CN108876781A - Surface defect recognition method based on SSD algorithm - Google Patents

Surface defect recognition method based on SSD algorithm Download PDF

Info

Publication number
CN108876781A
CN108876781A CN201810669959.2A CN201810669959A CN108876781A CN 108876781 A CN108876781 A CN 108876781A CN 201810669959 A CN201810669959 A CN 201810669959A CN 108876781 A CN108876781 A CN 108876781A
Authority
CN
China
Prior art keywords
image
convolutional network
depth convolutional
defect
layer
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN201810669959.2A
Other languages
Chinese (zh)
Inventor
王宏杰
黄运保
李海艳
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guangdong University of Technology
Original Assignee
Guangdong University of Technology
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by Guangdong University of Technology filed Critical Guangdong University of Technology
Priority to CN201810669959.2A priority Critical patent/CN108876781A/en
Publication of CN108876781A publication Critical patent/CN108876781A/en
Pending legal-status Critical Current

Links

Classifications

    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00Image analysis
    • G06T7/0002Inspection of images, e.g. flaw detection
    • G06T7/0004Industrial image inspection
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84Systems specially adapted for particular applications
    • G01N21/88Investigating the presence of flaws or contamination
    • G01N21/8851Scan 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
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06NCOMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00Computing arrangements based on biological models
    • G06N3/02Neural networks
    • G06N3/04Architecture, e.g. interconnection topology
    • G06N3/045Combinations of networks
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84Systems specially adapted for particular applications
    • G01N21/88Investigating the presence of flaws or contamination
    • G01N21/8851Scan 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
    • G01N2021/8854Grading and classifying of flaws
    • G01N2021/8877Proximity analysis, local statistics
    • GPHYSICS
    • G01MEASURING; TESTING
    • G01NINVESTIGATING OR ANALYSING MATERIALS BY DETERMINING THEIR CHEMICAL OR PHYSICAL PROPERTIES
    • G01N21/00Investigating or analysing materials by the use of optical means, i.e. using sub-millimetre waves, infrared, visible or ultraviolet light
    • G01N21/84Systems specially adapted for particular applications
    • G01N21/88Investigating the presence of flaws or contamination
    • G01N21/8851Scan 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
    • G01N2021/8887Scan 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
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20081Training; Learning
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/20Special algorithmic details
    • G06T2207/20084Artificial neural networks [ANN]
    • GPHYSICS
    • G06COMPUTING; CALCULATING OR COUNTING
    • G06TIMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00Indexing scheme for image analysis or image enhancement
    • G06T2207/30Subject of image; Context of image processing
    • G06T2207/30108Industrial image inspection
    • G06T2207/30164Workpiece; Machine component

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Data Mining & Analysis (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computational Linguistics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Signal Processing (AREA)
  • Chemical & Material Sciences (AREA)
  • Analytical Chemistry (AREA)
  • Biochemistry (AREA)
  • Immunology (AREA)
  • Pathology (AREA)
  • Quality & Reliability (AREA)
  • Image Analysis (AREA)

Abstract

The present invention provides a kind of surface defect recognition method based on SSD algorithm, includes the following steps:Handware defect image is acquired, defect image library is obtained;The defects of training set image input depth convolutional network is trained, obtains characteristics of image figure by construction depth convolutional network;Defects detection is carried out to characteristics of image figure, carries out error calculation;According to error update parameter, the deconditioning when error is less than setting error threshold obtains the depth convolutional network trained;The depth convolutional network that the input of the defects of test machine image has been trained is tested, judges whether image error meets the requirements;If so, the depth convolutional network that selection test image input has been trained is tested, recognition result is obtained.A kind of surface defect recognition method based on SSD algorithm provided by the invention carries out intelligentized identification to handware surface defect by training depth convolutional network, and high degree of automation effectively improves the efficiency to handware surface defects detection.

Description

Surface defect recognition method based on SSD algorithm
Technical field
The present invention relates to Machine Vision Detection fields, more particularly to a kind of Surface Defect Recognition based on SSD algorithm Method.
Background technique
Machine vision is also known as computer vision, is that research uses camera and computer to imitate human eye and brain respectively, with Just it replaces people to detect and judge with machine, completes the science of the tasks such as target identification and industrial detection.Machine vision technique collection A multi-disciplinary applied technology-oriented discipline such as Digital Image Processing, artificial intelligence, computer graphics is closed, in automatic metaplasia It is widely used in production.In recent years, constantly improve with the progress of computer technology and neural network theory, has pushed computer The fast development of vision.China's machine vision industry rapidly develops, and occupies in automated production detection field highly important Status.
It is easily obtained since handware has many advantages, such as easy to form, light weight, material, is suitble to produce in enormous quantities, be in The fields such as electricity, machinery, chemical industry, aviation are with very extensive.With more and more wider, the rapid shaping processing technology of application of handware Development be getting faster, requirement of the people to the quality of handware is also higher and higher.The quality of handware mainly has size, appearance Etc. requirement.Appearance is to guarantee an important link of handware quality, and artificial detection is generallyd use in actual production Mode carry out.Manual detection mode inefficiency, the degree of automation be not high, and accuracy rate is often passed through with the work of testing staff It tests related with attitude.Currently, handware production enterprise increasingly focuses on improving production automation level, to production efficiency It is required that higher and higher, manual detection mode is increasingly unable to meet demand.In addition, in the process of production and processing, due to raw material object Property Parameters variation, technological parameter be unreasonable and the factors such as processing equipment performance is bad, and hardware, which will appear, damages, sand holes, scrapes Wound lacks material, deformation, point, greasy dirt etc. surface defect.These surface defects can not only destroy the appearance of hardware, but also Will affect its performance lead to not using.The surface defects detection of current hardware and identification it is main manually based on, It is inefficient, the degree of automation is low.
Summary of the invention
The present invention is to overcome existing handware by the way of artificial detection, and there are inefficiency, the degree of automation are low The not high technological deficiency with accuracy provides a kind of surface defect recognition method based on SSD algorithm.
In order to solve the above technical problems, technical scheme is as follows:
A kind of surface defect recognition method based on SSD algorithm, includes the following steps:
S1:Handware defect image is acquired, defect image library is obtained;
S2:Defect image library is divided into training set and test set, defect type label is stamped on every image;
S3:The defects of training set image input depth convolutional network is trained, obtains image by construction depth convolutional network Characteristic pattern;
S4:Defects detection is carried out to characteristics of image figure, testing result is obtained compared with label and carries out error calculation;According to error Depth convolutional network parameter is updated, the deconditioning when error is less than depth convolutional network setting error threshold obtains having trained Depth convolutional network;
S5:The depth convolutional network that the input of the defects of test machine image has been trained is tested, whether judges image error It meets the requirements;If so, executing S7;If it is not, executing S3 re -training depth convolutional network;
S6:The depth convolutional network that selection test image input has been trained is tested, and recognition result is obtained.
Wherein, the defect image of handware is acquired in the step S1 by camera.
Wherein, in the step S2, defect image library is divided into training set and test set and accounts for defect image library picture number respectively 80% and 20%, each image all corresponds to the label of defective classification.
Wherein, in step s3, the depth convolutional network includes convolutional layer, excitation layer, pond layer;Wherein, the volume Lamination, excitation layer have multiple, and convolutional layer connects excitation layer, and excitation layer is followed by convolutional layer, excitation layer connect again after convolutional layer, through excessive After a continuous connection, excitation layer is connect with pond layer.
Wherein, the step S4 is specifically included:
S41:It is searched on low-dimensional characteristic pattern using small sliding window, by the defects of sliding window feature and the defect characteristic that in advance extracts It is compared, obtains similarity, if similarity reaches the threshold value of setting, illustrate that the feature in sliding window belongs to lacking for a certain classification Feature is fallen into, which includes the defect of a certain classification, to obtain image deflects;
S42:Obtained image deflects are compared with label, calculate one batch sample of input using batch gradient descent method The error obtained afterwards carries out right value update to depth convolutional network parameter according to loss function, until error is less than the threshold of setting Value exports the depth convolutional network trained.
Compared with prior art, the beneficial effect of technical solution of the present invention is:
A kind of surface defect recognition method based on SSD algorithm provided by the invention, by training depth convolutional network to hardware Part surface defect carries out intelligentized identification, and high degree of automation effectively improves the effect to handware surface defects detection Rate.
The present invention is also equipped with beneficial effect:Depth convolutional network is used in this algorithm, can extract number of drawbacks spy Sign eliminates complicated image processing algorithm, by increasing network depth, can extract more features, improve defect point The accuracy cut;It is identified on convolution characteristic pattern by search characteristics, the use of classifier is omitted, improves recognition efficiency, Significantly reduce calculation amount.
Detailed description of the invention
Fig. 1 is the surface defect recognition method flow chart based on SSD algorithm.
Fig. 2 is depth convolutional network feature structure figure.
Fig. 3 is the region connection schematic diagram of hidden layer neuron and input layer 3X3.
Fig. 4 is the functional image of ReLU function.
Fig. 5 is pond layer workflow schematic diagram.
Fig. 6 is characterized schematic diagram when figure carries out defects detection.
Specific embodiment
The attached figures are only used for illustrative purposes and cannot be understood as limitating the patent;
In order to better illustrate this embodiment, the certain components of attached drawing have omission, zoom in or out, and do not represent the ruler of actual product It is very little;
To those skilled in the art, the omitting of some known structures and their instructions in the attached drawings are understandable.
The following further describes the technical solution of the present invention with reference to the accompanying drawings and examples.
As shown in Figure 1, a kind of surface defect recognition method based on SSD algorithm, includes the following steps:
S1:Handware defect image is acquired, defect image library is obtained;
S2:Defect image library is divided into training set and test set, defect type label is stamped on every image;
S3:The defects of training set image input depth convolutional network is trained, obtains image by construction depth convolutional network Characteristic pattern;
S4:Defects detection is carried out to characteristics of image figure, testing result is obtained compared with label and carries out error calculation;According to error Depth convolutional network parameter is updated, the deconditioning when error is less than depth convolutional network setting error threshold obtains having trained Depth convolutional network;
S5:The depth convolutional network that the input of the defects of test machine image has been trained is tested, whether judges image error It meets the requirements;If so, executing S7;If it is not, executing S3 re -training depth convolutional network;
S6:The depth convolutional network that selection test image input has been trained is tested, and recognition result is obtained.
More specifically, the defect image of handware is acquired in the step S1 by camera.
More specifically, in the step S2, defect image library is divided into training set and test set and accounts for defect image library figure respectively As the 80% of number and 20%, each image all corresponds to the label of defective classification.
More specifically, in step s3, the depth convolutional network includes convolutional layer, excitation layer, pond layer;Wherein, institute State convolutional layer, excitation layer has multiple, convolutional layer connects excitation layer, and excitation layer is followed by convolutional layer, connects excitation layer after convolutional layer again, passes through It crosses after multiple continuous connections, excitation layer is connect with pond layer.
More specifically, the step S4 is specifically included:
S41:It is searched on low-dimensional characteristic pattern using small sliding window, by the defects of sliding window feature and the defect characteristic that in advance extracts It is compared, obtains similarity, if similarity reaches the threshold value of setting, illustrate that the feature in sliding window belongs to lacking for a certain classification Feature is fallen into, which includes the defect of a certain classification, to obtain image deflects;
S42:Obtained image deflects are compared with label, calculate one batch sample of input using batch gradient descent method The error obtained afterwards carries out right value update to depth convolutional network parameter according to loss function, until error is less than the threshold of setting Value exports the depth convolutional network trained.
Embodiment 1
Depth convolutional network feature structure as shown in Figure 2, including convolutional layer, excitation layer and pond layer.Under normal conditions, neural Input layer is indicated using a series of neurons in network, indicates more vivid with two-dimensional matrix in depth convolutional network Intuitively.As conventional neural networks, the neuron needs of input layer are connected with the neuron of hidden layer.But depth convolution Each input neuron is not connect by network with each hidden neuron, only in the partial zones of an image Domain creation connection.
As shown in figure 3, by taking size is the image of 7X7 as an example, if the one of the neuron of first hidden layer and input layer The region of a 3X3 connects.The region is called local sensing domain, 9 neurons in the local sensing domain and first hidden layer The same neuron connects, and has a weight in each connection, therefore local sensing domain shares 3X3 weight.If by local Domain is perceived along from left to right, sequence from top to bottom is slided, and will obtain neuron different in corresponding hidden layer, Fig. 3 is only Show only the connection of first neuron and input layer of first hidden layer.A benefit is slided to the right in local sensing domain It repays, is set as 2 here, will just input layer data and give second hidden neuron.It successively carries out, data can be completed from defeated Enter the transmission of layer to hidden layer.3X3 neuron in first hidden layer obtained above all uses same 3X3 weight, this Referred to as weight shares principle.In addition, each hidden neuron shares a deviation b, referred to as shared deviation.
The input of excitation layer neuron is similar with hidden layer input.Hidden layer data x is multiplied along with deviation b is obtained with weight w It is inputted to excitation layer, i.e.,.By the value input stimulus function of y, ReLU function, functional image are selected here As shown in Figure 4.
As shown in figure 5, pond layer uses max-pooling method, simplify the output of convolutional layer, the pond unit is simply By the maximum excitation output in the input domain of a 2X2.
As shown in fig. 6, construction depth convolutional network carries out defects detection.Depth convolutional network mainly has convolutional layer, excitation Layer, pond layer composition.Excitation layer is connected behind convolution, behind reconnect convolutional layer.By continuous several convolutional layers and excitation layer A pond layer can be connect later.After such structure is repeated several times, depth convolutional network is just generated.
It is formed after depth convolutional network, defect image is inputted into depth convolutional network, then can be obtained by image Characteristic pattern.It is searched on low-dimensional characteristic pattern using small sliding window, the defect characteristic that the feature in sliding window is extracted in advance is compared Compared with obtaining similarity.If similarity reaches the threshold value being previously set, the feature being considered as in the sliding window belongs to a certain classification Defect characteristic, the image include the defect of a certain classification.Mainly identify small defect, such as point on low-dimensional characteristic pattern, sand holes, this Sample prevents network too deep, has desalinated small defect.Utilize big sliding window in characteristic pattern in last characteristic pattern of depth network Upper search, main purpose are the big defects of detection, and such as greasy dirt lacks material, scratches.
Embodiment 2
In network training process, first the weight of network is initialized, and weight distribution is made to submit to mean value 0, variance For 0.01 Gaussian Profile.So that number of the weight greater than 0 is approximately equal to the number less than 0 simultaneously, it is a certain to be beneficial to prevent network item Whole handware images is divided into two major classes, i.e. training set and test set, accounts for 80% and 20% respectively by direction inclination.Every width figure As all corresponding to a label, the i.e. classification of defect.The sample for having label input network is subjected to the training for having supervision, uses batch Gradient descent method calculates the error obtained after one batch sample of input, according to the error update weight of loss function, until accidentally Difference is less than the threshold value of setting.
Obviously, the above embodiment of the present invention be only to clearly illustrate example of the present invention, and not be pair The restriction of embodiments of the present invention.For those of ordinary skill in the art, may be used also on the basis of the above description To make other variations or changes in different ways.There is no necessity and possibility to exhaust all the enbodiments.It is all this Made any modifications, equivalent replacements, and improvements etc., should be included in the claims in the present invention within the spirit and principle of invention Protection scope within.

Claims (5)

1. a kind of surface defect recognition method based on SSD algorithm, which is characterized in that include the following steps:
S1:Handware defect image is acquired, defect image library is obtained;
S2:Defect image library is divided into training set and test set, defect type label is stamped on every image;
S3:The defects of training set image input depth convolutional network is trained, obtains image by construction depth convolutional network Characteristic pattern;
S4:Defects detection is carried out to characteristics of image figure, testing result is obtained compared with label and carries out error calculation;According to error Depth convolutional network parameter is updated, the deconditioning when error is less than depth convolutional network setting error threshold obtains having trained Depth convolutional network;
S5:The depth convolutional network that the input of the defects of test machine image has been trained is tested, whether judges image error It meets the requirements;If so, executing S7;If it is not, executing S3 re -training depth convolutional network;
S6:The depth convolutional network that selection test image input has been trained is tested, and recognition result is obtained.
2. the surface defect recognition method according to claim 1 based on SSD algorithm, it is characterised in that:The step S1 In by camera acquire handware defect image.
3. the surface defect recognition method according to claim 1 based on SSD algorithm, it is characterised in that:The step S2 In, defect image library is divided into 80% and 20% that training set and test set account for defect image library picture number respectively, and each image is all right Answer the label of defective classification.
4. the surface defect recognition method according to claim 1 based on SSD algorithm, it is characterised in that:In step s3, The depth convolutional network includes convolutional layer, excitation layer, pond layer;Wherein, the convolutional layer, excitation layer have multiple, convolutional layer Excitation layer is connected, excitation layer is followed by convolutional layer, connects excitation layer again after convolutional layer, after multiple continuous connections, excitation layer It is connect with pond layer.
5. the surface defect recognition method according to claim 1 based on SSD algorithm, it is characterised in that:The step S4 It specifically includes:
S41:It is searched on low-dimensional characteristic pattern using small sliding window, by the defects of sliding window feature and the defect characteristic that in advance extracts It is compared, obtains similarity, if similarity reaches the threshold value of setting, illustrate that the feature in sliding window belongs to lacking for a certain classification Feature is fallen into, which includes the defect of a certain classification, to obtain image deflects;
S42:Obtained image deflects are compared with label, calculate one batch sample of input using batch gradient descent method The error obtained afterwards carries out right value update to depth convolutional network parameter according to loss function, until error is less than the threshold of setting Value exports the depth convolutional network trained.
CN201810669959.2A 2018-06-26 2018-06-26 Surface defect recognition method based on SSD algorithm Pending CN108876781A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810669959.2A CN108876781A (en) 2018-06-26 2018-06-26 Surface defect recognition method based on SSD algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810669959.2A CN108876781A (en) 2018-06-26 2018-06-26 Surface defect recognition method based on SSD algorithm

Publications (1)

Publication Number Publication Date
CN108876781A true CN108876781A (en) 2018-11-23

Family

ID=64294961

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810669959.2A Pending CN108876781A (en) 2018-06-26 2018-06-26 Surface defect recognition method based on SSD algorithm

Country Status (1)

Country Link
CN (1) CN108876781A (en)

Cited By (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109800697A (en) * 2019-01-09 2019-05-24 国网浙江省电力有限公司舟山供电公司 Transformer target detection and open defect recognition methods based on VGG-net Style Transfer
CN109886914A (en) * 2018-12-19 2019-06-14 刘咏晨 Paper sheet defect detection method based on local luminance invariance priori
CN110176001A (en) * 2019-06-03 2019-08-27 浙江大学 A kind of high iron catenary insulator breakage accurate positioning method based on Grad-CAM algorithm
CN110197477A (en) * 2019-05-07 2019-09-03 北京邮电大学 The method, apparatus and system of pavement crack detection
CN110378618A (en) * 2019-07-26 2019-10-25 杭州安脉盛智能技术有限公司 Quality evaluating method and system based on online pipe tobacco surface defects detection
CN110598767A (en) * 2019-08-29 2019-12-20 河南省收费还贷高速公路管理有限公司航空港分公司 SSD convolutional neural network-based underground drainage pipeline defect identification method
CN111079628A (en) * 2019-12-12 2020-04-28 哈尔滨市科佳通用机电股份有限公司 Floor damage fault image identification method
CN111127417A (en) * 2019-12-20 2020-05-08 江苏理工学院 Soft package coil stock printing defect detection method based on SIFT feature matching and improved SSD algorithm
CN111652227A (en) * 2020-05-21 2020-09-11 哈尔滨市科佳通用机电股份有限公司 Method for detecting damage fault of bottom floor of railway wagon
CN113111911A (en) * 2021-03-12 2021-07-13 东南大学 Defect depth detection method based on principal component analysis and gate control circulation unit network

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106568783A (en) * 2016-11-08 2017-04-19 广东工业大学 Hardware part defect detecting system and method
CN107481231A (en) * 2017-08-17 2017-12-15 广东工业大学 A kind of handware defect classifying identification method based on depth convolutional neural networks
CN107833220A (en) * 2017-11-28 2018-03-23 河海大学常州校区 Fabric defect detection method based on depth convolutional neural networks and vision significance
CN107966447A (en) * 2017-11-14 2018-04-27 浙江大学 A kind of Surface Flaw Detection method based on convolutional neural networks
CN108154508A (en) * 2018-01-09 2018-06-12 北京百度网讯科技有限公司 Method, apparatus, storage medium and the terminal device of product defects detection positioning

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106568783A (en) * 2016-11-08 2017-04-19 广东工业大学 Hardware part defect detecting system and method
CN107481231A (en) * 2017-08-17 2017-12-15 广东工业大学 A kind of handware defect classifying identification method based on depth convolutional neural networks
CN107966447A (en) * 2017-11-14 2018-04-27 浙江大学 A kind of Surface Flaw Detection method based on convolutional neural networks
CN107833220A (en) * 2017-11-28 2018-03-23 河海大学常州校区 Fabric defect detection method based on depth convolutional neural networks and vision significance
CN108154508A (en) * 2018-01-09 2018-06-12 北京百度网讯科技有限公司 Method, apparatus, storage medium and the terminal device of product defects detection positioning

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
LIU, W ET AL: "SSD: Single Shot MultiBox Detector", 《 IN PROCEEDINGS OF THE EUROPEAN CONFERENCE ON COMPUTER VISION》 *
颜伟鑫: "深度学习及其在工件缺陷自动检测中的应用研究", 《中国优秀硕士论文全文数据库》 *

Cited By (16)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109886914A (en) * 2018-12-19 2019-06-14 刘咏晨 Paper sheet defect detection method based on local luminance invariance priori
CN109886914B (en) * 2018-12-19 2020-01-17 浙江企银印务科技有限公司 Paper defect detection method based on local brightness invariance prior
CN109800697A (en) * 2019-01-09 2019-05-24 国网浙江省电力有限公司舟山供电公司 Transformer target detection and open defect recognition methods based on VGG-net Style Transfer
CN109800697B (en) * 2019-01-09 2022-10-25 国网浙江省电力有限公司舟山供电公司 Transformer target detection and appearance defect identification method based on VGG-net style migration
CN110197477A (en) * 2019-05-07 2019-09-03 北京邮电大学 The method, apparatus and system of pavement crack detection
CN110176001B (en) * 2019-06-03 2021-09-03 浙江大学 Grad-CAM algorithm-based high-speed rail contact net insulator damage accurate positioning method
CN110176001A (en) * 2019-06-03 2019-08-27 浙江大学 A kind of high iron catenary insulator breakage accurate positioning method based on Grad-CAM algorithm
CN110378618A (en) * 2019-07-26 2019-10-25 杭州安脉盛智能技术有限公司 Quality evaluating method and system based on online pipe tobacco surface defects detection
CN110598767A (en) * 2019-08-29 2019-12-20 河南省收费还贷高速公路管理有限公司航空港分公司 SSD convolutional neural network-based underground drainage pipeline defect identification method
CN111079628A (en) * 2019-12-12 2020-04-28 哈尔滨市科佳通用机电股份有限公司 Floor damage fault image identification method
CN111127417A (en) * 2019-12-20 2020-05-08 江苏理工学院 Soft package coil stock printing defect detection method based on SIFT feature matching and improved SSD algorithm
CN111127417B (en) * 2019-12-20 2024-02-06 江苏理工学院 Printing defect detection method based on SIFT feature matching and SSD algorithm improvement
CN111652227A (en) * 2020-05-21 2020-09-11 哈尔滨市科佳通用机电股份有限公司 Method for detecting damage fault of bottom floor of railway wagon
CN111652227B (en) * 2020-05-21 2021-01-22 哈尔滨市科佳通用机电股份有限公司 Method for detecting damage fault of bottom floor of railway wagon
CN113111911A (en) * 2021-03-12 2021-07-13 东南大学 Defect depth detection method based on principal component analysis and gate control circulation unit network
CN113111911B (en) * 2021-03-12 2024-02-06 东南大学 Defect depth detection method based on principal component analysis and gating circulation unit network

Similar Documents

Publication Publication Date Title
CN108876781A (en) Surface defect recognition method based on SSD algorithm
Kumar et al. Resnet-based approach for detection and classification of plant leaf diseases
CN107481231A (en) A kind of handware defect classifying identification method based on depth convolutional neural networks
CN110148104B (en) Infrared and visible light image fusion method based on significance analysis and low-rank representation
CN109740413A (en) Pedestrian recognition methods, device, computer equipment and computer storage medium again
CN104281853B (en) A kind of Activity recognition method based on 3D convolutional neural networks
CN105825235B (en) A kind of image-recognizing method based on multi-characteristic deep learning
Garg et al. Deep convolution neural network approach for defect inspection of textured surfaces
CN111222519B (en) Construction method, method and device of hierarchical colored drawing manuscript line extraction model
CN108537136A (en) The pedestrian's recognition methods again generated based on posture normalized image
CN113469119B (en) Cervical cell image classification method based on visual converter and image convolution network
CN105303150B (en) Realize the method and system of image procossing
CN110363253A (en) A kind of Surfaces of Hot Rolled Strip defect classification method based on convolutional neural networks
CN111539320B (en) Multi-view gait recognition method and system based on mutual learning network strategy
CN108257154A (en) Polarimetric SAR Image change detecting method based on area information and CNN
CN109214298A (en) A kind of Asia women face value Rating Model method based on depth convolutional network
CN109509171A (en) A kind of Fabric Defects Inspection detection method based on GMM and image pyramid
CN111161244B (en) Industrial product surface defect detection method based on FCN + FC-WXGboost
CN109325513B (en) Image classification network training method based on massive single-class images
CN108629370A (en) A kind of classification and identification algorithm and device based on depth confidence network
CN109344845A (en) A kind of feature matching method based on Triplet deep neural network structure
CN109919246A (en) Pedestrian's recognition methods again based on self-adaptive features cluster and multiple risks fusion
Liu et al. Strip steel surface defect detecting method combined with a multi-layer attention mechanism network
Wen et al. Multi-scene citrus detection based on multi-task deep learning network
CN107564013A (en) Merge the scene cut modification method and system of local message

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
RJ01 Rejection of invention patent application after publication
RJ01 Rejection of invention patent application after publication

Application publication date: 20181123