi, Inc. | Systems and methods for visual inspection based on augmented reality |
-
2021
- 2021-04-26 CN CN202110454981.7A patent/CN113362276B/en active Active
Patent Citations (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2018000731A1 (en) * | 2016-06-28 | 2018-01-04 | 华南理工大学 | Method for automatically detecting curved surface defect and device thereof |
| WO2018136262A1 (en) * | 2017-01-20 | 2018-07-26 | Aquifi, Inc. | Systems and methods for defect detection |
| EP3373251A1 (en) * | 2017-03-07 | 2018-09-12 | Trimble AB | Scan colorization with an uncalibrated camera |
| CN107052086A (en) * | 2017-06-01 | 2017-08-18 | 扬州苏星机器人科技有限公司 | Stamping parts surface defect detection apparatus and detection method based on 3D vision |
| CN108010036A (en) * | 2017-11-21 | 2018-05-08 | 江南大学 | A kind of object symmetry axis detection method based on RGB-D cameras |
| WO2019104767A1 (en) * | 2017-11-28 | 2019-06-06 | 河海大学常州校区 | Fabric defect detection method based on deep convolutional neural network and visual saliency |
| CN110308151A (en) * | 2019-07-22 | 2019-10-08 | 重庆大学 | A kind of surface of workpiece defect identification method and device based on machine vision |
| CN110533642A (en) * | 2019-08-21 | 2019-12-03 | 深圳新视达视讯工程有限公司 | A kind of detection method of insulator damage |
Also Published As
| Publication number | Publication date |
|---|---|
| CN113362276A (en) | 2021-09-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN113362276B (en) | Visual detection method and system for plates | |
| CN111915572B (en) | Adaptive gear pitting quantitative detection system and method based on deep learning | |
| CN103913468A (en) | Multi-vision defect detecting equipment and method for large-size LCD glass substrate in production line | |
| CN115294117B (en) | Defect detection method and related device for LED lamp beads | |
| CN113066079A (en) | Method, system and storage medium for automatically detecting wood defects | |
| CN110223267A (en) | The recognition methods of refractory brick deep defects based on height histogram divion | |
| CN115797354B (en) | Method for detecting appearance defects of laser welding seam | |
| CN112184648A (en) | Piston surface defect detection method and system based on deep learning | |
| CN117152161B (en) | Shaving board quality detection method and system based on image recognition | |
| CN114638797A (en) | Method and device for detecting copper surface defects based on linear array camera | |
| CN110992337A (en) | Container damage detection method and system | |
| CN110910382A (en) | Container detection system | |
| CN114581805A (en) | Coating roller surface defect detection method adopting 3D line laser profile technology | |
| CN114529510A (en) | Cathode copper online quality automatic detection and classification method | |
| CN114387262A (en) | Nut positioning detection method, device and system based on machine vision | |
| Mustaffar et al. | Automated pavement imaging program (APIP) for pavement cracks classification and quantification-a photogrammetric approach | |
| CN114494845A (en) | Artificial intelligence hidden danger troubleshooting system and method for construction project site | |
| CN116740449A (en) | Shaving form detection method and system based on AI (advanced technology attachment) computer vision technology | |
| CN107478656A (en) | Paper pulp mixing effect method of determination and evaluation based on machine vision, device, system | |
| JP2015163999A (en) | Analysis system and method of repeated pattern image, and analysis system and method of repeated structural drawing | |
| CN114529536A (en) | Solid wood quality detection method | |
| Orjuela et al. | Feature extraction of the wear label of carpets by using a novel 3d scanner | |
| CN113570587A (en) | Photovoltaic cell broken grid detection method and system based on computer vision | |
| Puan et al. | Automated pavement imaging program (APIP) for pavement cracks classification and quantification | |
| Marzouk et al. | Quality analysis using three-dimensional modelling and image processing techniques |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |