CN107657603A - A kind of industrial appearance detecting method based on intelligent vision - Google Patents
A kind of industrial appearance detecting method based on intelligent vision Download PDFInfo
- Publication number
- CN107657603A CN107657603A CN201710719954.1A CN201710719954A CN107657603A CN 107657603 A CN107657603 A CN 107657603A CN 201710719954 A CN201710719954 A CN 201710719954A CN 107657603 A CN107657603 A CN 107657603A
- Authority
- CN
- China
- Prior art keywords
- image
- appearance
- deep learning
- product
- flaw
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 33
- 238000001514 detection method Methods 0.000 claims abstract description 88
- 238000013135 deep learning Methods 0.000 claims abstract description 39
- 238000013136 deep learning model Methods 0.000 claims abstract description 28
- 238000011176 pooling Methods 0.000 claims description 27
- 230000007547 defect Effects 0.000 claims description 22
- 230000003213 activating effect Effects 0.000 claims description 12
- 238000004364 calculation method Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 9
- 239000011159 matrix material Substances 0.000 claims description 9
- 230000004913 activation Effects 0.000 claims description 3
- 230000002950 deficient Effects 0.000 claims description 3
- 238000002372 labelling Methods 0.000 abstract 1
- 239000004642 Polyimide Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 229920001721 polyimide Polymers 0.000 description 2
- 230000000007 visual effect Effects 0.000 description 2
- 241000282414 Homo sapiens Species 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 230000003252 repetitive effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
- G06T7/0004—Industrial image inspection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2415—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on parametric or probabilistic models, e.g. based on likelihood ratio or false acceptance rate versus a false rejection rate
-
- 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
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- 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
- G06T2207/20081—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]
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- General Health & Medical Sciences (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Molecular Biology (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Health & Medical Sciences (AREA)
- Probability & Statistics with Applications (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention provides a kind of industrial appearance detecting method based on intelligent vision, this method comprises the following steps：(1) multiple product appearance flaw samples, the image that the sample containing same flaw includes multiple different angles or position is shot, are obtained；(2), to the flaw image labeling in product appearance flaw sample, the affiliated classification of flaw in image is determined；(3), according to the classification of flaw in product appearance flaw sample, image, deep learning network training is carried out, obtains product appearance deep learning model；(4) test product appearance images, are obtained in real time, using product appearance deep learning model, take deep learning method to detect the product appearance image；(5), for the image of multiple angles and positions by after deep learning Model Identification, any image is identified flaw, you can thinks that the product has flaw.Treatment effeciency of the present invention is high, and accuracy of detection exceedes the accuracy of detection of the mankind, and examination criteria is unified, avoids repeated detection.
Description
Technical Field
The invention relates to an industrial appearance detection method based on intelligent vision, and belongs to the field of industrial appearance detection.
Background
In the field of industrial detection, appearance detection is an important content, the quality of the product is influenced to a certain extent by the quality of the appearance, the automation degree of the appearance detection at the present stage is low, most of the appearance detection depends on manual detection, and even in an automatic device which slightly utilizes image processing for detection, certain defects also exist.
The current industrial appearance detection mainly has the following problems:
1. the automatic detection equipment based on image processing is insufficient in extraction of appearance features, mainly the defects of appearance detection are various, single image processing cannot easily meet diversified detection requirements, and detection precision is low.
2. The existing deep learning intelligent identification model needs sample diversification due to characteristic extraction and cannot be directly used for industrial product detection, so that the detection of industrial appearance intelligently utilizes automatic means with low precision such as image processing and the like.
3. The manual detection brings the problems of high cost and low detection efficiency.
4. The detection standards are not uniform, the difference between the detection standards of people is large, and the detection result is unstable.
Disclosure of Invention
The technical problem to be solved by the invention is as follows: the defects of the prior art are overcome, an industrial appearance detection technology based on intelligent vision is provided, and a detection method of deep learning is adopted to rapidly detect the appearance of a product.
