CN114943684B - Curved surface anomaly detection method by using confrontation to generate self-coding neural network - Google Patents
Curved surface anomaly detection method by using confrontation to generate self-coding neural network Download PDFInfo
- Publication number
- CN114943684B CN114943684B CN202210396491.0A CN202210396491A CN114943684B CN 114943684 B CN114943684 B CN 114943684B CN 202210396491 A CN202210396491 A CN 202210396491A CN 114943684 B CN114943684 B CN 114943684B
- Authority
- CN
- China
- Prior art keywords
- detection
- image
- neural network
- defect
- self
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000001514 detection method Methods 0.000 title claims abstract description 86
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 31
- 230000007547 defect Effects 0.000 claims abstract description 42
- 238000012549 training Methods 0.000 claims abstract description 19
- 238000012545 processing Methods 0.000 claims abstract description 17
- 230000005856 abnormality Effects 0.000 claims abstract description 15
- 230000002159 abnormal effect Effects 0.000 claims abstract description 12
- 238000013527 convolutional neural network Methods 0.000 claims abstract description 9
- 238000000034 method Methods 0.000 claims description 30
- 238000001914 filtration Methods 0.000 claims description 10
- 238000009499 grossing Methods 0.000 claims description 10
- 238000004364 calculation method Methods 0.000 claims description 7
- 238000013499 data model Methods 0.000 claims description 6
- 238000003708 edge detection Methods 0.000 claims description 6
- 238000004891 communication Methods 0.000 claims description 5
- 238000005516 engineering process Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 5
- 238000007781 pre-processing Methods 0.000 claims description 4
- 238000012790 confirmation Methods 0.000 claims description 3
- 230000001629 suppression Effects 0.000 claims description 3
- 238000012360 testing method Methods 0.000 claims description 3
- 230000007306 turnover Effects 0.000 claims description 3
- 238000012795 verification Methods 0.000 claims description 3
- 230000005540 biological transmission Effects 0.000 claims description 2
- 230000002950 deficient Effects 0.000 abstract description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 238000004519 manufacturing process Methods 0.000 description 2
- 238000009966 trimming Methods 0.000 description 2
- 229910000831 Steel Inorganic materials 0.000 description 1
- 230000003042 antagnostic effect Effects 0.000 description 1
- 239000000919 ceramic Substances 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000000605 extraction Methods 0.000 description 1
- -1 floors Substances 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 239000011521 glass Substances 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 210000001503 joint Anatomy 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 230000003287 optical effect Effects 0.000 description 1
- 238000007639 printing Methods 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
- 238000006467 substitution reaction 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
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T5/00—Image enhancement or restoration
- G06T5/70—Denoising; Smoothing
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/11—Region-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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20112—Image segmentation details
- G06T2207/20132—Image cropping
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/30—Computing systems specially adapted for manufacturing
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Computing Systems (AREA)
- Medical Informatics (AREA)
- Multimedia (AREA)
- Databases & Information Systems (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Data Mining & Analysis (AREA)
- Molecular Biology (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
A curved surface anomaly detection method using a countermeasure generated self-coding neural network comprises the following steps: acquiring defect-free images by using a plurality of industrial camera groups and combining the images; importing the non-defective image into an abnormal detection neural network, pre-training, and storing the position information of the false detection defect; finally, storing the result of eliminating the false detection as the final result of the abnormal detection; and loading the final result preview of the anomaly detection into the shared memory, and returning an inference ending signal to the client through the Socket. The invention overcomes the defects of the prior art, and realizes efficient and accurate abnormality detection by applying the abnormality detection method combining the countermeasure generating network and the self-encoder in the field of abnormality detection and by processing the convolutional neural network and the traditional image.
Description
Technical Field
The invention relates to the technical field of defect image identification, in particular to a curved surface anomaly detection method for generating a self-coding neural network by using confrontation.
Background
The production involves a plurality of procedures, and the problem of surface defects of the curved surface is usually inevitable. The method is promoted by the development of industrial automation, the unmanned production process is basically realized at present, but the quality inspection process still needs to depend on a large amount of manual operation, and the method has the defects of high human resource consumption, low efficiency, high omission factor and the like. The detection of the surface anomaly of the curved surface is taken as an important link of product quality management, and the realization of automation and the improvement of detection accuracy rate become a technical bottleneck which needs to be broken through urgently.
The curved surface anomaly detection technology based on the convolutional neural network and the traditional image processing has become a mainstream method to be widely applied to curved surface anomaly detection of products such as ceramics, glass, floors, steel rails and the like, and has the characteristics of high efficiency and high reliability. And by combining an automatic detection platform, an anomaly detection scheme with strong stability, high detection speed and low cost can be realized. Common abnormity detection for complex curved surfaces such as tires is based on laser signals and vacuum extraction in a physical mode, and the like, and due to the fact that the non-plane surface of the curved surface has height difference and a large amount of complex patterns and marked character information, the problems of printing dislocation, texture dislocation, character error and the like of the curved surface cannot be identified by a traditional detection method, a large amount of manual detection cannot be avoided, and optical drawing of the surface of the full curved surface is very difficult.
Therefore, how to provide an automatic curved surface anomaly detection method based on a convolutional neural network is a problem that needs to be solved urgently by those skilled in the art.
Disclosure of Invention
Aiming at the defects of the prior art, the invention provides a curved surface anomaly detection method by utilizing a countermeasure generation self-coding neural network, which overcomes the defects of the prior art, and realizes efficient and accurate anomaly detection by applying an anomaly detection method combining the countermeasure generation network and a self-coder in the anomaly detection field and processing a convolutional neural network and a traditional image.
In order to achieve the purpose, the invention is realized by the following technical scheme:
a curved surface anomaly detection method using a countermeasure generated self-coding neural network comprises the following steps:
step S1: acquiring a complete curved surface characteristic image by using an industrial camera set, and establishing a defect image library;
step S2: performing automatic image cutting on the defect image acquired in the step S1 by using an edge detection algorithm, and accurately positioning an area; dividing the surface characteristic image of the curved surface into a plurality of detection sectors according to the repeatability of the patterns, intercepting areas, and combining to obtain an image;
and step S3: performing data enhancement processing on the image data set obtained in the step S2;
and step S4: establishing an abnormal discrimination neural network based on the surface defects of the curved surface of the convolutional neural network;
step S5: inputting the defect image subjected to enhancement processing into an abnormality discrimination neural network, and training the target abnormality discrimination neural network to obtain an optimized abnormality discrimination neural network;
step S6: solidifying the model parameters trained in the step S5, deploying the model parameters to a high-speed reasoning engine, realizing the butt joint with any language compiling software in a mode of combining Socket communication and shared memory, and simultaneously carrying out anomaly detection;
step S7: and eliminating false detection caused by complex lines.
Preferably, in step S1, the image library is composed of high-resolution images collected by various industrial cameras, and the main data is a high-precision scanned image.
Preferably, the step S2 specifically includes the following steps:
step S21: and performing Gaussian filtering smoothing processing on the defect image acquired in the step S1 to reduce the interference of the background noise, wherein the calculation formula of the Gaussian filtering smoothing is as follows:
where f (m, n) is the gray value of the position (m, n), σ is the width of the Gaussian filter, determines the degree of smoothing, g σ (m, n) is a gray value of the position (m, n) after Gaussian filtering;
step S22: and estimating the edge strength and the gradient direction of each point by calculating the gradient amplitude, wherein the calculation formula of the edge strength and the calculation formula of the gradient direction are respectively as follows:
wherein, g x (m, n) and g y (m, n) are gradient values in two directions;
step S23: according to the gradient direction, carrying out non-maximum suppression on the gradient amplitude, and then connecting the edges;
step S24: dividing the surface of the curved surface into a plurality of detection sectors according to the repeatability of the patterns, intercepting areas, and combining to obtain an image.
Preferably, the step S3 specifically includes the following steps:
step S31: carrying out data enhancement processing on the image by using random rotation and turnover operation to expand a defect data set;
step S32: and marking the enhanced defect image according to the defect type, and dividing the marked image into a training set, a verification set and a test set.
Preferably, in the step S4, the anomaly discrimination neural network is constructed by using a self-encoder and a convolutional neural network, and training is performed by using a countermeasure generation method.
Preferably, the step S6 of implementing the anomaly detection method includes:
step S61: carrying out edge detection and cutting on the scanned image;
s62, before each detection, a batch of defect-free pre-training is used as a data model for eliminating false detection;
s63, carrying out image reasoning by using an accelerated reasoning engine deployed to the TensorRT;
and step S64, finishing the confirmation of the detection result by using the data model for eliminating the false detection, and outputting the result.
Preferably, the result transferring step of the abnormality detecting method in step S6 is:
step S65: acquiring an instruction from a software end by using Socket communication, and acquiring a real-time image from a shared memory by adopting an Mmap technology;
step S66: carrying out preprocessing operations such as cutting, zooming, turning and the like on the image and then executing an inference process;
step S67: and storing the result locally, loading the result preview image into the shared memory and returning a Socket signal representing the completion of execution.
Preferably, in step S7, the step of eliminating the false detection caused by the complex texture includes:
step S71: loading a pattern template on the surface of the curved surface;
step S72, detecting a batch of defect-free targets, and recording the positions of all detected defect targets;
s73, adopting a clustering algorithm to the target position of each type of defect to obtain the coordinate position of each clustering center point to jointly form a false-detection-resisting model;
step S74, searching whether the detection target is close to a certain point in the anti-false-detection model by using a FLANN nearest neighbor algorithm for the detection result obtained by model inference, and judging whether the point is false-detected;
in step S75, the detection result excluding the erroneous detection is output.
The invention provides a curved surface anomaly detection method by using a confrontation self-coding neural network. The method has the following beneficial effects: the invention designs a neural network aiming at the characteristics of anomaly detection, and combines a TensorRT high-speed reasoning engine to realize high-efficiency surface anomaly detection of the curved surface. Compared with the traditional anomaly detection method, the method has the advantages that the detection accuracy is improved, the requirement for collecting a large number of marking defects is reduced, and the labor cost is greatly saved.
Drawings
In order to more clearly illustrate the present invention or the prior art solutions, the drawings that are needed in the description of the prior art will be briefly described below.
FIG. 1 is a general flowchart of a curved surface abnormality detection method according to the present invention (taking a tire as an example);
FIG. 2 is a flowchart of step S4 of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the technical solutions of the present invention will be described clearly and completely with reference to the accompanying drawings.
Examples
As shown in fig. 1 to 2, a method for detecting surface anomaly using a countermeasure generated self-coding neural network includes the following steps:
step S1: shooting at different angles by using an industrial camera set to obtain complete surface characteristic pictures of the curved surface, and establishing a defect image library; the image library consists of high-resolution images acquired by various industrial cameras, and main data are high-precision scanning images;
step S2: automatic image cutting is carried out on the defect image acquired in the step S1 by utilizing an edge detection algorithm in the traditional image processing, the background image is cut off, and the area is accurately positioned, so that the influence of the image background on abnormal detection can be avoided, and the detection range is reduced, thereby improving the detection efficiency; the specific method comprises the following steps:
firstly, gaussian filtering smoothing processing is carried out on the defect image acquired in the step S1, and background noise interference is reduced, wherein a calculation formula of the Gaussian filtering smoothing is as follows:
where f (m, n) is the gray value of the position (m, n), σ is the width of the Gaussian filter, determines the degree of smoothing, g σ (m, n) is a gray value of the position (m, n) after Gaussian filtering; the specific gray value g of each coordinate after Gaussian filtering is calculated through the formula σ (m, n), namely, the Gaussian smoothing processing of the whole image can be completed;
then, the gradient values g of each point of the image in two directions are calculated x (m, n) and g y (m,n)：
The edge strength of each point in the image estimated by the gradient value is calculated according to the following formula:
the gradient direction of each point in the image estimated by the gradient value is calculated according to the following formula:
then according to the gradient direction, carrying out non-maximum suppression on the gradient amplitude, and then connecting the edges;
and dividing the surface of the curved surface into a plurality of detection sectors according to the repeatability of the patterns, intercepting areas, and combining to obtain a defect image.
And step S3: performing data enhancement processing on the image data set obtained in the step S2; the method comprises the following steps:
firstly, carrying out data enhancement processing on an image by using random rotation and turnover operation to expand a defect data set;
and classifying and labeling the unclassified, qualified and unqualified images of the enhanced defect image, and dividing the labeled images and the corresponding labeling files into a training set, a verification set and a test set according to a proportion.
And step S4: establishing an abnormal discrimination neural network based on the surface defects of the curved surface of the convolutional neural network;
the invention utilizes an autoencoder and a convolution neural network to construct an abnormal discrimination neural network, utilizes an antagonistic generation mode to train, and adopts the following model construction mode:
in the coding stage, a multi-head self-attention mechanism and a ResNest18 basic structure of a variable convolutional layer are used, so that the sensing and extracting capability of the network on small target features and position information is improved and the small target features and the position information are used as a judgment network;
theory based on the classical GAN structure: in the early stage, an unmarked picture is used as a positive sample, a self-encoder restored picture is used as a negative sample, and the restoration capability of the self-encoder and the recognition capability of a judgment network pair are trained; after the training reaches a certain degree, the qualified sample passing through the self-encoder is taken as a positive sample, the unqualified sample passing through the self-encoder is taken as a negative sample, the confrontation training is further carried out, and the classification capability of the self-encoder to the positive and negative samples is trained:
step S5: inputting the training set and the defect marking file which are subjected to enhancement processing into an abnormal judgment neural network, training the abnormal judgment neural network, and obtaining an optimized abnormal judgment neural network through multiple times of model parameter tuning and training;
step S6: the method for realizing efficient anomaly detection by combining the neural network with the TensorRT comprises the following specific steps:
firstly, solidifying the model parameters trained in the step S5, and deploying the trained abnormal discrimination neural network to a TensorRT high-speed reasoning engine;
then, the function of the inference engine part is used as a server, and a connection relation is established with client software in a Socket communication mode;
then the client maps the defect image collected by the industrial camera set into the memory through the Mmap technology, the server reads the defect image scanned in real time from the shared memory, and performs preprocessing such as cutting, rotating, overturning and the like on the image;
then, using TensorRT to detect the abnormality;
finally, the defect picture and the inferred defect position information are stored locally;
and acquiring a final result of the anomaly detection, loading the result preview image into the shared memory, and returning an inference end signal to the client through the Socket.
In this step, the anomaly detection method is implemented by the steps of:
step S61: carrying out edge detection and cutting on the scanned image;
step S62, before each detection, a batch of non-defective pre-training is used as a data model for eliminating false detection;
s63, carrying out image reasoning by using an accelerated reasoning engine deployed to the TensorRT;
and step S64, finishing the confirmation of the detection result by using the data model for eliminating the false detection, and outputting the result.
In this step, the result transmission step of the abnormality detection method is:
step S65: acquiring an instruction from a software end by using Socket communication, and acquiring a real-time image from a shared memory by adopting an Mmap technology;
step S66: carrying out preprocessing operations such as cutting, zooming, turning and the like on the image and then executing an inference process;
step S67: and storing the result locally, loading the result preview image into the shared memory and returning a Socket signal representing the completion of execution.
Step S7: eliminating false detection caused by complex lines; the concrete implementation steps are as follows:
firstly, loading a pattern template on the surface of a curved surface;
detecting a batch of defect-free targets, and recording the positions of all detected defect targets;
then, a clustering algorithm is adopted for the target position of each type of defect to obtain the coordinate position of each clustering center point, and a false-detection resisting model is formed together;
then, for the detection result obtained by model inference, searching whether the detection target is close to a certain point in the anti-false-detection model by using a FLANN nearest neighbor algorithm, and judging whether the point is false-detected;
and finally, outputting the detection result after the error detection is eliminated.
In general, compared with the current artificial anomaly detection method, the technical scheme provided by the invention can achieve the following beneficial effects:
the method for detecting the surface anomaly of the curved surface provided by the invention is used for designing a neural network aiming at the anomaly detection characteristic and combining a TensorRT high-speed reasoning engine to realize high-efficiency surface anomaly detection of the curved surface.
As shown in fig. 1, taking the tire industry as an example, a large number of people are needed to perform manual sorting, trimming and finished product detection in the industry, and the human input for the process in each factory is counted by hundreds of people. And about 90% of the later stage does not need any treatment, and the part is selected in advance, so that the workload of manual trimming and sorting can be greatly reduced.
The above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (5)
1. A method for detecting surface anomalies using a countering self-encoding neural network, the method comprising: the method comprises the following steps:
step S1: acquiring a complete curved surface characteristic image by using an industrial camera set, and establishing a defect image library;
step S2: utilizing an edge detection algorithm to cut the defect image acquired in the step S1; dividing the surface characteristic image of the curved surface into a plurality of detection sectors according to the repeatability of the patterns, intercepting areas, and combining to obtain an image;
and step S3: performing data enhancement processing on the image data set obtained in the step S2;
and step S4: establishing an abnormal discrimination neural network based on the surface defects of the curved surface of the convolutional neural network;
step S5: in the early stage, an unmarked picture is used as a positive sample, a self-encoder restored picture is used as a negative sample, and the restoration capability of the self-encoder and the recognition capability of a judgment network pair are trained; after training to a certain degree, taking the qualified sample passing through the self-encoder as a positive sample, taking the unqualified sample passing through the self-encoder as a negative sample, further performing countermeasure training, and training the classification capability of the self-encoder on the positive and negative samples;
inputting the defect image subjected to enhancement processing into an abnormality discrimination neural network, and training the target abnormality discrimination neural network to obtain an optimized abnormality discrimination neural network;
step S6: solidifying the model parameters trained in the step S5, deploying the model parameters to a high-speed reasoning engine, and carrying out anomaly detection;
in the step S1, an image library consists of high-resolution images acquired by various industrial cameras, and main data are high-precision scanning images;
the step S2 specifically includes the following steps:
step S21: and performing Gaussian filtering smoothing processing on the defect image acquired in the step S1 to reduce the interference of the background noise, wherein the calculation formula of the Gaussian filtering smoothing is as follows:
where f (m, n) is the gray value of the position (m, n), σ is the width of the Gaussian filter, determines the degree of smoothing, g σ (m, n) is a gray value of the position (m, n) after Gaussian filtering;
step S22: and estimating the edge strength and the gradient direction of each point by calculating the gradient amplitude, wherein the calculation formula of the edge strength and the calculation formula of the gradient direction are respectively as follows:
wherein, g x (m, n) and g y (m, n) are gradient values in two directions;
step S23: according to the gradient direction, carrying out non-maximum suppression on the gradient amplitude, and then connecting the edges;
step S24: dividing the surface of the curved surface into a plurality of detection sectors according to the repeatability of patterns, intercepting areas, and combining to obtain an image;
further comprising step S7: eliminating false detection caused by complex lines; the method comprises the following implementation steps:
step S71: loading a pattern template on the surface of the curved surface;
step S72, detecting a batch of defect-free targets, and recording the positions of all detected defect targets;
s73, adopting a clustering algorithm to the target position of each type of defect to obtain the coordinate position of each clustering center point to jointly form a false-detection-resisting model;
step S74, searching whether the detection target is close to a certain point in the anti-false-detection model by using a FLANN nearest neighbor algorithm for the detection result obtained by model inference, and judging whether the point is false-detected;
in step S75, the detection result excluding the erroneous detection is output.
2. The method of claim 1, wherein the method comprises: the step S3 specifically includes the following steps:
step S31: carrying out data enhancement processing on the image by using random rotation and turnover operation to expand a defect data set;
step S32: and marking the enhanced defect image according to the defect type, and dividing the marked image into a training set, a verification set and a test set.
3. The method for detecting surface anomaly using robust self-coding neural network as claimed in claim 1, wherein: and S4, constructing an abnormal discrimination neural network by using the self-encoder and the convolutional neural network, and training by using a countermeasure generation mode.
4. The method of claim 1, wherein the method comprises: the method for detecting the abnormality in the step S6 comprises the following steps:
step S61: carrying out edge detection and cutting on the scanned image;
s62, before each detection, a batch of defect-free pre-training is used as a data model for eliminating false detection;
s63, carrying out image reasoning by using an accelerated reasoning engine deployed to the TensorRT;
and step S64, finishing the confirmation of the detection result by using the data model for eliminating the false detection, and outputting the result.
