CN112233067A - Hot rolled steel coil end face quality detection method and system - Google Patents
Hot rolled steel coil end face quality detection method and system Download PDFInfo
- Publication number
- CN112233067A CN112233067A CN202010997054.5A CN202010997054A CN112233067A CN 112233067 A CN112233067 A CN 112233067A CN 202010997054 A CN202010997054 A CN 202010997054A CN 112233067 A CN112233067 A CN 112233067A
- Authority
- CN
- China
- Prior art keywords
- image
- steel coil
- hot
- rolled steel
- detected
- 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.)
- Pending
Links
- 229910000831 Steel Inorganic materials 0.000 title claims abstract description 163
- 239000010959 steel Substances 0.000 title claims abstract description 163
- 238000001514 detection method Methods 0.000 title claims description 19
- 230000007547 defect Effects 0.000 claims abstract description 79
- 238000012545 processing Methods 0.000 claims abstract description 42
- 238000000034 method Methods 0.000 claims abstract description 34
- 238000005520 cutting process Methods 0.000 claims abstract description 33
- 238000000605 extraction Methods 0.000 claims abstract description 27
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 23
- 235000020061 kirsch Nutrition 0.000 claims abstract description 15
- 238000005286 illumination Methods 0.000 claims abstract description 14
- 239000013589 supplement Substances 0.000 claims abstract description 14
- 238000010845 search algorithm Methods 0.000 claims abstract description 10
- 239000011159 matrix material Substances 0.000 claims description 25
- 238000011176 pooling Methods 0.000 claims description 11
- 238000004422 calculation algorithm Methods 0.000 claims description 10
- 238000010586 diagram Methods 0.000 claims description 8
- 238000001914 filtration Methods 0.000 claims description 8
- 238000004364 calculation method Methods 0.000 claims description 6
- 230000000877 morphologic effect Effects 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 4
- 238000005098 hot rolling Methods 0.000 claims description 3
- 238000003708 edge detection Methods 0.000 description 5
- 238000012986 modification Methods 0.000 description 5
- 230000004048 modification Effects 0.000 description 5
- 239000000284 extract Substances 0.000 description 4
- 230000001502 supplementing effect Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 230000000694 effects Effects 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 230000004075 alteration Effects 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 230000006835 compression Effects 0.000 description 1
- 238000007906 compression Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000009432 framing Methods 0.000 description 1
- 238000003384 imaging method Methods 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000035772 mutation Effects 0.000 description 1
- 238000012634 optical imaging Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
Images
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
-
- 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
-
- G06T5/70—
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/12—Edge-based segmentation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- 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
- G06T2207/10—Image acquisition modality
- G06T2207/10004—Still image; Photographic image
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20024—Filtering details
- G06T2207/20032—Median filtering
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20092—Interactive image processing based on input by user
- G06T2207/20104—Interactive definition of region of interest [ROI]
-
- 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
- G06T2207/30168—Image quality inspection
Abstract
The invention discloses a method and a system for detecting the end surface quality of a hot-rolled steel coil, wherein the method comprises the following steps: acquiring an end face image of a hot-rolled steel coil shot by an industrial camera; eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter; searching and cutting the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold search algorithm, and extracting an image of a region to be detected; carrying out illumination equalization processing on the image of the region to be detected, improving the brightness of the image of the region to be detected, and obtaining a supplemented image; detecting the defects of the images after the light supplement by using a Kirsch operator, and eliminating noise points of the images to obtain preprocessed images; and extracting primary features from the preprocessed image by using a feature extraction layer, extracting an interested region from the primary features by using a regional network candidate layer, processing the interested region by using a convolutional neural network rapid region calibration layer, and extracting defect features in the end face image of the hot-rolled steel coil. The method and the system provided by the invention can be used for efficiently detecting the end surface quality of the hot-rolled steel coil.
Description
Technical Field
The invention relates to the technical field of quality detection, in particular to a method and a system for detecting the end face quality of a hot-rolled steel coil.
Background
The coiling quality of the hot-rolled strip steel is the most important quality index of a hot continuous rolling coiling area, the quality problems of edge loss, folding, riffled lines, tower shape and the like exist in the strip steel coiling process, and if the quality problems are not discovered and processed in time, great quality objectional loss is caused. The end surface texture patterns of the steel coil are complex, the defect forms are different, the defect detection is very difficult, and at present, a steel coil quality detection system with accurate classification and high identification rate is not available in China.
The existing hot rolling strip steel coil quality identification is completely determined by a skilled worker through naked eyes, and has the problems of low efficiency, easy omission and the like.
Therefore, a method for detecting the end face quality of the hot-rolled steel coil with high efficiency and capable of avoiding missing detection is needed.
Disclosure of Invention
The embodiment of the application provides the method and the system for detecting the end face quality of the hot-rolled steel coil, so that the end face quality of the hot-rolled steel coil can be efficiently detected.
The invention provides a method for detecting the end surface quality of a hot-rolled steel coil, which comprises the following steps:
acquiring an end face image of a hot-rolled steel coil shot by an industrial camera; the image corresponding to the end face of each hot rolled steel coil is shot by two industrial cameras;
eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
searching and cutting the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold search algorithm, and extracting an image of a region to be detected;
carrying out illumination equalization processing on the image of the region to be detected, improving the brightness of the image of the region to be detected, and obtaining a supplemented image;
detecting the defects of the images after the light supplement by using a Kirsch operator, and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain preprocessed images;
and extracting primary features from the preprocessed image by using a feature extraction layer, extracting an interested region from the primary features by using a regional network candidate layer, processing the interested region by using a convolutional neural network rapid region calibration layer, extracting defect features in the image of the end face of the hot-rolled steel coil, and outputting the defect features.
