CN106682694A - Sensitive image identification method based on depth learning - Google Patents

Sensitive image identification method based on depth learning Download PDF

Info

Publication number
CN106682694A
CN106682694A CN201611223417.XA CN201611223417A CN106682694A CN 106682694 A CN106682694 A CN 106682694A CN 201611223417 A CN201611223417 A CN 201611223417A CN 106682694 A CN106682694 A CN 106682694A
Authority
CN
China
Prior art keywords
training
network
test
sensitive
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
CN201611223417.XA
Other languages
Chinese (zh)
Inventor
胡二雷
杜姗姗
冯瑞
金城
薛向阳
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.)
Fudan University
Original Assignee
Fudan 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 Fudan University filed Critical Fudan University
Priority to CN201611223417.XA priority Critical patent/CN106682694A/en
Publication of CN106682694A publication Critical patent/CN106682694A/en
Pending legal-status Critical Current

Links

Classifications

    • 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
    • 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
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/40—Extraction of image or video features
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V30/00—Character recognition; Recognising digital ink; Document-oriented image-based pattern recognition
    • G06V30/10—Character recognition
    • G06V30/19—Recognition using electronic means
    • G06V30/192—Recognition using electronic means using simultaneous comparisons or correlations of the image signals with a plurality of references
    • G06V30/194—References adjustable by an adaptive method, e.g. learning

Landscapes

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

Abstract

The invention belongs to the technical field of digital image processing, and particularly relates to a sensitive image identification method based on depth learning. The method mainly comprises following steps of preprocessing sensitive images; dividing all preprocessed sensitive image databases into a training set and a test set, wherein the training set is divided into a train part and a validation part with proportion of 5:1, and using the images of the training set in a depth convolution neural network training, wherein parameters between layers of the convolution neural network are obtained through the training; after the training is finished, using a trained model to initializing a test neural network, wherein the test neural network is the same as a trained grid in structure; and inputting test sensitive images into the initialized depth neural network for identification test, thereby achieving identification of the sensitive images. According to the invention, functions of extraction and classification of features can be finished without manual participation and adjustment, and the reliable sensitive image identification method with high performance is provided.