The technical solution of the invention is as follows: an industrial appearance detection method based on intelligent vision comprises the following steps:
(1) obtaining M product appearance flaw samples, wherein M is more than or equal to 3000, the samples are product images containing appearance flaws, the product images containing the appearance flaws are obtained by shooting products containing the appearance flaws and fixed on a detection tool through a camera, and the samples containing the same flaws are multiple images shot at different angles or positions respectively;
(2) manually marking a flaw image in the product appearance flaw sample, and determining the category of the flaw in the image;
(3) performing deep learning network training according to the product appearance defect sample and the types of the defects in the image determined in the step (2) to obtain a product appearance deep learning model;
(4) acquiring an appearance image of a detected product in real time, detecting the appearance image of the product by adopting a product appearance deep learning model obtained in the step (3) and adopting a deep learning method to identify the defect type in the image, wherein the appearance image of the detected product is obtained by shooting the detected product fixed on a detection tool by a camera, and a plurality of appearance images of the same detected product are respectively shot at a plurality of different angles or positions;
(5) and after the product appearance images at a plurality of angles and positions are identified by the deep learning model, if any one or more images are identified to have defects, the product can be regarded as a defective product.
The detection tool at least comprises 3 degrees of freedom, and the detected product appearance images of a plurality of angles and positions can be obtained by adjusting the angles and the positions of the detection tool.
According to the product appearance defect sample and the category of the defects in the image determined in the step (2), deep learning network training is carried out, and the specific implementation of the product appearance deep learning model is as follows:
(3.1) calculating the sample by adopting a preset deep learning training model through a forward propagation method to obtain confidence coefficients of a plurality of classes preset in the deep learning training model;
(3.2) analyzing the confidence coefficient of each category, and selecting the category with the highest confidence coefficient as the category Z to which the flaw contained in the sample belongs;
(3.3) adjusting parameters of a deep learning training model by adopting a back propagation method according to the class Z to which the flaws contained in the sample in the step (3.2) belong, the actual class of the flaws and a preset training parameters α, and accumulating the number s of training rounds;
(3.4) repeating the steps (3.1) to (3.3) until the training round reaches a preset number a, reducing the preset training parameters, and turning to the step (3.5);
and (3.5) repeating the steps (3.1) to (3.3) until the training round reaches a preset number b.
The step (3.1) of calculating the sample by a forward propagation method to obtain confidence coefficients of a plurality of classes preset in the deep learning training model specifically comprises the following steps:
convolving the sample image, pooling the convolution result, and activating the pooling result to obtain the 1 st image feature layer of the image; then, performing convolution on the 1 st image feature layer, performing pooling on a convolution result, finally activating a pooling result to obtain a 2 nd image feature layer of the image, repeating the above steps to obtain an N-th image feature layer, and then performing forward calculation according to the N-th image feature layer through a Softmax function to obtain confidence coefficients of a plurality of categories preset in the deep learning training model, wherein N is greater than or equal to 6;
the result of the convolution of each layer isWhen i is 1, Li-1Representing the sample image, otherwise, Li-1Represents the i-1 th image feature layer, hi(k,j),j＝1～miA predetermined m is a i-th layer convolution kernel with a matrix of k × kiAnd (4) respectively.
The result of the treatment of each layer of pooling isMeans for performing an mxn downsampling operation on the matrix X;
the processing result of each layer activation is as follows: l isi,j＝max(0,Pi,j),i＝1～N,j＝1～miAnd max (0, X) denotes that the element in the matrix X is updated by comparing the element with 0 and taking a large value.
The specific processing process of the back propagation method for adjusting the parameters of the deep learning training model is as follows:
(3.3.1) comparing the category Y to which the flaws contained in the sample actually belong with the category Z to which the flaws contained in the sample obtained in the step (2) belong, and substituting the categories into a cost function f to obtain a loss value E:
E＝f(Y-Z)；
(3.3.2) updating deep learning model parameters according to the loss value E and the training parameters α, wherein the model parameters comprise a convolution kernel parameter h in convolution calculationi(k,j),j＝1～mi,i＝1～N：
Wherein,representing in the last training round
The step (4) of detecting the appearance image of the detected product by adopting a deep learning method, and the specific process of identifying the position and the category of the flaw in the image comprises the following steps: convolving the appearance image of the tested product, pooling the convolution result, and activating the pooling result to obtain the 1 st image characteristic layer of the image; and then, performing convolution on the 1 st image characteristic layer, performing pooling on the convolution result, finally activating the pooling result to obtain a 2 nd image characteristic layer of the image, repeating the above steps to obtain an Nth image characteristic layer, wherein N is more than or equal to 6, performing forward calculation by adopting a Softmax function to obtain confidence coefficients of a plurality of categories preset in a deep learning training model, selecting the category with the highest confidence coefficient as the category to which the flaws included in the appearance image of the tested product belong, and adopting a convolution kernel finally determined in the deep learning model in the convolution processing process of each layer.
Compared with the prior art, the invention has the beneficial effects that:
(1) the industrial appearance detection method based on intelligent vision is adopted, the image is used as system input, various appearance defects can be intelligently identified, the processing efficiency is high, the detection precision exceeds the detection precision of human beings, the detection standards are unified, and manual repeated detection is avoided.
(2) The invention solves the difficulties in the prior industrial automation field by utilizing intelligent visual detection in the industrial appearance detection field, not only solves the industrial problems and realizes high-speed and high-precision detection, but also can replace manual work, improve the detection efficiency and unify the detection standard.
(3) The invention uses the multi-degree-of-freedom detection tool to shoot the samples of the detected workpieces in multiple postures to obtain the samples which are easy to be used by the deep learning model, and then the accurate identification of the deep learning model is utilized, so that the product flaws can be detected with high precision, the detection precision is improved, and the problems that the detection features are not fully extracted and cannot be well identified due to single detection background when the intelligent detection based on the deep learning is directly used in industrial detection are solved.
Drawings
FIG. 1 is a block diagram of an implementation of deep learning identification detection in accordance with the present invention;
fig. 2 is a flowchart of an industrial appearance detection method based on smart vision according to an embodiment of the present invention.
Detailed Description
The invention is described in detail below with reference to the figures and specific examples of the specification.
As shown in fig. 1, the industrial appearance detection method based on intelligent vision provided by the invention is divided into two processes of offline training and online detection, wherein the offline training process obtains a product appearance deep learning model; and detecting whether the product has defects and the type of the defects on line.
As shown in fig. 2, the industrial appearance detection method based on smart vision provided by the present invention includes the following steps:
(1) obtaining M product appearance flaw samples, wherein M is more than or equal to 3000, the samples are product images containing appearance flaws, the product images containing the appearance flaws are obtained by shooting products containing the appearance flaws and fixed on a detection tool through a camera, and the samples containing the same flaws are multiple images shot at different angles or positions respectively; the detection tool at least comprises 3 degrees of freedom, and the detected product appearance images of a plurality of angles and positions can be obtained by adjusting the angles and the positions of the detection tool. The multi-degree-of-freedom detection tool is used for shooting samples of multi-posture detected workpieces to obtain samples which are easy to use by a deep learning model, then accurate identification of the deep learning model is utilized, product flaws can be detected with high precision, detection precision is improved, and the problems that detection features are not extracted sufficiently and cannot be identified well due to single detection background when the intelligent detection based on deep learning is directly used in industrial detection are solved.
(2) Manually marking a flaw image in the product appearance flaw sample, and determining the category of the flaw in the image;
(3) performing deep learning network training according to the product appearance defect sample and the types of the defects in the image determined in the step (2) to obtain a product appearance deep learning model;
the concrete implementation is as follows:
(3.1) calculating the sample by adopting a preset deep learning training model through a forward propagation method to obtain confidence coefficients of a plurality of classes preset in the deep learning training model;
convolving the sample image, pooling the convolution result, and activating the pooling result to obtain the 1 st image feature layer of the image; then, performing convolution on the 1 st image feature layer, performing pooling on a convolution result, finally activating a pooling result to obtain a 2 nd image feature layer of the image, repeating the above steps to obtain an N-th image feature layer, and then performing forward calculation according to the N-th image feature layer through a Softmax function to obtain confidence coefficients of a plurality of categories preset in the deep learning training model, wherein N is greater than or equal to 6;
the result of the convolution of each layer isWhen i is 1, Li-1Representing the sample image, otherwise, Li-1Represents the i-1 th image feature layer, hi(k,j),j＝1～miA predetermined m is a i-th layer convolution kernel with a matrix of k × kiOne, e.g., 128.
The result of the treatment of each layer of pooling isMeans for performing an mxn downsampling operation on the matrix X;
the processing result of each layer activation is as follows: l isi,j＝max(0,Pi,j),i＝1～N,j＝1～miAnd max (0, X) denotes that the element in the matrix X is updated by comparing the element with 0 and taking a large value.
(3.2) analyzing the confidence coefficient of each category, and selecting the category with the highest confidence coefficient as the category Z to which the flaw contained in the sample belongs;
(3.3) adjusting parameters of a deep learning training model by adopting a back propagation method according to the class Z to which the flaws contained in the sample in the step (3.2) belong, the actual class of the flaws and a preset training parameters α, and accumulating the number s of training rounds;
(3.3.1) comparing the category Y to which the flaws contained in the sample actually belong with the category Z to which the flaws contained in the sample obtained in the step (2) belong, and substituting the categories into a cost function f to obtain a loss value E:
e ═ f (Y-Z), for example:
(3.3.2) updating deep learning model parameters according to the loss value E and the training parameters α, wherein the model parameters comprise a convolution kernel parameter h in convolution calculationi(k,j),j＝1～mi,i＝1～N：
Wherein, among others,representing in the last training roundthe training parameter α may initially be 0.001, the reduction method may be a step method, and the step amount may be adjusted according to the actual situation.
(3.4) repeating the steps (3.1) to (3.3) until the training round reaches a preset number a, reducing the preset training parameters, and turning to the step (3.5);
and (3.5) repeating the steps (3.1) to (3.3) until the training round reaches a preset number b.
In the above process, a and b are adjusted according to the actual training error value, and generally, a is more than 8000 times, and b is more than 12000 times.
(4) And (3) acquiring an appearance image of the detected product in real time, adopting the product appearance deep learning model obtained in the step (3), detecting the appearance image of the product by adopting a deep learning method, and identifying the defect type in the image, wherein the appearance image of the detected product is obtained by shooting the detected product fixed on a detection tool by a camera, and a plurality of appearance images of the same detected product are respectively images shot at a plurality of different angles or positions.
The method comprises the following steps of detecting an appearance image of a detected product by adopting a deep learning method, and identifying the positions and the types of flaws in the image: convolving the appearance image of the tested product, pooling the convolution result, and activating the pooling result to obtain the 1 st image characteristic layer of the image; and then, performing convolution on the 1 st image characteristic layer, performing pooling on the convolution result, finally activating the pooling result to obtain a 2 nd image characteristic layer of the image, repeating the above steps to obtain an Nth image characteristic layer, wherein N is more than or equal to 6, performing forward calculation by adopting a Softmax function to obtain confidence coefficients of a plurality of categories preset in a deep learning training model, selecting the category with the highest confidence coefficient as the category to which the flaws included in the appearance image of the tested product belong, and adopting a convolution kernel finally determined in the deep learning model in the convolution processing process of each layer.
(5) And after the product appearance images at a plurality of angles and positions are identified by the deep learning model, if any one or more images are identified to have defects, the product can be regarded as a defective product.
In the field of industrial appearance detection, the invention utilizes intelligent visual detection to solve the difficulties in the past industrial automation field, thereby not only solving the industrial problems and realizing high-speed and high-precision detection, but also replacing manpower, improving the detection efficiency and the detection precision, having uniform detection standards and avoiding the manual repetitive detection.
Example (b):
according to the appearance detection equipment of a certain industrial polyimide product, the industrial appearance detection method based on intelligent vision is adopted, 5000 flaw samples of a detected workpiece are collected through a multi-degree-of-freedom detection tool, the samples are shot at multiple angles and positions and used for deep learning model training, and a product appearance detection deep learning model is obtained through 120000 rounds of deep learning training; the detection equipment obtains a detection result by utilizing a deep learning model and processing the image of the polyimide product in real time; the appearance detection accuracy of the deep learning training model reaches 93.2%, the precision of manual detection is exceeded, and the first industrial appearance detection device based on intelligent vision is realized.
The invention is not described in detail and is within the knowledge of a person skilled in the art.
Claims (6)
1. An industrial appearance detection method based on intelligent vision is characterized by comprising the following steps:
(1) obtaining M product appearance flaw samples, wherein M is more than or equal to 3000, the samples are product images containing appearance flaws, the product images containing the appearance flaws are obtained by shooting products containing the appearance flaws and fixed on a detection tool through a camera, and the samples containing the same flaws are multiple images shot at different angles or positions respectively;
(2) manually marking a flaw image in the product appearance flaw sample, and determining the category of the flaw in the image;
(3) performing deep learning network training according to the product appearance defect sample and the types of the defects in the image determined in the step (2) to obtain a product appearance deep learning model;
(4) acquiring an appearance image of a detected product in real time, detecting the appearance image of the product by adopting a product appearance deep learning model obtained in the step (3) and adopting a deep learning method to identify the defect type in the image, wherein the appearance image of the detected product is obtained by shooting the detected product fixed on a detection tool by a camera, and a plurality of appearance images of the same detected product are respectively shot at a plurality of different angles or positions;
(5) and after the product appearance images at a plurality of angles and positions are identified by the deep learning model, if any one or more images are identified to have defects, the product can be regarded as a defective product.
2. The industrial appearance detection method based on the intelligent vision is characterized in that the detection tool at least comprises 3 degrees of freedom, and the appearance images of the detected product at a plurality of angles and positions can be obtained by adjusting the angles and the positions of the detection tool.
3. The industrial appearance detection method based on intelligent vision according to claim 1, characterized in that according to the product appearance defect sample and the category of the defect in the image determined in step (2), deep learning network training is performed to obtain a product appearance deep learning model, which is specifically realized as follows:
(3.1) calculating the sample by adopting a preset deep learning training model through a forward propagation method to obtain confidence coefficients of a plurality of classes preset in the deep learning training model;
(3.2) analyzing the confidence coefficient of each category, and selecting the category with the highest confidence coefficient as the category Z to which the flaw contained in the sample belongs;
(3.3) adjusting parameters of a deep learning training model by adopting a back propagation method according to the class Z to which the flaws contained in the sample in the step (3.2) belong, the actual class of the flaws and a preset training parameters α, and accumulating the number s of training rounds;
(3.4) repeating the steps (3.1) to (3.3) until the training round reaches a preset number a, reducing the preset training parameters, and turning to the step (3.5);
and (3.5) repeating the steps (3.1) to (3.3) until the training round reaches a preset number b.
4. The industrial appearance detection method based on intelligent vision according to claim 1, characterized in that the step (3.1) of calculating the samples by a forward propagation method to obtain confidence levels of a plurality of classes preset in the deep learning training model specifically comprises:
convolving the sample image, pooling the convolution result, and activating the pooling result to obtain the 1 st image feature layer of the image; then, performing convolution on the 1 st image feature layer, performing pooling on a convolution result, finally activating a pooling result to obtain a 2 nd image feature layer of the image, repeating the above steps to obtain an N-th image feature layer, and then performing forward calculation according to the N-th image feature layer through a Softmax function to obtain confidence coefficients of a plurality of categories preset in the deep learning training model, wherein N is greater than or equal to 6;
the result of convolution processing for each layer is Gi,j：When i is 1, Li-1Representing the sample image, otherwise, Li-1Represents the i-1 th image feature layer, hi(k,j),j＝1～miA predetermined m is a i-th layer convolution kernel with a matrix of k × kiAnd (4) respectively.
The result of the treatment of each layer of pooling is Pi,j： Means for performing an mxn downsampling operation on the matrix X;
the processing result of each layer activation is as follows: l isi,j＝max(0,Pi,j),i＝1～N,j＝1～miAnd max (0, X) denotes that the element in the matrix X is updated by comparing the element with 0 and taking a large value.
5. The industrial appearance detection method based on intelligent vision according to claim 1, characterized in that the specific processing procedure of adjusting the deep learning training model parameters by the back propagation method is as follows:
(3.3.1) comparing the category Y to which the flaws contained in the sample actually belong with the category Z to which the flaws contained in the sample obtained in the step (2) belong, and substituting the categories into a cost function f to obtain a loss value E:
E＝f(Y-Z)；
(3.3.2) updating deep learning model parameters according to the loss value E and the training parameters α, wherein the model parameters comprise a convolution kernel parameter h in convolution calculationi(k,j),j＝1～mi,i＝1～N：
<mrow> <msub> <mi>h</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mi>k</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>=</mo> <msup> <msub> <mi>h</mi> <mi>i</mi> </msub> <mrow> <mo>(</mo> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </msup> <mrow> <mo>(</mo> <mi>k</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> <mo>-</mo> <mi>&alpha;</mi> <mo>&CenterDot;</mo> <mfrac> <mi>E</mi> <mrow> <msubsup> <mi>h</mi> <mi>i</mi> <mrow> <mo>(</mo> <mo>-</mo> <mn>1</mn> <mo>)</mo> </mrow> </msubsup> <mrow> <mo>(</mo> <mi>k</mi> <mo>,</mo> <mi>j</mi> <mo>)</mo> </mrow> </mrow> </mfrac> <mo>,</mo> <mi>j</mi> <mo>=</mo> <mn>1</mn> <mo>~</mo> <msub> <mi>m</mi> <mi>i</mi> </msub> <mo>,</mo> <mi>i</mi> <mo>=</mo> <mn>1</mn> <mo>~</mo> <mi>N</mi> <mo>;</mo> </mrow>
Wherein,representing in the last training round
6. The industrial appearance detection method based on intelligent vision according to claim 1, characterized in that the step (4) of detecting the appearance image of the product under test by a deep learning method, and the specific process of identifying the position and the category of the flaw in the image is as follows: convolving the appearance image of the tested product, pooling the convolution result, and activating the pooling result to obtain the 1 st image characteristic layer of the image; and then, performing convolution on the 1 st image characteristic layer, performing pooling on the convolution result, finally activating the pooling result to obtain a 2 nd image characteristic layer of the image, repeating the above steps to obtain an Nth image characteristic layer, wherein N is more than or equal to 6, performing forward calculation by adopting a Softmax function to obtain confidence coefficients of a plurality of categories preset in a deep learning training model, selecting the category with the highest confidence coefficient as the category to which the flaws included in the appearance image of the tested product belong, and adopting a convolution kernel finally determined in the deep learning model in the convolution processing process of each layer.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710719954.1A CN107657603B (en) | 2017-08-21 | 2017-08-21 | Industrial appearance detection method based on intelligent vision |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710719954.1A CN107657603B (en) | 2017-08-21 | 2017-08-21 | Industrial appearance detection method based on intelligent vision |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107657603A true CN107657603A (en) | 2018-02-02 |
| CN107657603B CN107657603B (en) | 2020-07-14 |
Family
ID=61128619
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710719954.1A Active CN107657603B (en) | 2017-08-21 | 2017-08-21 | Industrial appearance detection method based on intelligent vision |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107657603B (en) |
Cited By (22)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108227664A (en) * | 2018-02-05 | 2018-06-29 | 华侨大学 | Control of product quality equipment and method of quality control based on sample data training |
| CN108672316A (en) * | 2018-03-27 | 2018-10-19 | 哈尔滨理工大学 | A kind of micro parts quality detecting system based on convolutional neural networks |
| CN110310260A (en) * | 2019-06-19 | 2019-10-08 | 北京百度网讯科技有限公司 | Sub-material decision-making technique, equipment and storage medium based on machine learning model |
| CN110335272A (en) * | 2019-07-12 | 2019-10-15 | 创新奇智(广州)科技有限公司 | A kind of metal device intelligence quality inspection apparatus control system, method and electronic equipment based on machine vision |
| CN110659662A (en) * | 2018-06-29 | 2020-01-07 | 由田新技股份有限公司 | Flaw detection system and method using artificial intelligence |
| CN110738630A (en) * | 2018-07-02 | 2020-01-31 | 由田新技股份有限公司 | Training method and detection system of recursive deep learning system |
| CN110766664A (en) * | 2019-09-29 | 2020-02-07 | 杭州电子科技大学 | A detection method for defective appearance of electronic components based on deep learning |
| CN111077166A (en) * | 2018-10-19 | 2020-04-28 | 北京金山云网络技术有限公司 | Liquid crystal screen defect detection method, device and terminal equipment |
| CN111093140A (en) * | 2019-12-11 | 2020-05-01 | 上海闻泰信息技术有限公司 | Method, device, equipment and storage medium for detecting defects of microphone and earphone dust screen |
| CN111103306A (en) * | 2018-10-29 | 2020-05-05 | 所罗门股份有限公司 | Method for detecting and marking defects |
| CN111259887A (en) * | 2020-01-08 | 2020-06-09 | 网思科技股份有限公司 | Intelligent quality inspection method, system and device for dumb resource equipment |
| CN111507976A (en) * | 2020-04-28 | 2020-08-07 | 浙江省北大信息技术高等研究院 | Defect detection method and system based on multi-angle imaging |
| CN111642478A (en) * | 2020-05-19 | 2020-09-11 | 重庆赛格盾科技有限公司 | Pest identification system |
| US10861148B2 (en) | 2018-04-30 | 2020-12-08 | General Electric Company | Systems and methods for improved component inspection |
| CN112528860A (en) * | 2020-12-11 | 2021-03-19 | 浙江华电器材检测研究所有限公司 | Safety tool management method and system based on image recognition |
| CN112581462A (en) * | 2020-12-25 | 2021-03-30 | 北京邮电大学 | Method and device for detecting appearance defects of industrial products and storage medium |
| CN113245238A (en) * | 2021-05-13 | 2021-08-13 | 苏州迪宏人工智能科技有限公司 | Elbow welding flaw detection method, device and system |
| CN113267501A (en) * | 2020-02-17 | 2021-08-17 | 远传电信股份有限公司 | Sheet material rapid defect detection integration system and use method thereof |
| CN114088730A (en) * | 2022-01-24 | 2022-02-25 | 心鉴智控(深圳)科技有限公司 | Method and system for detecting aluminum-plastic bubble cap defects by using image processing |
| CN114264668A (en) * | 2022-03-03 | 2022-04-01 | 心鉴智控(深圳)科技有限公司 | Method and system for detecting three-stage flaws of medical packaging box by image processing |
| TWI843045B (en) * | 2022-01-19 | 2024-05-21 | 鴻海精密工業股份有限公司 | Image processing and detection method, computer device, and storage medium |
| TWI848462B (en) * | 2021-12-08 | 2024-07-11 | 大陸商先進半導體材料（深圳）有限公司 | Lead frame defect detection method |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850858A (en) * | 2015-05-15 | 2015-08-19 | 华中科技大学 | Injection-molded product defect detection and recognition method |
| CN104992449A (en) * | 2015-08-06 | 2015-10-21 | 西安冉科信息技术有限公司 | Information identification and surface defect on-line detection method based on machine visual sense |
| CN106022232A (en) * | 2016-05-12 | 2016-10-12 | 成都新舟锐视科技有限公司 | License plate detection method based on deep learning |
| CN106845529A (en) * | 2016-12-30 | 2017-06-13 | 北京柏惠维康科技有限公司 | Image feature recognition methods based on many visual field convolutional neural networks |
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A metal strip surface defect detection method and device based on Faster R-CNN network |
-
2017
- 2017-08-21 CN CN201710719954.1A patent/CN107657603B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104850858A (en) * | 2015-05-15 | 2015-08-19 | 华中科技大学 | Injection-molded product defect detection and recognition method |
| CN104992449A (en) * | 2015-08-06 | 2015-10-21 | 西安冉科信息技术有限公司 | Information identification and surface defect on-line detection method based on machine visual sense |
| CN106022232A (en) * | 2016-05-12 | 2016-10-12 | 成都新舟锐视科技有限公司 | License plate detection method based on deep learning |
| CN106845529A (en) * | 2016-12-30 | 2017-06-13 | 北京柏惠维康科技有限公司 | Image feature recognition methods based on many visual field convolutional neural networks |
| CN106952250A (en) * | 2017-02-28 | 2017-07-14 | 北京科技大学 | A metal strip surface defect detection method and device based on Faster R-CNN network |
Cited By (28)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108227664A (en) * | 2018-02-05 | 2018-06-29 | 华侨大学 | Control of product quality equipment and method of quality control based on sample data training |
| CN108672316A (en) * | 2018-03-27 | 2018-10-19 | 哈尔滨理工大学 | A kind of micro parts quality detecting system based on convolutional neural networks |
| US10861148B2 (en) | 2018-04-30 | 2020-12-08 | General Electric Company | Systems and methods for improved component inspection |
| CN110659662A (en) * | 2018-06-29 | 2020-01-07 | 由田新技股份有限公司 | Flaw detection system and method using artificial intelligence |
| CN110659662B (en) * | 2018-06-29 | 2023-05-30 | 由田新技股份有限公司 | Flaw detection system and method using artificial intelligence |
| CN110738630A (en) * | 2018-07-02 | 2020-01-31 | 由田新技股份有限公司 | Training method and detection system of recursive deep learning system |
| CN111077166A (en) * | 2018-10-19 | 2020-04-28 | 北京金山云网络技术有限公司 | Liquid crystal screen defect detection method, device and terminal equipment |
| CN111103306A (en) * | 2018-10-29 | 2020-05-05 | 所罗门股份有限公司 | Method for detecting and marking defects |
| CN110310260B (en) * | 2019-06-19 | 2021-10-15 | 北京百度网讯科技有限公司 | Material distribution decision method, equipment and storage medium based on machine learning model |
| CN110310260A (en) * | 2019-06-19 | 2019-10-08 | 北京百度网讯科技有限公司 | Sub-material decision-making technique, equipment and storage medium based on machine learning model |
| CN110335272A (en) * | 2019-07-12 | 2019-10-15 | 创新奇智(广州)科技有限公司 | A kind of metal device intelligence quality inspection apparatus control system, method and electronic equipment based on machine vision |
| CN110766664B (en) * | 2019-09-29 | 2021-09-14 | 杭州电子科技大学 | Method for detecting appearance defective products of electronic components based on deep learning |
| CN110766664A (en) * | 2019-09-29 | 2020-02-07 | 杭州电子科技大学 | A detection method for defective appearance of electronic components based on deep learning |
| CN111093140A (en) * | 2019-12-11 | 2020-05-01 | 上海闻泰信息技术有限公司 | Method, device, equipment and storage medium for detecting defects of microphone and earphone dust screen |
| CN111259887B (en) * | 2020-01-08 | 2021-08-06 | 网思科技股份有限公司 | Intelligent quality inspection method, system and device for dumb resource equipment |
| CN111259887A (en) * | 2020-01-08 | 2020-06-09 | 网思科技股份有限公司 | Intelligent quality inspection method, system and device for dumb resource equipment |
| CN113267501A (en) * | 2020-02-17 | 2021-08-17 | 远传电信股份有限公司 | Sheet material rapid defect detection integration system and use method thereof |
| CN111507976B (en) * | 2020-04-28 | 2023-08-18 | 杭州未名信科科技有限公司 | Defect detection method and system based on multi-angle imaging |
| CN111507976A (en) * | 2020-04-28 | 2020-08-07 | 浙江省北大信息技术高等研究院 | Defect detection method and system based on multi-angle imaging |
| CN111642478B (en) * | 2020-05-19 | 2022-02-01 | 重庆赛格盾科技有限公司 | Pest identification system |
| CN111642478A (en) * | 2020-05-19 | 2020-09-11 | 重庆赛格盾科技有限公司 | Pest identification system |
| CN112528860A (en) * | 2020-12-11 | 2021-03-19 | 浙江华电器材检测研究所有限公司 | Safety tool management method and system based on image recognition |
| CN112581462A (en) * | 2020-12-25 | 2021-03-30 | 北京邮电大学 | Method and device for detecting appearance defects of industrial products and storage medium |
| CN113245238A (en) * | 2021-05-13 | 2021-08-13 | 苏州迪宏人工智能科技有限公司 | Elbow welding flaw detection method, device and system |
| TWI848462B (en) * | 2021-12-08 | 2024-07-11 | 大陸商先進半導體材料（深圳）有限公司 | Lead frame defect detection method |
| TWI843045B (en) * | 2022-01-19 | 2024-05-21 | 鴻海精密工業股份有限公司 | Image processing and detection method, computer device, and storage medium |
| CN114088730A (en) * | 2022-01-24 | 2022-02-25 | 心鉴智控(深圳)科技有限公司 | Method and system for detecting aluminum-plastic bubble cap defects by using image processing |
| CN114264668A (en) * | 2022-03-03 | 2022-04-01 | 心鉴智控(深圳)科技有限公司 | Method and system for detecting three-stage flaws of medical packaging box by image processing |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107657603B (en) | 2020-07-14 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107657603B (en) | Industrial appearance detection method based on intelligent vision | |
| CN108074231B (en) | A method for surface defect detection of magnetic sheet based on convolutional neural network | |
| CN109859207B (en) | A defect detection method for high-density flexible substrates | |
| CN105678332B (en) | Converter steelmaking end point judgment method and system based on flame image CNN recognition modeling | |
| CN113239930B (en) | Glass paper defect identification method, system, device and storage medium | |
| CN109239102A (en) | A kind of flexible circuit board open defect detection method based on CNN | |
| CN110473173A (en) | A kind of defect inspection method based on deep learning semantic segmentation | |
| CN110806736A (en) | Method for detecting quality information of forge pieces of die forging forming intelligent manufacturing production line | |
| CN111179273A (en) | A method and system for automatic segmentation of leukocyte nucleus and cytoplasm based on deep learning | |
| CN110927171A (en) | A method for detecting defects on the chamfered surface of bearing rollers based on machine vision | |
| CN113822889B (en) | Method for detecting surface defects of hot-rolled steel plate | |
| CN113222913B (en) | Circuit board defect detection positioning method, device and storage medium | |
| CN108764134A (en) | A kind of automatic positioning of polymorphic type instrument and recognition methods suitable for crusing robot | |
| CN107622276B (en) | Deep learning training method based on combination of robot simulation and physical sampling | |
| CN110009617A (en) | An online quality inspection method for shaft parts | |
| CN111127417B (en) | Printing defect detection method based on SIFT feature matching and SSD algorithm improvement | |
| CN113516651A (en) | Welding joint defect detection method and device based on residual error network | |
| CN112819748A (en) | Training method and device for strip steel surface defect recognition model | |
| CN111784688A (en) | Flower automatic grading method based on deep learning | |
| CN110020691A (en) | LCD screen defect inspection method based on the training of convolutional neural networks confrontation type | |
| CN111879972A (en) | Method and system for detection of workpiece surface defects based on SSD network model | |
| Liu et al. | Research on surface defect detection based on semantic segmentation | |
| CN111178405A (en) | A Similar Object Recognition Method Integrating Multiple Neural Networks | |
| CN114998357B (en) | Industrial detection method, system, terminal and medium based on multi-information analysis | |
| CN114820543A (en) | Defect detection method and device |
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 |