Preferably, the method for searching and cutting the edge of the image of the end face of the hot-rolled steel coil by using the gray gradient threshold search algorithm to extract the image of the area to be detected comprises the following steps:
performing X-direction convolution on the end face image of the hot-rolled steel coil by using a Sobel operator to obtain a first Sobel result graph, and performing Y-direction convolution on the end face image of the hot-rolled steel coil by using the Sobel operator to obtain a second Sobel result graph;
and performing column projection on the first Sobel result graph to obtain a column projection matrix, searching left and right boundaries of a steel coil in the end face image of the hot-rolled steel coil from left and right initial positions of the column projection matrix, cutting after the left and right boundaries of the steel coil are searched, performing column projection on the second Sobel result graph to obtain a row projection matrix, searching upper and lower boundaries of the steel coil in the end face image of the hot-rolled steel coil from upper and lower initial positions of the row projection matrix, and cutting after the upper and lower boundaries of the steel coil are searched to obtain an image of the area to be detected.
Preferably, the method searches and cuts the edge of the image of the end face of the hot-rolled steel coil by using a gray gradient threshold search algorithm, extracts the image of the area to be detected, and further comprises the following steps:
and judging whether the actual height value of the image of the area to be detected is smaller than a preset height value or not, or whether the actual width value of the image of the area to be detected is smaller than a preset width value or not, and if the actual height value of the image of the area to be detected is smaller than the preset height value or the actual width value of the image of the area to be detected is smaller than the preset width value, determining that the image of the end face of the hot-rolled steel coil is invalid.
Preferentially, treat that the regional image of detecting carries out illumination balanced treatment, improves the luminance of treating the regional image of detecting, obtains the image after the light filling, includes:
performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the image of the area to be detected by a compensation coefficient C corresponding to the row of pixel pointsiObtaining a light-supplemented image;
wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the image of the area to be detected, and j is more than or equal to 1 and less than or equal to A.
Preferably, the extracting primary features from the preprocessed image by the feature extraction layer includes:
and processing the preprocessed image through double-layer sparse filtering and a VGG-16 network in sequence, and extracting primary features.
Preferably, the extracting the region of interest from the primary feature by using the local area network candidate layer includes:
sequentially carrying out convolution processing on the primary features by 5 multiplied by 5 and convolution processing by 1 multiplied by 1 to obtain a feature map;
convolving the characteristic diagram through a K-dimensional sliding window to obtain a 2 xK classification result and a 4 xK classification result, wherein K is more than or equal to 1;
and comparing the 2 xK classification result with the 4 xK classification result to perform sample screening to obtain the region of interest.
Preferably, the method for processing the region of interest by using the convolutional neural network fast region calibration layer and extracting the defect characteristics in the hot-rolled steel coil end face image comprises the following steps:
performing pooling treatment on the region of interest;
inputting the pooling treatment result of the region of interest into the full-connection layer for treatment;
and performing regression and classification processing on the processing result of the full connection layer to obtain a position offset and a probability vector, and obtaining defect characteristics in the end face image of the hot-rolled steel coil through the position offset and the probability vector.
The invention also provides a system for detecting the end surface quality of the hot-rolled steel coil, which comprises the following components:
the image acquisition module is used for acquiring an end face image of the hot-rolled steel coil shot by the industrial camera; the image corresponding to the end face of each hot rolled steel coil is shot by two industrial cameras;
the first denoising module is used for eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
the edge cutting module is used for searching and cutting the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold value searching algorithm and extracting an image of a region to be detected;
the image light supplementing module is used for carrying out illumination balance processing on the image of the area to be detected, improving the brightness of the image of the area to be detected and obtaining a supplemented image;
the second denoising module is used for detecting the defects of the images after the light supplement by using a Kirsch operator and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain preprocessed images;
and the defect feature extraction module is used for extracting primary features from the preprocessed image by using the feature extraction layer, extracting an interested region from the primary features by using the regional network candidate layer, processing the interested region by using the convolutional neural network rapid region calibration layer, extracting defect features in the image of the end face of the hot-rolled steel coil, and outputting the defect features.
Preferably, the edge cropping module includes:
the convolution operation unit is used for performing X-direction convolution on the end face image of the hot-rolled steel coil by using a Sobel operator to obtain a first Sobel result graph and performing Y-direction convolution on the end face image of the hot-rolled steel coil by using the Sobel operator to obtain a second Sobel result graph;
and the cutting unit is used for performing column projection on the first Sobel result graph to obtain a column projection matrix, searching left and right boundaries of the steel coil in the hot-rolled steel coil end surface image from left and right initial positions of the column projection matrix, cutting after the left and right boundaries of the steel coil are searched, performing line projection on the second Sobel result graph to obtain a line projection matrix, searching upper and lower boundaries of the steel coil in the hot-rolled steel coil end surface image from upper and lower initial positions of the line projection matrix, and cutting after the upper and lower boundaries of the steel coil are searched to obtain an image of the area to be detected.
Preferably, the image fill-in light module is configured to:
performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the image of the area to be detected by a compensation coefficient C corresponding to the row of pixel pointsiObtaining a light-supplemented image;
wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the image of the area to be detected, and j is more than or equal to 1 and less than or equal to A.
The method and the system provided by the invention have the following beneficial effects: according to the method, the images of the end faces of the steel coils are collected, the images corresponding to the end faces of each hot rolled steel coil are obtained by shooting through two industrial cameras, the image collection efficiency and the image collection precision can be improved, the median filter is used for conducting weighted average processing on the whole image, the noise of the image is reduced, irrelevant information in the image can be eliminated, useful real information is recovered, then the gray gradient threshold value search algorithm is adopted to complete the definition of the edge of the area to be detected, the illumination equalization algorithm is used for conducting optical compensation on the area to be detected, the brightness of the area to be detected is improved, the Kirsch operator is used for detecting defects after the completion of the weighted average processing, and the noise points of the defects after the Ki. And extracting primary features from the preprocessed image by using a feature extraction layer, extracting an interested region from the primary features by using a regional network candidate layer, extracting double-depth features as input of a fast region calibration layer of a convolutional neural network (convolutional neural network), and finally extracting defect features in the steel coil image by using the fast R-CNN to finish the identification and marking of defect types, improve the detection and identification precision of the fast R-CNN position, output indexes such as the type of detected defects, the upper left corner coordinate of a defect region, the defect width, the defect height, the defect shape, the number of pixels occupied by the defects and the like, and form a document which can be downloaded and stored. The invention can obtain the defect detection image of the end face of the steel coil with the detection rate of more than 90 percent in the environments of high temperature, high speed and the like.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings based on the drawings without creative efforts.
FIG. 1 is a flow chart of a method for detecting the end surface quality of a hot-rolled steel coil according to an embodiment of the present invention;
FIG. 2a is a schematic diagram of an arrangement of an industrial camera and a light source provided by an embodiment of the invention;
FIG. 2b is a schematic diagram of a side view industrial camera provided by an embodiment of the present invention;
fig. 2c is a schematic diagram of an industrial camera shooting an end surface area of a steel coil according to an embodiment of the present invention;
FIG. 3 is a flow chart of edge searching and cutting of an image of an end face of a hot-rolled steel coil according to an embodiment of the present invention;
FIG. 4a is a raw simulation image of a hot rolled steel coil provided in an embodiment of the present invention;
fig. 4b is a Sobel operator edge detection diagram corresponding to the original simulation image of the hot-rolled steel coil provided in the embodiment of the present invention;
fig. 5 is a flowchart illustrating illumination equalization processing performed on an image of a region to be detected according to an embodiment of the present invention;
fig. 6 is a flowchart of extracting the primary features, the regions of interest, and the defect features of the image of the end face of the hot-rolled steel coil in the method for detecting the quality of the end face of the hot-rolled steel coil according to the embodiment of the present invention.
Detailed Description
In order to make the present application more clearly understood by those skilled in the art to which the present application pertains, the following detailed description of the present application is made with reference to the accompanying drawings by way of specific embodiments.
The invention provides a method for detecting the end surface quality of a hot-rolled steel coil, which comprises the following steps of:
s1, acquiring an image of the end face of the hot-rolled steel coil shot by an industrial camera, wherein the industrial camera is distributed around the end face of the hot-rolled steel coil; the image corresponding to the end face of each hot rolled steel coil is shot by two industrial cameras;
s2, eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
s3, searching and cutting the edge of the hot-rolled steel coil end face image by utilizing a gray gradient threshold search algorithm, and extracting an image of a region to be detected;
s4, carrying out illumination balance processing on the image of the area to be detected, improving the brightness of the image of the area to be detected, and obtaining the image after light supplement;
s5, detecting the defects of the images after the light supplement by using a Kirsch operator, and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain preprocessed images;
s6, extracting primary features from the preprocessed image by using a Feature extraction layer (VFE), extracting a Region Of Interest (ROI) from the primary features by using a Region network candidate layer (RPN), processing the ROI by using a convolutional neural network fast Region calibration layer (Faster R-CNN), extracting defect features in the hot-rolled steel coil end face image, and outputting the defect features.
In another embodiment, after step S4, a median filter may be used to remove noise from the padded image.
The method provided by the invention relates to the identification of quality defects (including edge loss, folding, riffled lines and the like) and tower-shaped defects of the end surface of the steel coil, and is applied to the online detection of the quality of the steel coil on a hot-rolled plate production line.
Because the steel coil has larger size, a single industrial camera (CCD camera) is used for shooting the image of the end of the whole steel coil, and the required resolution cannot be achieved; and when a single industrial camera or more than three industrial cameras are used for shooting for many times, the problem of image splicing is caused, and the image acquisition efficiency and accuracy are reduced. Multiple shots taken with more than three industrial cameras increase the image data to be processed, thus reducing the final image acquisition efficiency. When a single industrial camera performs shooting, the obtained image data is insufficient, and thus the image accuracy is reduced.
In the method provided by the embodiment of the invention, when the end face of the steel coil is shot, 6 three light sources are fixedly arranged on each end face of the steel coil, as shown in fig. 2a, the 6 light sources are divided into an upper position, a middle position and a lower position, 2 light sources are arranged at each position, the emergent light of the upper light source and the emergent light of the lower light source form an included angle of 45 degrees with the end face, and the emergent light of the middle light source is perpendicular to the end face. Every terminal surface adopts two pixels to shoot for 2500 ten thousand industrial camera, and two cameras separately set up from top to bottom, are surrounded by 6 light sources and lie in same vertical line, as shown in fig. 2a and 2b, every camera shoots twice image, carries out the concatenation processing with the image of shooing again, can avoid reducing the problem of image acquisition efficiency and precision. The area of the end face of the steel coil shot by the camera is shown in fig. 2 c.
Noise sources on a steel coil production line mainly come from external environments (mechanical vibration, illumination conditions, electromagnetic interference, high-temperature atmospheric disturbance and the like), internal noises (electrical noise and thermal noise) of a CCD camera and an image cable transmission process, so that the noise type of a steel coil image is complex. In addition, since the end face image of the steel coil is a texture image containing more detailed information, the two-dimensional image generated by the optical imaging system contains various random noises and distortions. In order to eliminate irrelevant information in an image and recover useful real information, the invention adopts a median filtering method to remove mixed noise, and adopts a 3x3 template as a window of the median filtering, so that isolated noise points can be basically filtered.
As shown in fig. 3, step S3 includes:
performing X-direction convolution on the end face image of the hot-rolled steel coil by using a Sobel operator to obtain a first Sobel result graph, and performing Y-direction convolution on the end face image of the hot-rolled steel coil by using the Sobel operator to obtain a second Sobel result graph;
and performing column projection on the first Sobel result graph to obtain a column projection matrix, searching left and right boundaries of a steel coil in the end face image of the hot-rolled steel coil from left and right initial positions of the column projection matrix, cutting after the left and right boundaries of the steel coil are searched, performing column projection on the second Sobel result graph to obtain a row projection matrix, searching upper and lower boundaries of the steel coil in the end face image of the hot-rolled steel coil from upper and lower initial positions of the row projection matrix, and cutting after the upper and lower boundaries of the steel coil are searched to obtain an image of the area to be detected.
Sobel is a classical edge detection operator, mainly used for edge detection, and is a first-order discrete difference operator, which is used for calculating a gray scale approximation value of an image brightness function, and has two convolution templates, i.e. two groups of matrices of 3X3, which respectively correspond to an X direction (Gx) and a Y direction (Gy), wherein the templates are as follows:
the image is convolved with the image in a plane to obtain the horizontal and vertical brightness difference approximate values respectively. As can be seen from the template, the X-direction template is sensitive to the gray level mutation in the X direction, so that the vertical edge can be well detected, and a relatively obvious steel coil area can be obtained after the convolution result is binarized, which is more robust than the simple gray level threshold judgment.
After a Sobel result graph is obtained, column projection is carried out on the Sobel result graph, then searching is carried out from the left side and the right side to the middle side respectively, because a steel coil is generally positioned at the center position of an image, in order to avoid misjudgment caused by interference factors, the left and the right initial positions of the searching are determined at GrayScale X, the value is a configurable parameter, certain left and right areas are skipped according to the steel coil with different diameters, when sudden change occurs in a projection matrix, the edge point of the steel coil is reached, the position is used as the left (right) edge to carry out cutting, the upper and the lower boundaries are searched in the same way, but no area is skipped when the upper and the lower boundaries are searched, and when the row is detected to have more than GrayScale Y effective pixel points, the upper (.
And after the cutting is finished, if the width of the region to be detected is smaller than DropWidth or the height of the region to be detected is smaller than DropHeight, the image is considered invalid, and the detection is skipped. The edge is a place where the image gray scale change is severe, high values are generated by differentiating the place where the gray scale change is large, and Sobel operators are adopted in the digital image to carry out edge detection so as to obtain relatively prominent edge features. The original simulation image corresponding to the end face image of the hot-rolled steel coil is shown in fig. 4a, and the edge detection image of the Sobel operator is shown in fig. 4 b.
Step S3 further includes:
and judging whether the actual height value of the image of the area to be detected is smaller than a preset height value or not, or whether the actual width value of the image of the area to be detected is smaller than a preset width value or not, and if the actual height value of the image of the area to be detected is smaller than the preset height value or the actual width value of the image of the area to be detected is smaller than the preset width value, determining that the image of the end face of the hot-rolled steel coil is invalid.
Most of steel coil images shot by an industrial camera are located in the center of the images, and black backgrounds are arranged on two sides of the images, so that the first step of the algorithm is to extract an effective region to be detected and remove a background region. Due to the fact that the surface temperature of the steel coil is high, oil stain particles are doped in the air, and the field environment is complex, the quality of images shot by a camera is uneven, under the condition, the traditional simple gray level threshold value is used for judging that the boundary effect is unstable, the background misjudgment is possibly caused, and the subsequent detection process is misdetected, so that the gray level gradient threshold value searching algorithm is used in the method.
As shown in fig. 5, step S4 includes:
performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the image of the area to be detected by a compensation coefficient C corresponding to the row of pixel pointsiObtaining a light-supplemented image;
wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the image of the area to be detected, and j is more than or equal to 1 and less than or equal to A. Typically, a may be 128, which is a summary of the results obtained in trial and error. When A is 128, the variation difference between the columns of the image pixel points of the area to be detected is not too large, the brightness of the image is obviously improved on the whole, the details are well reserved, and then the original image is compensated according to the columns.
Because coil of strip diameter variation range is great, and the site environment is noisy, and the imaging effect difference of different specifications or different cameras is great, direct presentation is the inhomogeneous of image brightness, for guaranteeing the testing result, needs to carry out illumination equilibrium to it, promotes whole luminance.
Extracting primary features from the pre-processed image using a feature extraction layer, as shown in fig. 6, includes:
and processing the preprocessed image through double-layer sparse filtering and a VGG-16 network in sequence, and extracting primary features. The VGG in the VGG-16 Network is a visual Geometry Group Network (Geometry Group Network), and 16 indicates that the VGG Network structure has 13 convolutional layers and 3 full link layers.
Extracting the region of interest from the primary features by using the regional network candidate layer, comprising:
sequentially carrying out convolution processing on the primary features by 5 multiplied by 5 and convolution processing by 1 multiplied by 1 to obtain a feature map;
convolving the characteristic diagram through a K-dimensional sliding window to obtain a 2 xK classification result and a 4 xK classification result, wherein K is more than or equal to 1; here K may be 1,4,9,16, etc., each point of the feature map is mapped back to the central point of the receptive field of the original image as a reference point, and then K anchors (anchors) of different sizes and proportions are selected around this reference point.
And comparing the 2 xK classification result with the 4 xK classification result to perform sample screening to obtain the region of interest.
Utilize the quick regional calibration layer of convolution neural network to handle the region of interest, extract the defect characteristic in the hot rolling coil of strip terminal surface image, include:
performing pooling treatment on the region of interest; the pooling process is mainly used for feature dimension reduction, data and parameter quantity compression, overfitting reduction and model fault tolerance improvement;
inputting the pooling treatment result of the region of interest into the full-connection layer for treatment; after convolution and pooling for a plurality of times, the model can completely connect learned end face characteristics of a high-quality steel coil and send an output value to a classifier (such as a softmax classifier) when arriving at an output layer;
and performing regression and classification processing on the processing result of the full connection layer to obtain a position offset and a probability vector, and obtaining defect characteristics in the end face image of the hot-rolled steel coil through the position offset and the probability vector. The classification is to obtain the score of each class of the sample through a full connecting layer after a series of convolution layers and pooling layers, and then classify the sample by using, for example, softmax classification; regression is equivalent to framing the object to be identified with a rectangular box.
The key point of the steel coil defect image identification is the realization of a feature extraction algorithm of a target image. The invention uses the Faster R-CNN to extract the characteristics in the steel coil image, completes the identification and marking of the defect types, and the structure is shown in figure 6. The steel coil defect identification detection model comprises three parts, namely a Feature extraction layer VFE (visual Feature extraction), a region network candidate layer RPN (region probable network) and a fast region calibration layer based on a convolutional neural network. Wherein, the sparse filtering layer in the VFE extracts primary features as an input layer of a convolution part, the RPN layer preliminarily extracts an interested region, and finally, position marking and hidden type identification are carried out on a Fast R-CNN layer. In the preprocessing stage, a region planning method is provided to roughly cut out defect bodies so as to avoid generating a large number of redundant windows, thereby improving the detection speed and precision. The algorithm provided by the invention is combined with a data expansion method to increase the number of images, and the robustness of the algorithm is improved by dividing K fold intersection and verifying a data set; meanwhile, the sparse filtering idea is integrated into the convolutional neural network, and the dual depth features are extracted to be used as the input of the Faster R-CNN, so that the accuracy of the position detection and identification of the Faster R-CNN is improved.
After the extraction of the defect characteristics is completed, the defect types are identified and marked, indexes such as the type of the detected defect, the coordinates of the upper left corner of the defect area, the defect width, the defect height, the defect shape, the number of pixels occupied by the defect and the like are output, and a document which can be downloaded and stored is formed.
The invention also provides a system for detecting the end surface quality of the hot-rolled steel coil, which corresponds to the method for detecting the end surface quality of the hot-rolled steel coil, and comprises the following steps: the device comprises an image acquisition module, a first denoising module, an edge cutting module, an image light supplementing module, a second denoising module and a defect feature extraction module.
The image acquisition module is used for acquiring the end face image of the hot-rolled steel coil shot by the industrial camera, and the industrial camera is distributed around the end face of the hot-rolled steel coil. And the image corresponding to the end face of each hot rolled steel coil is shot by two industrial cameras.
The first denoising module is used for eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
and the edge cutting module is used for searching the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold value searching algorithm and cutting and extracting the image of the area to be detected.
The image light supplementing module is used for carrying out illumination balance processing on the image of the area to be detected, improving the brightness of the image of the area to be detected and obtaining the image after light supplementing.
And the second denoising module is used for detecting the defects of the images after the light supplement by using a Kirsch operator, and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain the preprocessed images.
The defect feature extraction module is used for extracting primary features from the preprocessed image by using the feature extraction layer, extracting an interested region from the primary features by using the regional network candidate layer, processing the interested region by using the convolutional neural network rapid regional calibration layer, extracting defect features in the image of the end face of the hot-rolled steel coil, and outputting the defect features.
The edge cropping module comprises: a convolution operation unit and a clipping unit.
The convolution operation unit is used for performing X-direction convolution on the end face image of the hot-rolled steel coil to obtain a first Sobel result graph, and performing Y-direction convolution on the end face image of the hot-rolled steel coil to obtain a second Sobel result graph.
The cutting unit is used for performing column projection on the first Sobel result graph to obtain a column projection matrix, searching left and right boundaries of a steel coil in the hot-rolled steel coil end surface image from left and right initial positions of the column projection matrix, cutting after the left and right boundaries of the steel coil are searched, performing column projection on the second Sobel result graph to obtain a row projection matrix, searching upper and lower boundaries of the steel coil in the hot-rolled steel coil end surface image from upper and lower initial positions of the row projection matrix, and cutting after the upper and lower boundaries of the steel coil are searched to obtain an image of the area to be detected.
The edge cropping module further comprises a judging unit. The judging unit is used for judging whether the actual height value of the image of the area to be detected is smaller than the preset height value or not, or whether the actual width value of the image of the area to be detected is smaller than the preset width value or not, and if the actual height value of the image of the area to be detected is smaller than the preset height value or the actual width value of the image of the area to be detected is smaller than the preset width value, determining that the image of the end face of the hot-rolled steel coil is.
The image light supplement module is used for: performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the image of the area to be detected by a compensation coefficient C corresponding to the row of pixel pointsiAnd obtaining the image after light supplement.
Wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the image of the area to be detected, and j is more than or equal to 1 and less than or equal to A.
Preferably, the defect feature extraction module includes a primary feature extraction unit, a region of interest extraction unit, and a defect feature extraction unit.
The primary feature extraction unit is used for processing the preprocessed image through double-layer sparse filtering and a VGG-16 network in sequence and extracting primary features.
The region-of-interest extraction unit is used for sequentially carrying out convolution processing on the primary features by 5 multiplied by 5 and convolution processing by 1 multiplied by 1 to obtain a feature map; convolving the characteristic diagram through a K-dimensional sliding window to obtain a 2 xK classification result and a 4 xK classification result, wherein K is more than or equal to 1; and comparing the 2 xK classification result with the 4 xK classification result to perform sample screening to obtain the region of interest.
The defect feature extraction unit is used for performing pooling treatment on the region of interest; inputting the pooling treatment result of the region of interest into the full-connection layer for treatment; and performing regression and classification processing on the processing result of the full connection layer to obtain a position offset and a probability vector, and obtaining defect characteristics in the end face image of the hot-rolled steel coil through the position offset and the probability vector.
In conclusion, the method provided by the invention collects the images of the end faces of the steel coils, the images corresponding to the end faces of each hot rolled steel coil are obtained by shooting through two industrial cameras, the image collection efficiency and the image collection precision can be improved, the median filter is utilized to carry out weighted average processing on the whole image, the noise of the image is reduced, irrelevant information in the image can be eliminated, useful real information is recovered, then the definition of the edge of the area to be detected is completed by adopting a gray gradient threshold value search algorithm, the illumination equalization algorithm is utilized to carry out optical compensation on the area to be detected, the brightness of the area to be detected is improved, the Kirsch operator is utilized to detect the defects after the completion of the optical compensation, and the noise points of the defects after the Ki. And extracting primary features from the preprocessed image by using a feature extraction layer, extracting an interested region from the primary features by using a regional network candidate layer, extracting double-depth features as input of the Faster R-CNN, and finally extracting defect features in the steel coil image by using the Faster R-CNN to finish the identification and marking of defect types, thereby improving the precision of the detection and identification of the position of the Faster R-CNN, outputting indexes such as the type of the detected defect, the upper left corner coordinate of the defect region, the defect width, the defect height, the defect shape, the number of pixels occupied by the defect and the like, and forming a document which can be downloaded and stored. The invention can obtain the defect detection image of the end face of the steel coil with the detection rate of more than 90 percent in the environments of high temperature, high speed and the like.
While preferred embodiments of the present invention have been described, additional variations and modifications in those embodiments may occur to those skilled in the art once they learn of the basic inventive concepts. Therefore, it is intended that the appended claims be interpreted as including preferred embodiments and all such alterations and modifications as fall within the scope of the invention.
It will be apparent to those skilled in the art that various changes and modifications may be made in the present invention without departing from the spirit and scope of the invention. Thus, if such modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.
Claims (10)
1. A method for detecting the end surface quality of a hot-rolled steel coil is characterized by comprising the following steps:
acquiring an end face image of a hot-rolled steel coil shot by an industrial camera; the image corresponding to the end face of each hot-rolled steel coil is shot by two industrial cameras;
eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
searching and cutting the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold search algorithm, and extracting an image of a region to be detected;
carrying out illumination equalization processing on the image of the area to be detected, improving the brightness of the image of the area to be detected, and obtaining a supplemented image;
detecting the defects of the images after the light supplement by using a Kirsch operator, and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain preprocessed images;
extracting primary features from the preprocessed image by using a feature extraction layer, extracting an interested region from the primary features by using a regional network candidate layer, processing the interested region by using a convolutional neural network rapid regional calibration layer, extracting defect features in the image of the end face of the hot-rolled steel coil, and outputting the defect features.
2. The method for detecting the end face quality of the hot-rolled steel coil according to claim 1, wherein the step of searching and cutting the edge of the image of the end face of the hot-rolled steel coil by using a gray gradient threshold search algorithm to extract the image of the area to be detected comprises the following steps:
performing X-direction convolution on the end face image of the hot-rolled steel coil by using a Sobel operator to obtain a first Sobel result graph, and performing Y-direction convolution on the end face image of the hot-rolled steel coil by using the Sobel operator to obtain a second Sobel result graph;
and performing column projection on the first Sobel result graph to obtain a column projection matrix, searching left and right boundaries of a steel coil in the hot-rolled steel coil end surface image from left and right initial positions of the column projection matrix, cutting after the left and right boundaries of the steel coil are searched, performing column projection on the second Sobel result graph to obtain a row projection matrix, searching upper and lower boundaries of the steel coil in the hot-rolled steel coil end surface image from upper and lower initial positions of the row projection matrix, and cutting after the upper and lower boundaries of the steel coil are searched to obtain the to-be-detected region image.
3. The method for detecting the end face quality of the hot-rolled steel coil according to claim 2, wherein the method for searching and cutting the edge of the image of the end face of the hot-rolled steel coil by using the gray gradient threshold search algorithm to extract the image of the area to be detected further comprises:
and judging whether the actual height value of the image of the area to be detected is smaller than a preset height value or not, or whether the actual width value of the image of the area to be detected is smaller than a preset width value or not, and if the actual height value of the image of the area to be detected is smaller than the preset height value or the actual width value of the image of the area to be detected is smaller than the preset width value, determining that the image of the end face of the hot-rolled steel coil is invalid.
4. The method for detecting the end face quality of the hot-rolled steel coil according to claim 1, wherein the step of performing illumination equalization processing on the image of the area to be detected to improve the brightness of the image of the area to be detected and obtain the image after light supplement comprises the steps of:
performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the to-be-detected region image by a compensation coefficient C corresponding to the row of pixel pointsiObtaining the image after the light supplement;
wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the to-be-detected region image, and j is more than or equal to 1 and less than or equal to A.
5. The method for detecting the end face quality of the hot-rolled steel coil as claimed in claim 1, wherein the extracting primary features from the preprocessed image by using a feature extraction layer comprises:
and processing the preprocessed image sequentially through double-layer sparse filtering and a VGG-16 network, and extracting the primary features.
6. The method for detecting the end face quality of the hot-rolled steel coil as claimed in claim 1, wherein the extracting the region of interest from the primary features by using the local area network candidate layer comprises:
sequentially carrying out convolution processing on the primary features by 5 multiplied by 5 and convolution processing by 1 multiplied by 1 to obtain a feature map;
convolving the characteristic diagram through a K-dimensional sliding window to obtain a 2 xK classification result and a 4 xK classification result, wherein K is more than or equal to 1;
and comparing the 2 xK classification result with the 4 xK classification result to perform sample screening to obtain the region of interest.
7. The method for detecting the end face quality of the hot-rolled steel coil according to claim 1, wherein the step of processing the region of interest by using a convolutional neural network fast region calibration layer to extract defect features in the image of the end face of the hot-rolled steel coil comprises the following steps:
performing pooling treatment on the region of interest;
inputting the pooling treatment result of the region of interest into a full-connection layer for treatment;
and performing regression and classification processing on the processing result of the full connection layer to obtain a position offset and a probability vector, and obtaining defect characteristics in the end face image of the hot-rolled steel coil according to the position offset and the probability vector.
8. The utility model provides a hot rolling coil of strip terminal surface quality detection system which characterized in that includes:
the image acquisition module is used for acquiring an end face image of the hot-rolled steel coil shot by the industrial camera; the image corresponding to the end face of each hot-rolled steel coil is shot by two industrial cameras;
the first denoising module is used for eliminating the noise of the image of the end face of the hot-rolled steel coil by using a median filter;
the edge cutting module is used for searching and cutting the edge of the image of the end face of the hot-rolled steel coil by utilizing a gray gradient threshold value searching algorithm and extracting an image of a region to be detected;
the image light supplement module is used for carrying out illumination balance processing on the image of the area to be detected, improving the brightness of the image of the area to be detected and obtaining a supplemented image;
the second denoising module is used for detecting the defects of the images after the light supplement by using a Kirsch operator and eliminating noise points of the defects detected by the Kirsch operator by using morphological operation to obtain preprocessed images;
and the defect feature extraction module is used for extracting primary features from the preprocessed image by using the feature extraction layer, extracting an interested region from the primary features by using the regional network candidate layer, processing the interested region by using the convolutional neural network rapid region calibration layer, extracting defect features in the hot-rolled steel coil end face image and outputting the defect features.
9. The system for detecting the end face quality of the hot-rolled steel coil as claimed in claim 8, wherein the edge cutting module comprises:
the convolution operation unit is used for performing X-direction convolution on the end face image of the hot-rolled steel coil by using a Sobel operator to obtain a first Sobel result graph and performing Y-direction convolution on the end face image of the hot-rolled steel coil by using the Sobel operator to obtain a second Sobel result graph;
and the cutting unit is used for performing column projection on the first Sobel result graph to obtain a column projection matrix, starting to search the left and right boundaries of the steel coil in the hot-rolled steel coil end surface image from the left and right initial positions of the column projection matrix, cutting the left and right boundaries of the steel coil after the left and right boundaries of the steel coil are searched, performing column projection on the second Sobel result graph to obtain a row projection matrix, starting to search the upper and lower boundaries of the steel coil in the hot-rolled steel coil end surface image from the upper and lower initial positions of the row projection matrix, and cutting the upper and lower boundaries of the steel coil after the upper and lower boundaries of the steel coil are searched to obtain the to-be-detected region image.
10. The system for detecting the end surface quality of the hot-rolled steel coil according to claim 8, wherein the image light supplement module is configured to:
performing column projection on the image of the area to be detected, and calculating the average gray value G of each column projection of the image of the area to be detectedjAnd according to the average gray value GjCalculating a compensation coefficient CiMultiplying each row of pixel points of the to-be-detected region image by a compensation coefficient C corresponding to the row of pixel pointsiTo obtain the fill lightA post image;
wherein the compensation coefficient CiThe calculation formula of (2) is as follows:
a is the number of pixel point lines of the to-be-detected region image, and j is more than or equal to 1 and less than or equal to A.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010997054.5A CN112233067A (en) | 2020-09-21 | 2020-09-21 | Hot rolled steel coil end face quality detection method and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010997054.5A CN112233067A (en) | 2020-09-21 | 2020-09-21 | Hot rolled steel coil end face quality detection method and system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN112233067A true CN112233067A (en) | 2021-01-15 |
Family
ID=74108470
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010997054.5A Pending CN112233067A (en) | 2020-09-21 | 2020-09-21 | Hot rolled steel coil end face quality detection method and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112233067A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884797A (en) * | 2021-02-02 | 2021-06-01 | 武汉钢铁有限公司 | Image background removing method and device and electronic equipment |
| CN113240628A (en) * | 2021-04-14 | 2021-08-10 | 首钢集团有限公司 | Method, device and system for judging quality of steel coil |
| CN113269759A (en) * | 2021-05-28 | 2021-08-17 | 中冶赛迪重庆信息技术有限公司 | Steel coil information detection method, system, medium and terminal based on image recognition |
| CN113284115A (en) * | 2021-05-28 | 2021-08-20 | 中冶赛迪重庆信息技术有限公司 | Steel coil tower shape identification method, system, medium and terminal |
| CN113465511A (en) * | 2021-06-19 | 2021-10-01 | 精锐视觉智能科技(上海)有限公司 | Online measurement and omnibearing end surface defect online detection method for steel coil size |
| CN113554628A (en) * | 2021-07-27 | 2021-10-26 | 苏州微景医学科技有限公司 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN113989257A (en) * | 2021-11-09 | 2022-01-28 | 国网江苏省电力有限公司南通供电分公司 | Electric power comprehensive pipe gallery settlement crack identification method based on artificial intelligence technology |
| CN114693684A (en) * | 2022-06-01 | 2022-07-01 | 领伟创新智能系统(浙江)有限公司 | Airborne fan blade defect detection method |
| CN114792369A (en) * | 2022-06-29 | 2022-07-26 | 上海启迪睿视智能科技有限公司 | Cigarette carton filling state detection method and system based on light projection |
| CN117115082A (en) * | 2023-07-12 | 2023-11-24 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108629775A (en) * | 2018-05-14 | 2018-10-09 | 华中科技大学 | A kind of hot high-speed rod surface image processing method |
| CN110276754A (en) * | 2019-06-21 | 2019-09-24 | 厦门大学 | A kind of detection method of surface flaw, terminal device and storage medium |
-
2020
- 2020-09-21 CN CN202010997054.5A patent/CN112233067A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108629775A (en) * | 2018-05-14 | 2018-10-09 | 华中科技大学 | A kind of hot high-speed rod surface image processing method |
| CN110276754A (en) * | 2019-06-21 | 2019-09-24 | 厦门大学 | A kind of detection method of surface flaw, terminal device and storage medium |
Non-Patent Citations (2)
| Title |
|---|
| 李宜汀，等: "基于 Faster R-CNN 的表面缺陷检测方法研究", 《计算机集成制造系统》 * |
| 管声启等: "基于图像预处理的神经网络带钢缺陷检测", 《钢铁研究》 * |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112884797A (en) * | 2021-02-02 | 2021-06-01 | 武汉钢铁有限公司 | Image background removing method and device and electronic equipment |
| CN112884797B (en) * | 2021-02-02 | 2023-12-08 | 武汉钢铁有限公司 | Image background removing method and device and electronic equipment |
| CN113240628A (en) * | 2021-04-14 | 2021-08-10 | 首钢集团有限公司 | Method, device and system for judging quality of steel coil |
| CN113269759A (en) * | 2021-05-28 | 2021-08-17 | 中冶赛迪重庆信息技术有限公司 | Steel coil information detection method, system, medium and terminal based on image recognition |
| CN113284115A (en) * | 2021-05-28 | 2021-08-20 | 中冶赛迪重庆信息技术有限公司 | Steel coil tower shape identification method, system, medium and terminal |
| CN113465511B (en) * | 2021-06-19 | 2022-12-13 | 精锐视觉智能科技(上海)有限公司 | Steel coil size online measurement and omnibearing end surface defect online detection method |
| CN113465511A (en) * | 2021-06-19 | 2021-10-01 | 精锐视觉智能科技(上海)有限公司 | Online measurement and omnibearing end surface defect online detection method for steel coil size |
| CN113554628A (en) * | 2021-07-27 | 2021-10-26 | 苏州微景医学科技有限公司 | Image processing method, image processing apparatus, and computer-readable storage medium |
| CN113989257A (en) * | 2021-11-09 | 2022-01-28 | 国网江苏省电力有限公司南通供电分公司 | Electric power comprehensive pipe gallery settlement crack identification method based on artificial intelligence technology |
| CN114693684A (en) * | 2022-06-01 | 2022-07-01 | 领伟创新智能系统(浙江)有限公司 | Airborne fan blade defect detection method |
| CN114792369A (en) * | 2022-06-29 | 2022-07-26 | 上海启迪睿视智能科技有限公司 | Cigarette carton filling state detection method and system based on light projection |
| CN117115082A (en) * | 2023-07-12 | 2023-11-24 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
| CN117115082B (en) * | 2023-07-12 | 2024-04-05 | 钛玛科（北京）工业科技有限公司 | Method and equipment for detecting overlap quality of tire |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112233067A (en) | Hot rolled steel coil end face quality detection method and system | |
| CN111325713B (en) | Neural network-based wood defect detection method, system and storage medium | |
| CN106875373B (en) | Mobile phone screen MURA defect detection method based on convolutional neural network pruning algorithm | |
| CN111223093A (en) | AOI defect detection method | |
| CN111402226A (en) | Surface defect detection method based on cascade convolution neural network | |
| CN109034184B (en) | Grading ring detection and identification method based on deep learning | |
| CN111383209A (en) | Unsupervised flaw detection method based on full convolution self-encoder network | |
| CN111915704A (en) | Apple hierarchical identification method based on deep learning | |
| CN113554631B (en) | Chip surface defect detection method based on improved network | |
| CN106355579A (en) | Defect detecting method of cigarette carton surface wrinkles | |
| CN111368825B (en) | Pointer positioning method based on semantic segmentation | |
| CN112348787A (en) | Training method of object defect detection model, object defect detection method and device | |
| CN112132196B (en) | Cigarette case defect identification method combining deep learning and image processing | |
| CN113240626A (en) | Neural network-based method for detecting and classifying concave-convex flaws of glass cover plate | |
| CN111127417B (en) | Printing defect detection method based on SIFT feature matching and SSD algorithm improvement | |
| CN112200790B (en) | Cloth defect detection method, device and medium | |
| CN115830004A (en) | Surface defect detection method, device, computer equipment and storage medium | |
| CN114255212A (en) | FPC surface defect detection method and system based on CNN | |
| CN104484679B (en) | Non- standard rifle shooting warhead mark image automatic identifying method | |
| CN115511775A (en) | Light-weight ceramic tile surface defect detection method based on semantic segmentation | |
| CN114065798A (en) | Visual identification method and device based on machine identification | |
| CN117197682A (en) | Method for blind pixel detection and removal by long-wave infrared remote sensing image | |
| CN112750113B (en) | Glass bottle defect detection method and device based on deep learning and linear detection | |
| CN107123105A (en) | Images match defect inspection method based on FAST algorithms | |
| CN114092441A (en) | Product surface defect detection method and system based on dual neural network |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20210115 |
|
| RJ01 | Rejection of invention patent application after publication |