5. The method of claim 4, wherein the method comprises: the result transmission step of the abnormality detection method in step S6 is:
step S65: acquiring an instruction from a software end by using Socket communication, and acquiring a real-time image from a shared memory by adopting an Mmap technology;
step S66: carrying out preprocessing operations such as cutting, zooming, turning and the like on the image and then executing an inference process;
step S67: and storing the result locally, loading the result preview image into the shared memory and returning a Socket signal representing the completion of execution.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210396491.0A CN114943684B (en) | 2022-04-15 | 2022-04-15 | Curved surface anomaly detection method by using confrontation to generate self-coding neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202210396491.0A CN114943684B (en) | 2022-04-15 | 2022-04-15 | Curved surface anomaly detection method by using confrontation to generate self-coding neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN114943684A CN114943684A (en) | 2022-08-26 |
| CN114943684B true CN114943684B (en) | 2023-04-07 |
Family
ID=82907827
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202210396491.0A Active CN114943684B (en) | 2022-04-15 | 2022-04-15 | Curved surface anomaly detection method by using confrontation to generate self-coding neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN114943684B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113901947A (en) * | 2021-11-04 | 2022-01-07 | 浙江大学高端装备研究院 | Intelligent identification method for tire surface flaws under small sample |
| CN113935953A (en) * | 2021-09-18 | 2022-01-14 | 南通豪派金属制品有限公司 | Steel coil defect detection method based on image processing |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101808237A (en) * | 2010-03-09 | 2010-08-18 | 西安科技大学 | Image acquisition terminal for embedded system web server and image acquisition method |
| JP5443435B2 (en) * | 2011-05-17 | 2014-03-19 | シャープ株式会社 | Tire defect detection method |
| CN105430006B (en) * | 2015-12-25 | 2019-05-14 | 深圳市研唐科技有限公司 | A kind of method of dynamic realtime adjustment spice image quality |
| CN107680086B (en) * | 2017-09-27 | 2020-10-23 | 电子科技大学 | Method for detecting material contour defects with arc-shaped edges and linear edges |
| CN107886509A (en) * | 2017-11-24 | 2018-04-06 | 苏州珂锐铁电气科技有限公司 | A kind of image deflects recognition methods, electronic equipment, storage medium and system |
| CN108961217B (en) * | 2018-06-08 | 2022-09-16 | 南京大学 | Surface defect detection method based on regular training |
| EP3916635B1 (en) * | 2020-05-26 | 2023-05-10 | Fujitsu Limited | Defect detection method and apparatus |
| CN112528975A (en) * | 2021-02-08 | 2021-03-19 | 常州微亿智造科技有限公司 | Industrial quality inspection method, device and computer readable storage medium |
-
2022
- 2022-04-15 CN CN202210396491.0A patent/CN114943684B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113935953A (en) * | 2021-09-18 | 2022-01-14 | 南通豪派金属制品有限公司 | Steel coil defect detection method based on image processing |
| CN113901947A (en) * | 2021-11-04 | 2022-01-07 | 浙江大学高端装备研究院 | Intelligent identification method for tire surface flaws under small sample |
Also Published As
| Publication number | Publication date |
|---|---|
| CN114943684A (en) | 2022-08-26 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110543878B (en) | Pointer instrument reading identification method based on neural network | |
| US10803573B2 (en) | Method for automated detection of defects in cast wheel products | |
| CN109840900B (en) | Fault online detection system and detection method applied to intelligent manufacturing workshop | |
| CN111325713A (en) | Wood defect detection method, system and storage medium based on neural network | |
| CN113362326A (en) | Method and device for detecting welding spot defects of battery | |
| CN111815601A (en) | Texture image surface defect detection method based on depth convolution self-encoder | |
| CN112037219A (en) | Metal surface defect detection method based on two-stage convolution neural network | |
| CN110751604A (en) | Machine vision-based steel pipe weld defect online detection method | |
| CN114994061A (en) | Machine vision-based steel rail intelligent detection method and system | |
| CN113627435A (en) | Method and system for detecting and identifying flaws of ceramic tiles | |
| CN109693140A (en) | A kind of intelligent flexible production line and its working method | |
| Wu et al. | An end-to-end learning method for industrial defect detection | |
| CN111524154B (en) | Image-based tunnel segment automatic segmentation method | |
| CN115131268A (en) | Automatic welding system based on image feature extraction and three-dimensional model matching | |
| CN115035092A (en) | Image-based bottle detection method, device, equipment and storage medium | |
| Kähler et al. | Anomaly detection for industrial surface inspection: Application in maintenance of aircraft components | |
| Ekambaram et al. | Identification of defects in casting products by using a convolutional neural network | |
| CN103926255A (en) | Method for detecting surface defects of cloth based on wavelet neural network | |
| CN114943684B (en) | Curved surface anomaly detection method by using confrontation to generate self-coding neural network | |
| CN113705564A (en) | Pointer type instrument identification reading method | |
| TWI822968B (en) | Color filter inspection device, inspection device, color filter inspection method, and inspection method | |
| JP2021060962A (en) | Method of generating teacher data, trained learning model, and system | |
| CN115100110A (en) | Defect detection method, device and equipment for polarized lens and readable storage medium | |
| CN113421334B (en) | Multi-focus image three-dimensional reconstruction method based on deep learning | |
| Cateni et al. | Fuzzy inference systems applied to image classification in the industrial field |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |