CN109034204B - Weld defect identification method based on improved convolutional neural network - Google Patents

Weld defect identification method based on improved convolutional neural network Download PDF

Info

Publication number
CN109034204B
CN109034204B CN201810709776.9A CN201810709776A CN109034204B CN 109034204 B CN109034204 B CN 109034204B CN 201810709776 A CN201810709776 A CN 201810709776A CN 109034204 B CN109034204 B CN 109034204B
Authority
CN
China
Prior art keywords
pooling
neural network
feature
samples
domain
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
Application number
CN201810709776.9A
Other languages
Chinese (zh)
Other versions
CN109034204A (en
Inventor
姜洪权
高建民
王晓桥
王泉生
夏锋社
贺帅
程雷
李华
昌亚胜
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.)
Xian Jiaotong University
Original Assignee
Xian Jiaotong University
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 Xian Jiaotong University filed Critical Xian Jiaotong University
Priority to CN201810709776.9A priority Critical patent/CN109034204B/en
Publication of CN109034204A publication Critical patent/CN109034204A/en
Application granted granted Critical
Publication of CN109034204B publication Critical patent/CN109034204B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • 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
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • 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

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Artificial Intelligence (AREA)
  • General Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Health & Medical Sciences (AREA)
  • Software Systems (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Mathematical Physics (AREA)
  • Computational Linguistics (AREA)
  • Health & Medical Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Evolutionary Biology (AREA)
  • Image Analysis (AREA)

Abstract

The invention discloses a weld defect identification method based on an improved convolutional neural network, which comprises the steps of establishing a pooling model comprehensively considering pooling domains and characteristic map characteristic distribution of the pooling domains, introducing a correction factor mu to correct maximum pooling characteristics, and combining a Relieff algorithm and the neural network as a characteristic selection method; and constructing a deep convolution neural network with the pooling model and the feature selection method, and iterating by taking a minimized cost function as a target to train and form a weld defect identification neural network so as to realize the identification of the type of the weld defect. The method avoids the process that the traditional welding seam defect identification method needs manual feature extraction, and the defect identification rate of the convolutional neural network model is further improved compared with the traditional method before improvement.

Description

Weld defect identification method based on improved convolutional neural network
Technical Field
The invention belongs to the technical field of automatic identification of weld defects, and particularly relates to a weld defect identification method based on an improved convolutional neural network.
Background
In the field of automatic identification of weld defects, the traditional method inevitably needs to go through a process of manually selecting, extracting and characterizing, the process is time-consuming and labor-consuming, whether the characteristics are reasonably selected has great subjectivity, and the accuracy of identification is greatly influenced.
The convolutional neural network classical pooling models (maximum pooling model and average pooling model) lack dynamic adaptivity when performing feature extraction on different feature distribution pooling domains, resulting in inaccurate extracted features.
The weld defect data belongs to small sample and non-mass data, and the learning of the convolutional neural network is insufficient under the condition of the non-mass data, so that the model cannot achieve the optimal feature selection.
Disclosure of Invention
The technical problem to be solved by the invention is to provide a weld defect identification method based on an improved convolutional neural network aiming at the defects in the prior art, so that the process of manual feature selection in the traditional method is avoided; the classical pooling model and the feature selection method of the neural network are improved to obtain a weld defect identification method based on an improved convolutional neural network, and the weld image to be identified is sent to a trained deep neural network to realize automatic identification of the type of the weld defect.
The invention adopts the following technical scheme:
a weld defect identification method based on an improved convolutional neural network is characterized by establishing a pooling model comprehensively considering pooling domains and characteristic map characteristic distribution of the pooling domains, and introducing a correction factor mu to correct maximum pooling characteristics; combining a Relieff algorithm with a neural network as a feature selection method; and constructing a deep convolution neural network with the pooling model and the feature selection method, performing iteration by taking a minimum cost function as a target to train and form a weld defect identification neural network, and sending the weld image to be identified into the trained deep convolution neural network to realize the identification of the type of the weld defect.
Specifically, the specific steps for constructing the pooling model are as follows:
s101, obtaining a feature variance sigma of a current pooling domainPFeature variance sigma with the feature map of the pooling domainFM；
S102, constructing a pooling model comprehensively considering the pooling domain and the feature distribution of the feature map where the pooling domain is located.
Further, in step S102, the pooling model is expressed as follows:
Figure BDA0001716380660000021
wherein: the size of the pooling domain is nxn, S is the pooling characteristic extracted by the pooling model, FijRepresenting the eigenvalues, σ, of the input eigenmap F at (i, j)PFor the variance, σ, of features in the pooling domainFMMu is a correction factor for the variance of the characteristic diagram of the pooling domain.
Further, the correction factor μ is a ratio of a difference between the sum of the characteristic values and the characteristic range and the sum of the characteristic values in the pooling domain, and is specifically calculated as follows:
μ＝(tsum-tmax+tmin)/tsum
wherein, tsumIs the sum of the eigenvalues in the pooling domain, tminIs the minimum eigenvalue, t, in the pooling domainmaxIs the maximum eigenvalue within the pooling domain.
Specifically, the steps of establishing the feature selection method based on the combination of the Relieff algorithm and the neural network are as follows:
s201, obtaining a sub-sample set M by an intra-class random sampling method according to the proportion between classes;
s202, obtaining a feature set T which is extracted from a sample set and used for classification, and processing the feature set by adopting a Relieff algorithm to obtain an initial feature weight vector L0；
S203, mixing L0Setting the element with negative weight value as 0 to obtain new weight value vector L1Is prepared by mixing L1After normalization processing, the final product is obtainedA final weight vector L;
and S204, correspondingly giving the feature weight in the L to the feature selection layer to obtain a feature selection layer parameter.
Further, in step S201, the inter-class ratio refers to a ratio of the number of times samples in each class are extracted to the total number of times samples are extracted and a ratio of the number of samples in the class to the total number of all samples, and the intra-class random refers to randomly selecting samples in the class C.
Further, the ratio of the number of samples extracted in each class to the total number of samples and the ratio of the number of samples in the class to the total number of all samples are calculated as follows:
Figure BDA0001716380660000031
where n is the total number of samples in class C, m is the total number of samples in all classes, CnFor the number of times the samples in class C were extracted, CmThe total number of draws for all samples.
Compared with the prior art, the invention has at least the following beneficial effects:
the invention relates to a weld defect identification method based on an improved convolutional neural network, which introduces a correction factor mu to correct the maximum pooling characteristic according to the pooling domain and the characteristic distribution of a characteristic diagram where the pooling domain is located, builds up the convolutional neural network, and builds a characteristic selection method based on the combination of a Relieff algorithm and the neural network; iteration is carried out by taking the minimum cost function as a target, a weld defect recognition neural network is formed through training, the weld image to be recognized is sent to the trained deep neural network, the recognition of the type of the weld defect is realized, the process of manually selecting the characteristics is avoided, and the higher correct defect recognition rate is obtained.
Furthermore, the pooling model has important influence on the types of the welding seam defects based on the welding seam defects and the peripheral areas thereof, and comprehensively considers the pooling areas and the characteristic distribution of the characteristic diagram where the pooling areas are located, so that the pooling model has certain adaptivity when facing different pooling areas and has the characteristics of maximum pooling and average pooling.
Furthermore, the traditional feature evaluation method Relieff algorithm is combined with the understanding of the deep military aircraft neural network on feature importance, and the problem that the deep convolutional neural network is difficult to achieve optimal feature selection under non-mass data is solved to a certain extent.
Furthermore, by adopting an inter-class proportion intra-class random sampling method, the phenomenon that the samples are densely distributed in certain classes due to the traditional Relieff algorithm is avoided. The disadvantage of not being able to completely cover the entire sample set.
In conclusion, the method avoids the process that the traditional welding seam defect identification method needs to manually extract features, and the defect identification rate of the convolutional neural network model is further improved compared with that of the traditional improved convolutional neural network model.
The technical solution of the present invention is further described in detail by the accompanying drawings and embodiments.
Drawings
FIG. 1 is a diagram of the steps of the method of the present invention;
FIG. 2 is an image of a weld of an experimental sample;
FIG. 3 is an initial convolutional neural network model;
FIG. 4 is a model of an improved convolutional neural network.
Detailed Description
The invention provides a weld defect identification method based on an improved convolutional neural network, which is used for constructing the convolutional neural network with a specific framework, wherein the specific framework is a pooling model which comprehensively considers the pooling domain and the characteristic distribution of the characteristic diagram where the pooling domain is located is provided on a pooling layer, and on the basis of a classical maximum pooling model, a correction factor mu is introduced to correct the maximum pooling characteristic according to the pooling domain and the characteristic distribution of the characteristic diagram where the pooling domain is located; a feature selection method combining a traditional feature evaluation method, a Relief algorithm and a neural network is provided; and (4) iterating by taking the minimum cost function as a target to train and form the weld defect identification neural network, and sending the weld image to be identified into the trained deep neural network to realize the identification of the weld defect type. The method avoids the process that the traditional welding seam defect identification method needs manual feature extraction, and the defect identification rate of the convolutional neural network model is further improved compared with the traditional method before improvement.
Referring to fig. 1, the method for identifying the weld defect based on the improved convolutional neural network of the present invention includes the following specific steps:
s1, aiming at the characteristics that weld defects and peripheral areas thereof have important influence on defect type identification, and the problem that characteristic extraction is inaccurate in the characteristic extraction process of a classical pooling model (maximum pooling and average pooling) in a convolutional neural network, improving the classical pooling model, and establishing a pooling model comprehensively considering the pooling domain and the characteristic distribution of a characteristic diagram where the pooling domain is located;
s101, obtaining a feature variance sigma of a current pooling domainPFeature variance sigma with the feature map of the pooling domainFM；
S102, constructing a pooling model comprehensively considering the pooling domain and the feature distribution of the feature map where the pooling domain is located, wherein the expression is as follows:
Figure BDA0001716380660000041
wherein: the size of the pooling domain is nxn, S is the pooling characteristic extracted by the pooling model, FijRepresenting the eigenvalues, σ, of the input eigenmap F at (i, j)PFor the variance, σ, of features in the pooling domainFMAnd mu is a correction factor, and the mathematical expression of the variance of the characteristic diagram of the pooling domain is as follows:
μ＝(tsum-tmax+tmin)/tsum
wherein, tsumIs the sum of the eigenvalues in the pooling domain, tminIs the minimum eigenvalue, t, in the pooling domainmaxIs the maximum eigenvalue within the pooling domain.
S2, aiming at the situation that under the condition of non-mass data, the deep neural network model features are not sufficiently learned, and the optimal feature selection cannot be achieved, establishing a feature selection method combining a traditional feature evaluation method Relieff algorithm and a deep neural network;
s201, obtaining a sub-sample set M by an intra-class random sampling method according to the proportion between classes;
in the process of processing the feature set T by applying the Relieff algorithm, a sampling method of 'inter-class proportion and intra-class random' is adopted, and the 'inter-class proportion' refers to the ratio of the number of extracted samples in each class to the total number of sampled samples in the class to the ratio of the number of samples in the class to the total number of all samples, namely:
Figure BDA0001716380660000051
where n is the total number of samples in class C, m is the total number of samples in all classes, CnFor the number of times the samples in class C were extracted, CmTotal number of draws for all samples; "random within class" means that samples are randomly selected in class C.
S202, obtaining a feature set T which is extracted from a sample set and used for classification, and processing the feature set by adopting a Relieff algorithm to obtain an initial feature weight vector L0；
S203, mixing L0Setting the element with negative weight value as 0 to obtain new weight value vector L1Is prepared by mixing L1Carrying out normalization processing to obtain a final weight vector L;
and S204, correspondingly giving the feature weight in the L to the feature selection layer to obtain a feature selection layer parameter.
And S3, iterating by taking the minimum cost function as a target to train and form a weld defect recognition neural network, and inputting the image to be classified into the trained improved convolution neural network to realize the recognition of the type of the weld defect.
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of the embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations. Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The method comprises the steps of preprocessing an original welding seam image, intercepting a 3232 defect-containing region on the original welding seam image as an interested region, taking the image of the interested region as the input of a neural network, and selecting 557 interested region welding seam images, wherein the 557 interested region welding seam images comprise 6 types of images including 134 crack images, 60 unfused images, 144 unfused images, 52 long defect images, 84 circular defects and 74 defect-free images. The images are divided into a training set and a testing set according to the ratio of 4:1, 445 images are obtained for training, and 112 images are obtained for testing. A partial image as an experimental sample is shown in fig. 2.
First, an initial convolutional neural network model (in the figure, C represents a convolutional layer, S represents a pooling layer, and FC represents a fully-connected layer) as shown in fig. 3 is constructed, which is composed of two convolutional-pooling models, two fully-connected layers, and one output layer. And recording the convolutional neural network obtained when the largest pooling model is selected in the pooling layer as CNN-1, recording the convolutional neural network obtained when the average pooling model is selected in the pooling layer as CNN-2, and recording the convolutional neural network obtained when the pooling model provided by the invention is selected in the pooling layer as CNN-3.
Constructing an initial convolutional neural network model as shown in fig. 4: and adding a characteristic selection layer of a Relieff algorithm between the FC6 th layer and an output layer of the initial convolutional neural network, and marking the obtained convolutional neural network as CNN-4.
In order to verify the effectiveness of the pooling model provided by the invention, experiments are carried out on 3 convolutional neural networks CNN-1, CNN-2 and CNN-3, the correct recognition rates of the three models are compared under different iteration times, and the experimental results are shown in the following table 1:
Figure BDA0001716380660000071
experiments prove that the pooling model provided by the invention has higher recognition rate than the maximum pooling model and the average pooling model under the condition of different iteration times, and has more obvious advantages under the condition of less iteration times. In order to verify the effectiveness of the pooling model provided by the invention, the features extracted by the deep neural network are processed by adopting a Relieff algorithm to obtain a weight vector, the negative weight value of the weight vector is set to be 0, then normalization processing is carried out, the weight vector L is shown in table 3, and the obtained weight values of each feature (namely the weight values of each parameter of the feature selection layer in the attached figure 4) under 200 iteration times are as follows:
Figure BDA0001716380660000072
Figure BDA0001716380660000081
experiments are carried out on the convolutional neural network CNN-4, the correct recognition rates of the models on the CNN-1, the CNN-2, the CNN-3 and the CNN-4 are compared under different iteration times, and the experimental results are shown in the following table 3:
Figure BDA0001716380660000082
experiments prove that the recognition rate of the feature selection method provided by the invention is improved to a certain extent before improvement. In summary, the improved convolutional neural network model (including the improved pooling model and the feature selection method) related in the invention has a certain improvement on the seam defect recognition rate before improvement.
The above-mentioned contents are only for illustrating the technical idea of the present invention, and the protection scope of the present invention is not limited thereby, and any modification made on the basis of the technical idea of the present invention falls within the protection scope of the claims of the present invention.

Claims (4)

1. A weld defect identification method based on an improved convolutional neural network is characterized in that a pooling model comprehensively considering the feature distribution of a pooling domain and a feature map where the pooling domain is located is established, and a correction factor mu is introduced to correct the maximum pooling feature; combining a Relieff algorithm with a neural network as a feature selection method; constructing a deep convolution neural network with the pooling model and the feature selection method, performing iteration by taking a minimum cost function as a target to train and form a weld defect identification neural network, and sending a weld image to be identified into the trained deep convolution neural network to realize identification of the type of the weld defect;
the concrete steps for constructing the pooling model are as follows:
s101, obtaining a feature variance sigma of a current pooling domainPFeature variance sigma with the feature map of the pooling domainFM；
S102, constructing a pooling model comprehensively considering the pooling domain and the characteristic distribution of the characteristic diagram where the pooling domain is located, wherein the pooling model is expressed as follows:
Figure FDA0003306059980000011
wherein: the size of the pooling domain is nxn, S is the pooling characteristic extracted by the pooling model, FijRepresenting the eigenvalues, σ, of the input eigenmap F at (i, j)PFor the variance, σ, of features in the pooling domainFMThe variance of the characteristic diagram of the pooling domain is shown, and mu is a correction factor;
the steps for establishing the feature selection method based on the combination of the Relieff algorithm and the neural network are as follows:
s201, obtaining a sub-sample set M by an intra-class random sampling method according to the proportion between classes;
s202, obtaining a feature set T which is extracted from a sample set and used for classification, and processing the feature set by adopting a Relieff algorithm to obtain an initial feature weight vector L0；
S203, mixing L0Setting the element with negative weight value as 0 to obtain new weight value vector L1Is prepared by mixing L1Go on to unityObtaining a final weight vector L after the quantization processing;
and S204, correspondingly giving the feature weight in the L to the feature selection layer to obtain a feature selection layer parameter.
2. The method for identifying the weld defect based on the improved convolutional neural network as claimed in claim 1, wherein in step S102, the correction factor μ is a ratio of a difference between a sum of characteristic values and a range of characteristic differences to the sum of characteristic values in the pooling domain, and is specifically calculated as follows:
μ＝(tsum-tmax+tmin)/tsum
wherein, tsumIs the sum of the eigenvalues in the pooling domain, tminIs the minimum eigenvalue, t, in the pooling domainmaxIs the maximum eigenvalue within the pooling domain.
3. The method for identifying the weld defect based on the improved convolutional neural network as claimed in claim 1, wherein in step S201, the inter-class ratio refers to a ratio of the number of times of samples extracted in each class to the total number of samples extracted in the class to a ratio of the number of samples in the class to the total number of all samples, and the intra-class random refers to randomly selecting samples in the class C.
4. The method for identifying the weld defects based on the improved convolutional neural network as claimed in claim 3, wherein the ratio of the number of the samples extracted in each category to the total number of the samples in the category to the total number of all the samples is calculated as follows:
Figure FDA0003306059980000021
where n is the total number of samples in class C, m is the total number of samples in all classes, CnFor the number of times the samples in class C were extracted, CmThe total number of draws for all samples.
CN201810709776.9A 2018-07-02 2018-07-02 Weld defect identification method based on improved convolutional neural network Active CN109034204B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201810709776.9A CN109034204B (en) 2018-07-02 2018-07-02 Weld defect identification method based on improved convolutional neural network

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201810709776.9A CN109034204B (en) 2018-07-02 2018-07-02 Weld defect identification method based on improved convolutional neural network

Publications (2)

Publication Number Publication Date
CN109034204A CN109034204A (en) 2018-12-18
CN109034204B true CN109034204B (en) 2022-01-25

Family

ID=65521240

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201810709776.9A Active CN109034204B (en) 2018-07-02 2018-07-02 Weld defect identification method based on improved convolutional neural network

Country Status (1)

Country Link
CN (1) CN109034204B (en)

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2020129617A1 (en) * 2018-12-19 2020-06-25 パナソニックＩｐマネジメント株式会社 Visual inspection device, method for improving accuracy of determination for existence/nonexistence of shape failure of welding portion and kind thereof using same, welding system, and work welding method using same
CN112440039A (en) * 2019-08-31 2021-03-05 南京理工大学 Intelligent photoelectric tracking system and measuring method for welding seam based on multi-line structured light projection
CN110781741A (en) * 2019-09-20 2020-02-11 中国地质大学（武汉） Face recognition method based on Relief feature filtering method
CN110991458B (en) * 2019-11-25 2023-05-23 创新奇智(北京)科技有限公司 Image feature-based artificial intelligent recognition result sampling system and sampling method
CN112819806B (en) * 2021-02-23 2024-05-28 江苏科技大学 Ship weld defect detection method based on deep convolutional neural network model
CN113129266B (en) * 2021-03-22 2022-03-29 太原科技大学 Stainless steel weld defect detection method based on multi-domain expression data enhancement and model self-optimization
CN113362900A (en) * 2021-06-15 2021-09-07 邵阳学院 Mixed model for predicting N4-acetylcytidine
CN115392317A (en) * 2022-08-31 2022-11-25 广东石油化工学院 Petrochemical unit rotating machinery fault diagnosis method, device, equipment and medium

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102779281A (en) * 2012-06-25 2012-11-14 同济大学 Vehicle type identification method based on support vector machine and used for earth inductor
US8516584B2 (en) * 2007-01-29 2013-08-20 Deutsche Telekom Ag Method and system for detecting malicious behavioral patterns in a computer, using machine learning
CN104809069A (en) * 2015-05-11 2015-07-29 中国电力科学研究院 Source node loophole detection method based on integrated neural network
CN107220606A (en) * 2017-05-22 2017-09-29 西安电子科技大学 The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks
CN107766865A (en) * 2017-11-06 2018-03-06 北京旷视科技有限公司 Pond method, object detecting method, device, system and computer-readable medium
CN107871136A (en) * 2017-03-22 2018-04-03 中山大学 The image-recognizing method of convolutional neural networks based on openness random pool

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8516584B2 (en) * 2007-01-29 2013-08-20 Deutsche Telekom Ag Method and system for detecting malicious behavioral patterns in a computer, using machine learning
CN102779281A (en) * 2012-06-25 2012-11-14 同济大学 Vehicle type identification method based on support vector machine and used for earth inductor
CN104809069A (en) * 2015-05-11 2015-07-29 中国电力科学研究院 Source node loophole detection method based on integrated neural network
CN107871136A (en) * 2017-03-22 2018-04-03 中山大学 The image-recognizing method of convolutional neural networks based on openness random pool
CN107220606A (en) * 2017-05-22 2017-09-29 西安电子科技大学 The recognition methods of radar emitter signal based on one-dimensional convolutional neural networks
CN107766865A (en) * 2017-11-06 2018-03-06 北京旷视科技有限公司 Pond method, object detecting method, device, system and computer-readable medium

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
不同池化模型的卷积神经网络学习性能研究;刘万军，梁雪剑，曲海成;《中国图象图形学报》;20160930;第1178-1190页 *
基于X射线图像和卷积神经网络的石油钢管焊缝缺陷检测与识别;刘涵，郭润元;《仪器仪表学报》;20180430;第247-256页 *
基于改进的ReliefF 算法的神经网络集成分类模型;朱远枫，章晶，史娜;《电脑知识与技术》;20090331;第1699-1700页 *

Also Published As

Publication number Publication date
CN109034204A (en) 2018-12-18

Similar Documents

Publication Publication Date Title
CN109034204B (en) Weld defect identification method based on improved convolutional neural network
CN110516596B (en) Octave convolution-based spatial spectrum attention hyperspectral image classification method
CN107633255B (en) Rock lithology automatic identification and classification method under deep learning mode
WO2016165082A1 (en) Image stego-detection method based on deep learning
CN109299741B (en) Network attack type identification method based on multi-layer detection
CN109615604B (en) Part appearance flaw detection method based on image reconstruction convolutional neural network
CN104766097B (en) Surface of aluminum plate defect classification method based on BP neural network and SVMs
CN108492294B (en) Method and device for evaluating harmony degree of image colors
CN113486981A (en) RGB image classification method based on multi-scale feature attention fusion network
CN108596892A (en) A kind of identification of Weld Defects based on improvement LeNet-5 models
CN108710893B (en) Digital image camera source model classification method based on feature fusion
CN109543760A (en) Confrontation sample testing method based on image filters algorithm
CN110084812B (en) Terahertz image defect detection method, device and system and storage medium
CN111105389B (en) Road surface crack detection method integrating Gabor filter and convolutional neural network
CN111222545B (en) Image classification method based on linear programming incremental learning
CN105718932A (en) Colorful image classification method based on fruit fly optimization algorithm and smooth twinborn support vector machine and system thereof
CN111985554A (en) Model training method, bracelet identification method and corresponding device
CN111597932A (en) Road crack image identification method, device and system based on convolutional neural network
CN116699096B (en) Water quality detection method and system based on deep learning
CN114331946A (en) Image data processing method, device and medium
CN112507888A (en) Building identification method and device
CN115471463A (en) Solar cell defect detection method based on countermeasure type generation network
CN111652307A (en) Intelligent nondestructive identification method and device for redwood furniture based on convolutional neural network
CN111598854A (en) Complex texture small defect segmentation method based on rich robust convolution characteristic model
CN108537762B (en) Depth multi-scale network-based secondary JPEG compressed image evidence obtaining method

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