CN114414578A - Industrial hole wall defect detection system and identification algorithm based on AI - Google Patents
Industrial hole wall defect detection system and identification algorithm based on AI Download PDFInfo
- Publication number
- CN114414578A CN114414578A CN202210064160.7A CN202210064160A CN114414578A CN 114414578 A CN114414578 A CN 114414578A CN 202210064160 A CN202210064160 A CN 202210064160A CN 114414578 A CN114414578 A CN 114414578A
- Authority
- CN
- China
- Prior art keywords
- unit
- defect
- image
- defect detection
- loss function
- 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.)
- Withdrawn
Links
- 230000007547 defect Effects 0.000 title claims abstract description 96
- 238000001514 detection method Methods 0.000 title claims abstract description 57
- 238000000605 extraction Methods 0.000 claims abstract description 22
- 238000007781 pre-processing Methods 0.000 claims abstract description 19
- 238000012549 training Methods 0.000 claims abstract description 10
- 238000012360 testing method Methods 0.000 claims abstract description 7
- 230000006870 function Effects 0.000 claims description 39
- 238000013473 artificial intelligence Methods 0.000 claims description 20
- 238000012545 processing Methods 0.000 claims description 15
- 230000004927 fusion Effects 0.000 claims description 14
- 230000005540 biological transmission Effects 0.000 claims description 12
- 230000003287 optical effect Effects 0.000 claims description 12
- 238000000034 method Methods 0.000 claims description 11
- 238000004364 calculation method Methods 0.000 claims description 6
- 239000011148 porous material Substances 0.000 claims description 5
- 238000010606 normalization Methods 0.000 claims description 4
- 238000011478 gradient descent method Methods 0.000 claims description 3
- 238000013519 translation Methods 0.000 claims description 3
- 230000000007 visual effect Effects 0.000 abstract description 2
- 230000036544 posture Effects 0.000 description 9
- 238000010586 diagram Methods 0.000 description 3
- 230000003321 amplification Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 238000006073 displacement reaction Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000003199 nucleic acid amplification method Methods 0.000 description 2
- 238000011084 recovery Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/95—Investigating the presence of flaws or contamination characterised by the material or shape of the object to be examined
- G01N21/954—Inspecting the inner surface of hollow bodies, e.g. bores
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/253—Fusion techniques of extracted features
-
- 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
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T1/00—General purpose image data processing
- G06T1/0014—Image feed-back for automatic industrial control, e.g. robot with camera
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/60—Analysis of geometric attributes
- G06T7/66—Analysis of geometric attributes of image moments or centre of gravity
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8854—Grading and classifying of flaws
- G01N2021/8874—Taking dimensions of defect into account
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01N—INVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
- G01N21/00—Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
- G01N21/84—Systems specially adapted for particular applications
- G01N21/88—Investigating the presence of flaws or contamination
- G01N21/8851—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges
- G01N2021/8887—Scan or image signal processing specially adapted therefor, e.g. for scan signal adjustment, for detecting different kinds of defects, for compensating for structures, markings, edges based on image processing techniques
Abstract
The invention relates to the technical field of defect detection, in particular to an industrial hole wall defect detection system and an identification algorithm based on AI. The system comprises a data acquisition unit, a data preprocessing unit, a feature extraction unit, a defect detection unit and a loss function unit; the data preprocessing unit is used for amplifying the image signals acquired by the data acquisition unit and subtracting the mean value from the images in the training set and the test set, and the feature extraction unit is used for optimizing the weight coefficient and the bias of each node of the image; the defect detection unit is used for determining the central point of the detected defect position according to the image characteristics in the characteristic extraction unit; according to the invention, the position and the type of the defect can be obtained only by converting the picture into an image signal, the visual observation is liberated, the detection standard is unified, the consistency of the product quality is ensured, and the detection speed is increased, so that the detection speed of the defect on the inner wall of the industrial product is increased.
Description
Technical Field
The invention relates to the technical field of defect detection, in particular to an industrial hole wall defect detection system and an identification algorithm based on AI.
Background
At present, most of the defects of the inner wall of the hole of the industrial product depend on manpower, the detection speed of the scheme is slow, meanwhile, the detection result is influenced by the experience and skill level of inspectors and some subjective factors, the consistency and the reliability are lacked, the traditional technology for detecting the defects of the inner wall of the industrial product is mainly checked in an inclined way through a microscope with super depth of field, the method is difficult to realize automation, generates image distortion, has low detection efficiency and high detection error rate, in addition, the method for detecting the defects of the inner wall of the industrial product by utilizing the deep learning method mainly utilizes a two-stage target detection algorithm, the method firstly generates a large number of candidate pictures, then uses the convolutional neural network to classify and regress the pictures, the method is time-consuming, and in view of the above, an industrial hole wall defect detection system and an identification algorithm based on AI are provided.
Disclosure of Invention
The invention aims to provide an industrial hole wall defect detection system and an industrial hole wall defect identification algorithm based on AI (artificial intelligence) so as to solve the problems in the background technology.
In order to achieve the above object, in one aspect, the present invention provides an AI-based industrial hole wall defect detection system, which includes a data acquisition unit, a data preprocessing unit, a feature extraction unit, a defect detection unit, and a loss function unit;
the data acquisition unit is used for converting a shot target into an image signal and transmitting the image signal to the image processing system, the image signal is converted into a digital signal by adopting artificial intelligence according to information such as pixel distribution, brightness and color, the image processing system performs operation on the digital signal to extract the characteristics of the target, and then a detection result is obtained or feedback control is realized;
the data preprocessing unit is used for amplifying the image signals acquired by the data acquisition unit through rotation, translation, mirror image and picture brightness adjustment modes, increasing the robustness of the algorithm, and subtracting the mean value from the images in the training set and the test set to normalize the images;
the feature extraction unit is used for optimizing each node weight coefficient and bias of the image preprocessed by the data preprocessing unit, so that the network features are more suitable for specific projects, and the use precision of the network features is improved;
the defect detection unit is used for determining the center point of the detected defect position according to the image characteristics in the characteristic extraction unit, regressing the size and the offset of a defect frame at the center point position, and detecting the position and the type of the defect;
the loss function unit is used for optimizing and specifying the direction of the network parameters of the defect position in the defect detection unit.
As a further improvement of the technical scheme, the data acquisition unit comprises an optical information transmission module and an image acquisition module;
the optical information transmission module is used for realizing the leading-in of an external light source and the leading-out of an image inside the hole wall, the optical information transmission part comprises a camera, a lens, a light source, a video tube and the like, and during measurement, the video tube scans the measured hole wall under the driving of the electric control displacement guide rail;
the image acquisition module is used for converting a shot target into an image signal.
As a further improvement of the technical solution, an image expansion width calculation formula of the optical information transmission module is as follows:
s＝d*p*m
wherein d is the image development width, p is the pixel equivalent, m is the camera frame rate, and s is the advancing speed, and the configuration of key parameters and the research of influencing factors are very important, so that the coordination and the stability of the work can be improved.
As a further improvement of the technical solution, the feature extraction unit adopts a network learning weight algorithm, and calculates the formula as follows:
Yi＝αiX1i+βiX2i
Yicharacteristic of the fused ith channel, X1iFeatures of the ith channel in the first feature map before fusion, X2iFeatures of the ith channel, α, in the second feature map before fusioniAnd betaiThe weighting coefficients of different channels are obtained through automatic network learning, and the importance among different channels can be automatically learned by utilizing the strategy network, so that the precision is improved.
As a further improvement of the technical scheme, the defect detection unit comprises an acquisition module, a semantic feature acquisition module and a feature fusion module;
the acquisition module is used for expanding the spatial resolution;
the semantic feature acquisition module is used for extracting features and reducing the spatial resolution at the same time to acquire high-level semantic features;
the feature fusion module is used for fusing the feature layers of the upper half part under different resolutions while expanding the spatial resolution of the acquisition module, so that a defect central point heat map, central point position offset, size and defect type can be output.
As a further improvement of the technical solution, the acquisition module adopts a deconvolution mode, and a deconvolution calculation formula is as follows:
f*g＝h
where f is the expected recovery signal, g is the driving force, and h is the convolution.
As a further improvement of the technical solution, the feature fusion module adopts a network algorithm for fusing multi-layer features, and includes the following steps:
firstly, training a model by adopting a back propagation method, and updating network parameters by using a random gradient descent method;
secondly, setting an initial learning rate, finishing the initial learning rate for multiple times, and adjusting the learning rate for one time after iteration;
and thirdly, setting the quantity of the batch input network samples.
As a further improvement of the present technical solution, the loss function unit includes the following postures:
posture one: the central point location heat map loss, using Focal length as the loss function, is expressed as follows:
wherein alpha and beta are hyper-parameters, and the problem of class imbalance can be effectively solved through the hyper-parameters;
and (5) posture II: defect position center offset loss, adopting smooth L1 loss as a loss function;
posture three: the defect size loss function also adopts smooth L1 loss as a loss function, and the expression of the loss function is as follows:
and (4) posture IV: taking the bisection cross entropy as a loss function, the expression is as follows:
where y ∈ {0,1}, the loss function can be used to determine the type of defect.
On the other hand, the invention also provides an identification algorithm for detecting the industrial pore wall defect based on the AI, which comprises any one of the above-mentioned industrial pore wall defect detection systems based on the AI, and the operation steps are as follows:
s1, data acquisition: converting a shot target into an image signal by adopting a data acquisition unit, transmitting the image signal to an image processing system, and converting the image signal into a digital signal by adopting artificial intelligence according to information such as pixel distribution, brightness, color and the like;
s2, preprocessing data: adopting a data preprocessing unit to perform data amplification on the acquired image signals in a rotating, translating, mirroring and picture brightness adjusting mode, simultaneously adopting an image processing algorithm to subtract an average value from the images in the training set and the test set, and performing normalization processing on the images;
s3, feature extraction: the preprocessed image is subjected to optimization of each node weight coefficient and bias by a feature extraction unit and a network learning weight algorithm, so that the network features are more suitable for specific projects, and the use precision of the network features is improved;
s4, defect detection: the method comprises the steps that a defect detection unit is adopted, a network algorithm with the weight fused with multilayer features is adopted according to image features, the central point of a detected defect position is determined, the size and the offset of a defect frame are regressed at the central point position, the position and the type of the defect are detected, the convolution features under different resolutions are fused, the extracted features can be better used for identifying the type and the position of the defect, the defect is mainly regarded as one point, the defect is detected in a non-preset frame mode, and a model is end-to-end differentiable, so that the method is simpler, quicker and more accurate;
s5, determining a loss function: and judging a proper loss function by adopting the loss function unit, and optimizing the network parameter of the defect position to an appointed direction.
Compared with the prior art, the invention has the beneficial effects that:
1. in the industrial hole wall defect detection system and the identification algorithm based on AI, through the data acquisition unit and the data preprocessing unit, the position and the type of the defect can be obtained only by converting the picture into an image signal, the visual observation is liberated, the detection standard is unified, the consistency of the product quality is ensured, and the detection speed is improved, so that the detection speed of the defect on the inner wall of the industrial product is improved.
2. In the industrial pore wall defect detection system and the identification algorithm based on AI, by adopting the network algorithm with the right fused with the multilayer characteristics, a preset frame is not required to be set, and the model is end-to-end differentiable, simpler, faster and more accurate.
Drawings
FIG. 1 is a block diagram of AI-based industrial hole wall defect detection of example 1;
FIG. 2 is a block diagram of a data acquisition unit according to embodiment 1;
FIG. 3 is a block diagram of a defect detection unit of embodiment 1;
fig. 4 is a flow chart of the AI-based industrial hole wall defect detection identification algorithm of embodiment 1.
The various reference numbers in the figures mean:
100. a data acquisition unit; 110. an optical information transmission module; 120. an image acquisition module;
200. a data preprocessing unit;
300. a feature extraction unit;
400. a defect detection unit; 410. an acquisition module; 420. a semantic feature acquisition module; 430. a feature fusion module;
500. and a loss function unit.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
Referring to fig. 1-3, in one aspect, the present invention provides an AI-based industrial hole wall defect detection system, which includes a data acquisition unit 100, a data preprocessing unit 200, a feature extraction unit 300, a defect detection unit 400, and a loss function unit 500;
the data acquisition unit 100 is used for converting a shot target into an image signal and transmitting the image signal to an image processing system, the image signal is converted into a digital signal by adopting artificial intelligence according to information such as pixel distribution, brightness and color, and the like, and the image processing system performs operation on the digital signal to extract the characteristics of the target so as to obtain a detection result or realize feedback control;
the data preprocessing unit 200 is configured to perform data amplification on the image signals acquired by the data acquisition unit 100 in a manner of rotation, translation, mirroring, and picture brightness adjustment, increase robustness of an algorithm, and subtract an average value from the images in the training set and the test set, so as to perform normalization processing on the images;
the feature extraction unit 300 is configured to optimize each node weight coefficient and bias for the image preprocessed by the data preprocessing unit 200, so that the network features are more suitable for specific projects, and the use accuracy of the network features is improved;
the defect detection unit 400 is configured to determine a center point of a detected defect position according to the image features in the feature extraction unit 300, return the size and offset of a defect frame at the center point, and detect the position and type of the defect;
the loss function unit 500 is used to optimize the specified direction for the network parameters of the defect locations in the defect detection unit 400.
As a further improvement of the present technical solution, the data acquisition unit 100 includes an optical information transmission module 110 and an image acquisition module 120;
the optical information transmission module 110 is used for guiding an external light source in and guiding out an image inside the hole wall, the optical information transmission part comprises a camera, a lens, a light source, a video tube and the like, and during measurement, the video tube scans the measured hole wall under the driving of the electric control displacement guide rail;
the image acquisition module 120 is used for converting the captured object into an image signal.
In this embodiment, the calculation formula of the image expansion width of the optical information transmission module 110 is as follows:
s＝d*p*m
wherein d is the image development width, p is the pixel equivalent, m is the camera frame rate, and s is the advancing speed, and the configuration of key parameters and the research of influencing factors are very important, so that the coordination and the stability of the work can be improved.
It should be noted that the feature extraction unit 300 adopts a network learning weight algorithm, and calculates the formula as follows:
Yi＝αiX1i+βiX2i
Yicharacteristic of the fused ith channel, X1iFeatures of the ith channel in the first feature map before fusion, X2iFeatures of the ith channel, α, in the second feature map before fusioniAnd betaiThe weighting coefficients of different channels are obtained through automatic network learning, and the importance among different channels can be automatically learned by utilizing the strategy network, so that the precision is improved.
Further, the defect detection unit 400 includes an acquisition module 410, a semantic feature acquisition module 420, and a feature fusion module 430;
the acquisition module 410 is used to expand the spatial resolution;
the semantic feature obtaining module 420 is configured to extract features and reduce spatial resolution at the same time, so as to obtain high-level semantic features;
the feature fusion module 430 is configured to expand the spatial resolution of the acquisition module 410 and fuse feature layers of the upper half at different resolutions, so as to output a defect center point thermal map, a center point position offset, a size and a defect type.
Specifically, the acquisition module 410 adopts a deconvolution mode, and the deconvolution calculation formula is as follows:
f*g＝h
where f is the expected recovery signal, g is the driving force, and h is the convolution.
Further, the feature fusion module 430 adopts a network algorithm for fusing multi-layer features, and includes the following steps:
firstly, training a model by adopting a back propagation method, and updating network parameters by using a random gradient descent method;
secondly, setting an initial learning rate, finishing the initial learning rate for multiple times, and adjusting the learning rate for one time after iteration;
and thirdly, setting the quantity of the batch input network samples.
In addition, the loss function unit 500 includes the following poses:
posture one: the central point location heat map loss, using Focal length as the loss function, is expressed as follows:
wherein alpha and beta are hyper-parameters, and the problem of class imbalance can be effectively solved through the hyper-parameters;
and (5) posture II: defect position center offset loss, adopting smooth L1 loss as a loss function;
posture three: the defect size loss function also adopts smooth L1 loss as a loss function, and the expression of the loss function is as follows:
and (4) posture IV: taking the bisection cross entropy as a loss function, the expression is as follows:
where y ∈ {0,1}, the loss function can be used to determine the type of defect.
On the other hand, as shown in fig. 4, the present invention further provides an AI-based industrial hole wall defect detection recognition algorithm, which includes any one of the above-mentioned AI-based industrial hole wall defect detection systems, and the operation steps thereof are as follows:
s1, data acquisition: the data acquisition unit 100 is adopted to convert the shot target into an image signal, the image signal is transmitted to an image processing system, and the image signal is converted into a digital signal by adopting artificial intelligence according to information such as pixel distribution, brightness, color and the like;
s2, preprocessing data: the data preprocessing unit 200 is adopted to amplify the acquired image signals in a rotating, translating, mirroring and picture brightness adjusting mode, meanwhile, an image processing algorithm is adopted to subtract the average value of the images in the training set and the test set, and the images are subjected to normalization processing;
s3, feature extraction: the feature extraction unit 300 is adopted to optimize the weight coefficient and the bias of each node by adopting a network learning weight algorithm on the preprocessed image, so that the network features are more suitable for specific projects, and the use precision of the network features is improved;
s4, defect detection: the defect detection unit 400 is adopted, a network algorithm with the weight fused with multilayer features is adopted according to image features, the central point of the detected defect position is determined, the size and the offset of a defect frame are regressed at the central point position, the position and the type of the defect are detected, the convolution features under different resolutions are fused, the extracted features can be better used for identifying the type and the position of the defect, the defect is mainly regarded as one point, the defect is detected in a non-preset frame mode, and the model is end-to-end differentiable, so that the method is simpler, quicker and more accurate;
s5, determining a loss function: and judging a proper loss function by using the loss function unit 500, and optimizing the specified direction of the network parameters of the defect position.
The foregoing shows and describes the general principles, essential features, and advantages of the invention. It will be understood by those skilled in the art that the present invention is not limited to the embodiments described above, and the preferred embodiments of the present invention are described in the above embodiments and the description, and are not intended to limit the present invention. The scope of the invention is defined by the appended claims and equivalents thereof.
Claims (4)
1. An AI-based industrial pore wall defect detection system is characterized by comprising a data acquisition unit (100), a data preprocessing unit (200), a feature extraction unit (300), a defect detection unit (400) and a loss function unit (500);
the data acquisition unit (100) is used for converting a shot target into an image signal and transmitting the image signal to an image processing system;
the data preprocessing unit (200) is used for amplifying the image signals acquired by the data acquisition unit (100) through rotation, translation, mirror image and picture brightness adjustment modes, and subtracting the average value from the images in the training set and the test set to normalize the images;
the feature extraction unit (300) is used for optimizing each node weight coefficient and bias of the image preprocessed by the data preprocessing unit (200);
the defect detection unit (400) is used for determining the central point of the detected defect position according to the image characteristics in the characteristic extraction unit (300), regressing the size and the offset of a defect frame at the central point position, and detecting the position and the type of the defect;
the loss function unit (500) is used for optimizing a specified direction for network parameters of a defect position in the defect detection unit (400);
the data acquisition unit (100) comprises an optical information transmission module (110) and an image acquisition module (120);
the optical information transmission module (110) is used for realizing the leading-in of an external light source and the leading-out of an image inside the hole wall;
the image acquisition module (120) is used for converting a shot target into an image signal;
the image expansion width calculation formula of the optical information transmission module (110) is as follows:
s＝d*p*m
wherein d is the image expansion width, p is the pixel equivalent, m is the camera frame rate, and s is the advancing speed;
the feature extraction unit (300) adopts a network learning weight algorithm, and the calculation formula is as follows:
Yi＝αiX1i+βiX2i
Yicharacteristic of the fused ith channel, X1iFeatures of the ith channel in the first feature map before fusion, X2iFeatures of the ith channel, α, in the second feature map before fusioniAnd betaiWeighting coefficients for different channels;
the defect detection unit (400) comprises an acquisition module (410), a semantic feature acquisition module (420) and a feature fusion module (430);
the acquisition module (410) is configured to expand spatial resolution;
the semantic feature acquisition module (420) is used for extracting features, reducing the spatial resolution and acquiring high-level semantic features;
the feature fusion module (430) is used for expanding the spatial resolution of the acquisition module (410) and fusing feature layers of the upper part under different resolutions.
2. The AI-based industrial aperture wall defect detection system of claim 1, wherein: the feature fusion module (430) adopts a network algorithm for fusing multilayer features, and comprises the following steps:
firstly, training a model by adopting a back propagation method, and updating network parameters by using a random gradient descent method;
secondly, setting an initial learning rate, finishing the initial learning rate for multiple times, and adjusting the learning rate for one time after iteration;
and thirdly, setting the quantity of the batch input network samples.
3. The AI-based industrial aperture wall defect detection system of claim 1, wherein: the loss function unit (500) comprises the following poses:
posture one: the central point location heat map loss, using Focal length as the loss function, is expressed as follows:
wherein α and β are hyperparameters;
and (5) posture II: defect position center offset loss, adopting smooth L1 loss as a loss function;
posture three: the defect size loss function also adopts smooth L1 loss as a loss function, and the expression of the loss function is as follows:
and (4) posture IV: taking the bisection cross entropy as a loss function, the expression is as follows:
where y is equal to 0, 1.
4. An identification algorithm for industrial hole wall defect detection is characterized in that: the AI-based industrial pore wall defect detection system including any one of claims 1-3, operative by:
s1, data acquisition: a data acquisition unit (100) is adopted to convert a shot target into an image signal, the image signal is transmitted to an image processing system, and the image signal is converted into a digital signal by adopting artificial intelligence according to pixel distribution, brightness and color information;
s2, preprocessing data: the data preprocessing unit (200) is adopted to amplify the acquired image signals in a rotating, translating, mirroring and picture brightness adjusting mode, meanwhile, an image processing algorithm is adopted to subtract the mean value of the images in the training set and the test set, and the images are subjected to normalization processing;
s3, feature extraction: optimizing the weight coefficient and the bias of each node by adopting a network learning weight algorithm on the preprocessed image by adopting a feature extraction unit (300);
s4, defect detection: determining the central point of the detected defect position by adopting a network algorithm of weight fusion of multilayer characteristics according to the image characteristics by adopting a defect detection unit (400), and regressing the size and the offset of a defect frame at the central point position to detect the position and the type of the defect;
s5, determining a loss function: and judging a proper loss function by using a loss function unit (500) to optimize and specify the direction of the network parameters of the defect position.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210064160.7A CN114414578A (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210064160.7A CN114414578A (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
| CN202110060827.1A CN112903692B (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
Related Parent Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110060827.1A Division CN112903692B (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN114414578A true CN114414578A (en) | 2022-04-29 |
Family
ID=76114494
Family Applications (2)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210064160.7A Withdrawn CN114414578A (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
| CN202110060827.1A Active CN112903692B (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
Family Applications After (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202110060827.1A Active CN112903692B (en) | 2021-01-18 | 2021-01-18 | Industrial hole wall defect detection system and identification algorithm based on AI |
Country Status (1)
| Country | Link |
|---|---|
| CN (2) | CN114414578A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256372A (en) * | 2023-05-16 | 2023-06-13 | 征图新视(江苏)科技股份有限公司 | Optical film front-end Cheng Juancai full-detection device and detection method thereof |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN114804550B (en) * | 2022-06-10 | 2023-06-02 | 安徽新宇环保科技股份有限公司 | Sewage treatment regulation and control system based on neural network model |
Family Cites Families (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102608124A (en) * | 2012-04-06 | 2012-07-25 | 天津大学 | Micro-pipe internal flaw and appearance measurement device and method |
| CN107238603A (en) * | 2017-07-14 | 2017-10-10 | 深圳市恒工科技有限公司 | A kind of inner hole wall optical detection apparatus and its lamp guide |
| CN111126472B (en) * | 2019-12-18 | 2023-07-25 | 南京信息工程大学 | SSD (solid State disk) -based improved target detection method |
| CN112132156B (en) * | 2020-08-18 | 2023-08-22 | 山东大学 | Image saliency target detection method and system based on multi-depth feature fusion |
-
2021
- 2021-01-18 CN CN202210064160.7A patent/CN114414578A/en not_active Withdrawn
- 2021-01-18 CN CN202110060827.1A patent/CN112903692B/en active Active
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116256372A (en) * | 2023-05-16 | 2023-06-13 | 征图新视(江苏)科技股份有限公司 | Optical film front-end Cheng Juancai full-detection device and detection method thereof |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112903692A (en) | 2021-06-04 |
| CN112903692B (en) | 2022-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112903692B (en) | Industrial hole wall defect detection system and identification algorithm based on AI | |
| CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
| CN113436169B (en) | Industrial equipment surface crack detection method and system based on semi-supervised semantic segmentation | |
| CN115908311A (en) | Lens forming detection equipment based on machine vision and method thereof | |
| CN113313107A (en) | Intelligent detection and identification method for multiple types of diseases on cable surface of cable-stayed bridge | |
| CN113554638A (en) | Method and system for establishing chip surface defect detection model | |
| CN112529931A (en) | Foreground segmentation method and system | |
| CN115639248A (en) | System and method for detecting quality of building outer wall | |
| CN111160100A (en) | Lightweight depth model aerial photography vehicle detection method based on sample generation | |
| CN114463619A (en) | Infrared dim target detection method based on integrated fusion features | |
| CN116091506B (en) | Machine vision defect quality inspection method based on YOLOV5 | |
| CN117152094A (en) | Method, device and system for analyzing surface defects of steel plate based on computer vision | |
| CN114913086B (en) | Face image quality enhancement method based on generation countermeasure network | |
| CN116824319A (en) | Fusion method, device and storage medium of infrared image and visible light image | |
| CN115035168B (en) | Multi-constraint-based photovoltaic panel multi-source image registration method, device and system | |
| CN114612468B (en) | Equipment external defect detection method based on positive sample | |
| CN116563262A (en) | Building crack detection algorithm based on multiple modes | |
| CN112801071B (en) | Elevator asynchronous door opening recognition system and method based on deep learning | |
| CN112508946B (en) | Cable tunnel anomaly detection method based on antagonistic neural network | |
| CN115330743A (en) | Method for detecting defects based on double lights and corresponding system | |
| CN114663795A (en) | Target detection method for obtaining rear image of glass curtain wall by range gating imaging equipment | |
| CN114119479A (en) | Industrial production line quality monitoring method based on image recognition | |
| CN112150432A (en) | Optical excitation infrared nondestructive testing method based on generation countermeasure network | |
| TWI792291B (en) | Method of automatically setting optical parameters and automated optical inspection system using the same | |
| CN113591933B (en) | Remote sensing image change detection method and system based on correlation measurement |
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 | ||
| WW01 | Invention patent application withdrawn after publication | ||
| WW01 | Invention patent application withdrawn after publication |
Application publication date: 20220429 |