Description

A kind of nude picture detection method based on deep learning
Technical field
The invention belongs to digital image processing techniques field, and in particular to a kind of nude picture detection based on deep learning Method.
Background technology
As the Internet is particularly the rapid popularization of mobile Internet, the multi-medium data with image, video as representative is just Increased with surprising rapidity.How to detect that the sensitive image in internet content becomes the focus of research.Sensitive image Refer mainly to：Pornographic, violence, reaction, sensitive personage are (such as：Dalai) etc., traditional monitoring method relies primarily on manpower, with many matchmakers The explosive increase held in vivo, the mode for manually checking cannot meet and be actually needed.On the one hand it is the number such as video image According in explosive growth, manual type cannot undertake huge verification workload；On the other hand be exist it is a large amount of out-of-date Or the data being checked, cause a large amount of useless repeated workloads.
Deep learning belongs to machine learning, and deep learning is the focus in machine learning field in recent years, in speech recognition, figure As the fields such as identification achieve huge success.
The research that deep learning technology has very long developing history, early stage artificial neural network promotes deep learning Produce.General, deep learning refers to the Deep model of input layer, multiple hidden layers and output layer.Backpropagation techniques exist The eighties in 20th century is especially popular, is the algorithm of weight between famous learning network layers.Yann LeCun are adopted at first Handwritten numeral image is recognized with the back propagation convolutional network for having supervision of depth.In recent years, using depth convolution net Network identification handwritten numeral image has become the research topic of computer vision and machine learning field most worthy, at the two In field, deep learning technology all achieves state-of-the-art achievement.Convolutional neural networks (CNN) are carried by Krizhevsky et al. Go out, and won the champion of ImageNet image classifications using the technology in 2012.ImageNet is a large-scale number According to collection, possess 15,000,000 markd images, belong to 22000 classifications.Image inside ImageNet is received from the Internet Collection, is made marks using the instrument Mechanical Turk crowd-source of Amazon through artificial mode,
From the beginning of 2010, ImageNet data sets as the extensive visual identity challenges (ILSVRC) of ImageNet one Part, holds every year once, and using 1000 classifications in ImageNet, each classification includes 1000 images to ILSVRC, 120 Used as training set, 50000 images have 8 to image as checking collection, 150,000 images as test set, the model for using at that time Layer, first 5 layers is convolutional layer, and afterwards three layers is full articulamentum, and deep learning framework caffe is exactly according to this model realization Open Framework, the deep learning Open Framework that we use in the present invention is exactly caffe.
The content of the invention
The purpose of the present invention is exactly to overcome the shortcomings of that artificial cognition sensitive image is wasted time and energy, there is provided one kind is based on deep The method of degree study, improves the efficiency of nude picture detection.
The nude picture detection method based on deep learning that the present invention is provided, concretely comprises the following steps：
Step 1, from collect sensitive image data base in extract sensitive image information；
Step 2, sensitive image information is trained before pretreatment；
Step 3, training set and test set will be divided into through pretreated whole sensitive images；
Step 4, the training that training set is used for deep neural network (DNN)；
The neutral net of step 5, the training result initialization test of deep neural network (DNN)；
Step 6, the identification test that the neutral net that test set is input to after initialization is carried out sensitive image.
Wherein：
In step 2, the pretreatment of sensitive image, including：
Each sub-picture in sensitive image data base is sampled, and extracts the part containing sensitive information, while will Image is converted into jpg forms, is scaled 256*256 sizes, generates the text of image path/label form, then out of order arrangement Information in text.
In step 4, the training of depth convolutional neural networks (DNN), including：
The training of depth convolutional neural networks has two kinds of forms, pre-training (pre-training) and tuning (fine- Tuning), the time loss of pre-training is longer, and it cannot be guaranteed that training model can restrain, so the present invention adopts tuning Formal discipline model, the model initialization neutral net that fine-tuning is trained using identical network, then according to training Parameter between sample fine setting neutral net, increases the learning capacity to training sample, and the time not only trained is shorter, Er Qieke To ensure convergence.
Step 5 kind, tests the initialization of neutral net, including：
The neutral net of test is identical with the neural network structure of training, last layer network softmax outputs It is probability that test image belongs to each classification.Initialization test network, exactly by the deep neural network after training Each layer of hidden layer network weight weight values, be directly transmitted to test network equivalent layer so that test network and training model With identical hidden layers numbers, the number of hidden nodes and each layer of network weight weight values.
In step 6, the output of nude picture detection result, including：
For the sample in sensitive image test set, carried out using deep neural network (DNN) model after initialization The identification test of sensitive image, exports the result of nude picture detection.The characteristics of here with softmax layers, softmax layers The n-dimensional vector of output is the probability that test image belongs to each classification, so, correspond in the maximum subscript of n-dimensional vector intermediate value Corresponding that classification of label in training sample.
The tuning (fine-tunning) of depth convolutional neural networks (DNN) of the present invention, including：Using from top The method of downward supervised learning is learnt and tuning, i.e., be trained using the sample data that has label, error from push up to Lower transmission, to network tuning is carried out.After tuning so that each layer of hidden layer of depth convolutional neural networks (DNN) model Network weight weight values can be optimal value.Concrete tuning (fine-tuning) process is as follows：
According to input data and the error function of output result, the weighted value of network is updated using back-propagation algorithm, made Obtain network and reach global optimum.Cost function between input data and output result is：
The purpose of weight attenuation parameter λ is two impact proportions to total cost in control formula.Wherein：J(W,b；x,y) It is single sample (x(i),y(i)) calculated cost function；J (W, b) be whole samples cost function, it contains weight and declines Deduction item.
Gradient descent method mainly updates weight parameter W and offset parameter b, and the formula of calculating is as follows：
First, need to be calculated using back-propagation algorithm before calculatingWith This two, this two be single sample input (x, y) cost function J (W, b；X, y) partial derivative.The partial derivative is obtained, so Overall cost function J (W, partial derivative b) are calculated afterwards：
In the formula, above formula is one more than following formula, because weight decay will act on weight parameter W rather than bias b.
The main thought of back-propagation algorithm is as follows：
Input sample (x, y), calculates according to forward methods, obtains all of activation value, i.e. h in networkW,b(x)It is defeated Go out value.Afterwards for each node i of l layers, calculate its " residual error "Difference i.e. between output valve and desired value. Last output node is exactly the corresponding label of training sample due to desired value, and the residual error of calculating is(value is known ).For hidden unit, will be calculated based on the weighted mean of node (l+1 node layers) residual errorThese nodes with As input.The details of reverse conduction algorithm is presented below：
1st, forward conduction calculating is carried out, using forward conduction formula, is calculated L2, L3... until output layer LnlSwash Value living.
2nd, for each output unit i of the n-th l layers (output layer), residual error is calculated according to below equation：
3rd, to l=nl-1, nl-2, nl-3 ..., 2 each layer, the residual computations method of i-th node of l layers is such as Under：
4th, the partial derivative for needing is calculated, computational methods are as follows：
5th, according to above method, iteration is reducing cost function J (W, value b), and then solve neutral net Parameter.
Compared with prior art, advantages of the present invention and effect have：
1st, for the extraction of sensitive features, directly using the original pixels characteristic information of sensitive image, method is simple, is not required to Want artificial design feature.
2nd, using depth convolutional neural networks, it is to avoid the artificial design feature of traditional pattern recognition, deep learning is learned automatically Practise effective feature.
3rd, by regulation parameter and Optimized model, the feature learning ability of depth convolutional neural networks is given full play to, there is provided A kind of method of the high performance nude picture detection based on deep learning
4th, the present invention increased incremental training on the basis of tuning (fine-tuning), continue in the model for training Upper increase sample continues to train, and strengthens the learning capacity to sensitive image.
Description of the drawings
Fig. 1 is the nude picture detection training pattern and the block diagram of nude picture detection based on deep learning.
Fig. 2 is neural network diagram.
Specific embodiment
Technical scheme of the present invention is further illustrated below.
1st, sensitive image information is extracted from the sensitive image data base for collecting, is mainly cut out containing in sensitive image The part of sensitive information, reduces the impact of noise components.
2nd, the pretreatment operation before being trained to sensitive image information, is mainly converted into jpg format-patterns by image, It is 256*256 sizes by image scaling, ultimately produces the text of image path/label (0,1,2...m) form.
3rd, training set and test set will be divided into through pretreated whole sensitive images, i.e., according to training set：Test set About 10:All images are divided into training set and test set by 1 ratio
The 4th, training set is used for the training of deep neural network (DNN), is provided using deep learning Open Framework caffe Teaching interface, starts training pattern.
5th, sensitive image test, specifically, the input image to be recognized, using model initialization neutral net, through god The computing of Jing networks, finally exports the result of identification.
List of references：
[1]Krizhevsky A,Sutskever I,Hinton G E.ImageNet Classification with Deep Convolutional Neural Networks[J].Advances in Neural Information Processing Systems,2012,25(2):2012.
[2]Chen Y,Lin Z,Zhao X,et al.Deep Learning-Based Classification of Hyperspectral Data[J].IEEE Journal of Selected Topics in Applied Earth Observations&Remote Sensing,2014,7(6):2094-2107.
[3]Levy E,David O E,Netanyahu N S.Genetic algorithms and deep learning for automatic painter classification[C]//Conference on Genetic& Evolutionary Computation.ACM,2014:1143-1150.
[4]An X,Kuang D,Guo X,et al.A Deep Learning Method for Classification of EEG Data Based on Motor Imagery[M]//Intelligent Computing in Bioinformatics.2014:203-210.
[5] Kang Xiaodong, Wang Hao, Guo Jun, etc. unsupervised deep learning color image recognition method [J]. computer utility, 2015,35(9):2636-2639.
[6]Marmanis D,Datcu M,Esch T,et al.Deep Learning Earth Observation Classification Using ImageNet Pretrained Networks[J].IEEE Geoscience&Remote Sensing Letters,2015,13(1):1-5.
[7]Ji Wan,Dayong Wang,Steven C.H.Hoi,Pengcheng Wu,Jianke Zhu,Yongdong Zhang,Jintao Li.Deep Learning for Content-Based Image Retrieval:A Comprehensive Study.2014。

Claims (2)

1. the nude picture detection method of deep learning is based on, it is characterised in that concretely comprised the following steps：
Step 1, from collect sensitive image data base in extract sensitive image information；
Step 2, sensitive image information is trained before pretreatment；
Step 3, training set and test set will be divided into through pretreated whole sensitive images；
Step 4, the training that training set is used for deep neural network；
The neutral net of step 5, the training result initialization test of deep neural network；
Step 6, the identification test that the neutral net that test set is input to after initialization is carried out sensitive image；
Wherein：
In step 2, the pretreatment of sensitive image, including：
Each sub-picture in sensitive image data base is sampled, and extracts the part containing sensitive information, while by image Jpg forms are converted into, 256*256 sizes are scaled, the text of image path/label form is generated, then out of order arrangement text In information；
In step 4, the training of depth convolutional neural networks, using the formal discipline model of tuning, tuning is instructed using identical network The model initialization neutral net perfected, then finely tunes the parameter between neutral net according to training sample, increases to training sample This learning capacity；
In step 5, the initialization of neutral net is tested, the neutral net of test is identical with the neural network structure of training, What last layer network softmax was exported is the probability that test image belongs to each classification；Initialization test network, being exactly will instruction The network weight weight values of each layer of hidden layer in the deep neural network after perfecting, are directly transmitted to test network equivalent layer, so as to So that test network and the model of training have identical hidden layers numbers, the number of hidden nodes and each layer of network weight weight values；
In step 6, for the sample in sensitive image test set, carried out using the deep neural network model after initialization quick The identification test of sense image, exports the result of nude picture detection, the characteristics of here with softmax layers, the output of softmax layers N-dimensional vector be probability that test image belongs to each classification, so, correspond to training in the maximum subscript of n-dimensional vector intermediate value Corresponding that classification of label in sample.
2. the nude picture detection method based on deep learning according to claim 1, it is characterised in that be depth volume The tuning of product neutral net, is learnt and tuning using the method for top-down supervised learning, i.e., using the sample for having label Notebook data is trained, the top-down transmission of error, and to network tuning is carried out；After tuning so that depth convolutional Neural The network weight weight values of each layer of hidden layer of network model can be optimal value；Concrete evolutionary process is as follows：
According to input data and the error function of output result, the weighted value of network is updated using back-propagation algorithm so that net Network reaches global optimum；Cost function between input data and output result is：
（1）
Wherein,Weight attenuation parameter, for controlling formula in two impact proportions to total cost； It is single sampleCalculated cost function；It is the cost function of whole samples, it contains weight and declines Deduction item；
Using gradient descent method, for updating weight parameterAnd offset parameter, the formula of calculating is as follows：
（2）
（3）
Need to be calculated using back-propagation algorithm before calculatingWithThis two, this two Item is single sample inputCost functionPartial derivative；Then overall cost function is calculatedPartial derivative：
（4）
Described back-propagation algorithm is：Input sample, it is calculated according to forward methods all of sharp in network Value living, i.e.,Output valve；Afterwards for each node i of l layers, calculate its " residual error ", i.e., output valve and Difference between desired value；Last output node is exactly the corresponding label of training sample due to desired value, and the residual error of calculating is；For hidden unit, the weighted mean of residual error for based on node being l+1 node layers is calculated, these nodes withAs input；
The idiographic flow of reverse conduction algorithm is：
（1）Forward conduction calculating is carried out, using forward conduction formula, is calculatedUntil output layerActivation Value；
（2）ForLayer is each output unit i of output layer, and according to below equation residual error is calculated：
；
（3）It is rightEach layer, the residual computations of i-th node of l layers are as follows：
；
（4）Calculate the partial derivative for needing：
；
（5）According to above method, iteration is reducing cost functionValue, and then solve neutral net ginseng Number.
CN201611223417.XA 2016-12-27 2016-12-27 Sensitive image identification method based on depth learning Pending CN106682694A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201611223417.XA CN106682694A (en) 2016-12-27 2016-12-27 Sensitive image identification method based on depth learning

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201611223417.XA CN106682694A (en) 2016-12-27 2016-12-27 Sensitive image identification method based on depth learning

Publications (1)

Publication Number Publication Date
CN106682694A true CN106682694A (en) 2017-05-17

Family

ID=58871668

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201611223417.XA Pending CN106682694A (en) 2016-12-27 2016-12-27 Sensitive image identification method based on depth learning

Country Status (1)

Country Link
CN (1) CN106682694A (en)

Cited By (31)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107330519A (en) * 2017-06-26 2017-11-07 西北工业大学 Fault Locating Method based on deep neural network
CN107369086A (en) * 2017-07-06 2017-11-21 上海你我贷互联网金融信息服务有限公司 A kind of identity card stamp system and method
CN107437096A (en) * 2017-07-28 2017-12-05 北京大学 Image classification method based on the efficient depth residual error network model of parameter
CN107547555A (en) * 2017-09-11 2018-01-05 北京匠数科技有限公司 A kind of web portal security monitoring method and device
CN107704877A (en) * 2017-10-09 2018-02-16 哈尔滨工业大学深圳研究生院 A kind of image privacy cognitive method based on deep learning
CN108038543A (en) * 2017-10-24 2018-05-15 华南师范大学 It is expected and anti-desired depth learning method and nerve network system
CN108154442A (en) * 2017-12-25 2018-06-12 杭州七炅信息科技有限公司 The anti-fraud detection algorithm of property insurance
CN108334902A (en) * 2018-02-02 2018-07-27 北京华纵科技有限公司 A kind of track train equipment room smog fireproof monitoring method based on deep learning
CN108415807A (en) * 2018-02-12 2018-08-17 成都睿码科技有限责任公司 A method of crawling whether monitoring electronic equipment browses flame
CN108427972A (en) * 2018-04-24 2018-08-21 云南佳叶现代农业发展有限公司 Tobacco classification method based on on-line study and its system
CN108648188A (en) * 2018-05-15 2018-10-12 南京邮电大学 A kind of non-reference picture quality appraisement method based on generation confrontation network
CN108764293A (en) * 2018-04-28 2018-11-06 重庆交通大学 A kind of vehicle checking method and system based on image
CN108847223A (en) * 2018-06-20 2018-11-20 陕西科技大学 A kind of audio recognition method based on depth residual error neural network
CN108959794A (en) * 2018-07-13 2018-12-07 北京航空航天大学 A kind of structural frequency response modification methodology of dynamics model based on deep learning
CN109034217A (en) * 2018-07-10 2018-12-18 成都先进金属材料产业技术研究院有限公司 Grain size intelligence ranking method based on image recognition depth learning technology
CN109035221A (en) * 2018-07-10 2018-12-18 成都先进金属材料产业技术研究院有限公司 Macrostructure intelligence ranking method based on image recognition depth learning technology
CN109360196A (en) * 2018-09-30 2019-02-19 北京羽医甘蓝信息技术有限公司 Method and device based on deep learning processing oral cavity radiation image
CN109359551A (en) * 2018-09-21 2019-02-19 深圳市璇玑实验室有限公司 A kind of nude picture detection method and system based on machine learning
CN109447140A (en) * 2018-10-19 2019-03-08 广州四十五度科技有限公司 A method of the image recognition based on neural network deep learning simultaneously recommends cognition
CN109583344A (en) * 2018-11-21 2019-04-05 红相股份有限公司 A kind of power equipment Infrared Image Processing Method based on deep learning
CN109635945A (en) * 2018-11-21 2019-04-16 华中科技大学 A kind of training method of the deep neural network for image classification
CN109657596A (en) * 2018-12-12 2019-04-19 天津卡达克数据有限公司 A kind of vehicle appearance component identification method based on deep learning
CN109711377A (en) * 2018-12-30 2019-05-03 陕西师范大学 Standardize examinee's positioning and method of counting in the single-frame images of examination hall monitoring
CN110070119A (en) * 2019-04-11 2019-07-30 北京工业大学 A kind of handwritten numeral image recognition classification method based on binaryzation deep neural network
CN110110780A (en) * 2019-04-30 2019-08-09 南开大学 A kind of picture classification method based on confrontation neural network and magnanimity noise data
WO2020006898A1 (en) * 2018-07-05 2020-01-09 平安科技（深圳）有限公司 Method and device for recognizing audio data of instrument, electronic apparatus, and storage medium
CN110674678A (en) * 2019-08-07 2020-01-10 国家计算机网络与信息安全管理中心 Method and device for identifying sensitive mark in video
CN110892417A (en) * 2017-06-05 2020-03-17 D5Ai有限责任公司 Asynchronous agent with learning coach and structurally modifying deep neural network without degrading performance
CN111639359A (en) * 2020-04-22 2020-09-08 中国科学院计算技术研究所 Method and system for detecting and early warning privacy risks of social network pictures
CN113033796A (en) * 2020-12-11 2021-06-25 北京信息科技大学 Image identification method of all-optical nonlinear diffraction neural network
CN114998659A (en) * 2022-06-17 2022-09-02 北京大学 Image data classification method for training impulse neural network model on line along with time

Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101055621A (en) * 2006-04-10 2007-10-17 中国科学院自动化研究所 Content based sensitive web page identification method
CN101281521A (en) * 2007-04-05 2008-10-08 中国科学院自动化研究所 Method and system for filtering sensitive web page based on multiple classifier amalgamation
CN101661559A (en) * 2009-09-16 2010-03-03 中国科学院计算技术研究所 Digital image training and detecting methods
CN101853377A (en) * 2010-05-13 2010-10-06 复旦大学 Method for identifying content of digital video
CN101877064A (en) * 2009-04-30 2010-11-03 索尼株式会社 Image classification method and image classification device
CN102201064A (en) * 2010-03-25 2011-09-28 北京中星微电子有限公司 Method and device for filtering sensitive video
CN102306287A (en) * 2011-08-24 2012-01-04 百度在线网络技术（北京）有限公司 Method and equipment for identifying sensitive image
CN102938054A (en) * 2012-09-06 2013-02-20 北京工业大学 Method for recognizing compressed-domain sensitive images based on visual attention models
CN103093236A (en) * 2013-01-15 2013-05-08 北京工业大学 Movable terminal porn filtering method based on analyzing image and semantics
CN103294813A (en) * 2013-06-07 2013-09-11 北京捷成世纪科技股份有限公司 Sensitive image search method and device
CN103605992A (en) * 2013-11-28 2014-02-26 国家电网公司 Sensitive image recognizing method in interaction of inner and outer power networks
CN103793718A (en) * 2013-12-11 2014-05-14 台州学院 Deep study-based facial expression recognition method
CN104850845A (en) * 2015-05-30 2015-08-19 大连理工大学 Traffic sign recognition method based on asymmetric convolution neural network
CN105787510A (en) * 2016-02-26 2016-07-20 华东理工大学 System and method for realizing subway scene classification based on deep learning
CN106096616A (en) * 2016-06-08 2016-11-09 四川大学华西医院 A kind of nuclear magnetic resonance image feature extraction based on degree of depth study and sorting technique

Patent Citations (15)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101055621A (en) * 2006-04-10 2007-10-17 中国科学院自动化研究所 Content based sensitive web page identification method
CN101281521A (en) * 2007-04-05 2008-10-08 中国科学院自动化研究所 Method and system for filtering sensitive web page based on multiple classifier amalgamation
CN101877064A (en) * 2009-04-30 2010-11-03 索尼株式会社 Image classification method and image classification device
CN101661559A (en) * 2009-09-16 2010-03-03 中国科学院计算技术研究所 Digital image training and detecting methods
CN102201064A (en) * 2010-03-25 2011-09-28 北京中星微电子有限公司 Method and device for filtering sensitive video
CN101853377A (en) * 2010-05-13 2010-10-06 复旦大学 Method for identifying content of digital video
CN102306287A (en) * 2011-08-24 2012-01-04 百度在线网络技术（北京）有限公司 Method and equipment for identifying sensitive image
CN102938054A (en) * 2012-09-06 2013-02-20 北京工业大学 Method for recognizing compressed-domain sensitive images based on visual attention models
CN103093236A (en) * 2013-01-15 2013-05-08 北京工业大学 Movable terminal porn filtering method based on analyzing image and semantics
CN103294813A (en) * 2013-06-07 2013-09-11 北京捷成世纪科技股份有限公司 Sensitive image search method and device
CN103605992A (en) * 2013-11-28 2014-02-26 国家电网公司 Sensitive image recognizing method in interaction of inner and outer power networks
CN103793718A (en) * 2013-12-11 2014-05-14 台州学院 Deep study-based facial expression recognition method
CN104850845A (en) * 2015-05-30 2015-08-19 大连理工大学 Traffic sign recognition method based on asymmetric convolution neural network
CN105787510A (en) * 2016-02-26 2016-07-20 华东理工大学 System and method for realizing subway scene classification based on deep learning
CN106096616A (en) * 2016-06-08 2016-11-09 四川大学华西医院 A kind of nuclear magnetic resonance image feature extraction based on degree of depth study and sorting technique

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
吴正文: "《卷积神经网络在图像分类中的应用研究》", 《中国优秀硕士学位论文全文数据库 信息科技辑》 *

Cited By (46)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110892417A (en) * 2017-06-05 2020-03-17 D5Ai有限责任公司 Asynchronous agent with learning coach and structurally modifying deep neural network without degrading performance
CN110892417B (en) * 2017-06-05 2024-02-20 D5Ai有限责任公司 Asynchronous agent with learning coaches and structurally modifying deep neural networks without degrading performance
CN107330519A (en) * 2017-06-26 2017-11-07 西北工业大学 Fault Locating Method based on deep neural network
CN107369086A (en) * 2017-07-06 2017-11-21 上海你我贷互联网金融信息服务有限公司 A kind of identity card stamp system and method
CN107437096A (en) * 2017-07-28 2017-12-05 北京大学 Image classification method based on the efficient depth residual error network model of parameter
CN107437096B (en) * 2017-07-28 2020-06-26 北京大学 Image classification method based on parameter efficient depth residual error network model
CN107547555A (en) * 2017-09-11 2018-01-05 北京匠数科技有限公司 A kind of web portal security monitoring method and device
CN107704877A (en) * 2017-10-09 2018-02-16 哈尔滨工业大学深圳研究生院 A kind of image privacy cognitive method based on deep learning
WO2019071754A1 (en) * 2017-10-09 2019-04-18 哈尔滨工业大学深圳研究生院 Method for sensing image privacy on the basis of deep learning
CN108038543B (en) * 2017-10-24 2021-01-22 华南师范大学 Expectation and anti-expectation deep learning method and neural network system
CN108038543A (en) * 2017-10-24 2018-05-15 华南师范大学 It is expected and anti-desired depth learning method and nerve network system
CN108154442A (en) * 2017-12-25 2018-06-12 杭州七炅信息科技有限公司 The anti-fraud detection algorithm of property insurance
CN108334902A (en) * 2018-02-02 2018-07-27 北京华纵科技有限公司 A kind of track train equipment room smog fireproof monitoring method based on deep learning
CN108415807A (en) * 2018-02-12 2018-08-17 成都睿码科技有限责任公司 A method of crawling whether monitoring electronic equipment browses flame
CN108427972A (en) * 2018-04-24 2018-08-21 云南佳叶现代农业发展有限公司 Tobacco classification method based on on-line study and its system
CN108427972B (en) * 2018-04-24 2024-06-07 云南佳叶现代农业发展有限公司 Tobacco leaf classification method and system based on online learning
CN108764293A (en) * 2018-04-28 2018-11-06 重庆交通大学 A kind of vehicle checking method and system based on image
CN108648188A (en) * 2018-05-15 2018-10-12 南京邮电大学 A kind of non-reference picture quality appraisement method based on generation confrontation network
CN108648188B (en) * 2018-05-15 2022-02-11 南京邮电大学 No-reference image quality evaluation method based on generation countermeasure network
CN108847223B (en) * 2018-06-20 2020-09-29 陕西科技大学 Voice recognition method based on deep residual error neural network
CN108847223A (en) * 2018-06-20 2018-11-20 陕西科技大学 A kind of audio recognition method based on depth residual error neural network
WO2020006898A1 (en) * 2018-07-05 2020-01-09 平安科技（深圳）有限公司 Method and device for recognizing audio data of instrument, electronic apparatus, and storage medium
CN109034217A (en) * 2018-07-10 2018-12-18 成都先进金属材料产业技术研究院有限公司 Grain size intelligence ranking method based on image recognition depth learning technology
CN109035221A (en) * 2018-07-10 2018-12-18 成都先进金属材料产业技术研究院有限公司 Macrostructure intelligence ranking method based on image recognition depth learning technology
CN108959794B (en) * 2018-07-13 2023-04-07 北京航空航天大学 Structural frequency response dynamic model correction method based on deep learning
CN108959794A (en) * 2018-07-13 2018-12-07 北京航空航天大学 A kind of structural frequency response modification methodology of dynamics model based on deep learning
CN109359551A (en) * 2018-09-21 2019-02-19 深圳市璇玑实验室有限公司 A kind of nude picture detection method and system based on machine learning
CN109360196A (en) * 2018-09-30 2019-02-19 北京羽医甘蓝信息技术有限公司 Method and device based on deep learning processing oral cavity radiation image
CN109447140A (en) * 2018-10-19 2019-03-08 广州四十五度科技有限公司 A method of the image recognition based on neural network deep learning simultaneously recommends cognition
CN109447140B (en) * 2018-10-19 2021-10-12 广州四十五度科技有限公司 Image identification and cognition recommendation method based on neural network deep learning
CN109635945A (en) * 2018-11-21 2019-04-16 华中科技大学 A kind of training method of the deep neural network for image classification
CN109583344A (en) * 2018-11-21 2019-04-05 红相股份有限公司 A kind of power equipment Infrared Image Processing Method based on deep learning
CN109635945B (en) * 2018-11-21 2022-12-02 华中科技大学 Deep neural network training method for image classification
CN109583344B (en) * 2018-11-21 2019-11-19 红相股份有限公司 A kind of power equipment Infrared Image Processing Method based on deep learning
CN109657596A (en) * 2018-12-12 2019-04-19 天津卡达克数据有限公司 A kind of vehicle appearance component identification method based on deep learning
CN109711377B (en) * 2018-12-30 2023-04-18 陕西师范大学 Method for positioning and counting examinees in single-frame image monitored by standardized examination room
CN109711377A (en) * 2018-12-30 2019-05-03 陕西师范大学 Standardize examinee's positioning and method of counting in the single-frame images of examination hall monitoring
CN110070119A (en) * 2019-04-11 2019-07-30 北京工业大学 A kind of handwritten numeral image recognition classification method based on binaryzation deep neural network
CN110110780B (en) * 2019-04-30 2023-04-07 南开大学 Image classification method based on antagonistic neural network and massive noise data
CN110110780A (en) * 2019-04-30 2019-08-09 南开大学 A kind of picture classification method based on confrontation neural network and magnanimity noise data
CN110674678A (en) * 2019-08-07 2020-01-10 国家计算机网络与信息安全管理中心 Method and device for identifying sensitive mark in video
CN111639359A (en) * 2020-04-22 2020-09-08 中国科学院计算技术研究所 Method and system for detecting and early warning privacy risks of social network pictures
CN111639359B (en) * 2020-04-22 2023-09-12 中国科学院计算技术研究所 Method and system for detecting and early warning privacy risk of social network picture
CN113033796A (en) * 2020-12-11 2021-06-25 北京信息科技大学 Image identification method of all-optical nonlinear diffraction neural network
CN114998659A (en) * 2022-06-17 2022-09-02 北京大学 Image data classification method for training impulse neural network model on line along with time
CN114998659B (en) * 2022-06-17 2024-05-24 北京大学 Image data classification method for training impulse neural network model on line along with time

Similar Documents

Publication Publication Date Title
CN106682694A (en) Sensitive image identification method based on depth learning
CN112308158B (en) Multi-source field self-adaptive model and method based on partial feature alignment
CN111126386B (en) Sequence domain adaptation method based on countermeasure learning in scene text recognition
CN109886141B (en) Pedestrian re-identification method based on uncertainty optimization
CN105205448B (en) Text region model training method and recognition methods based on deep learning
CN112883738A (en) Medical entity relation extraction method based on neural network and self-attention mechanism
Dong et al. Research and discussion on image recognition and classification algorithm based on deep learning
CN107330355B (en) Deep pedestrian re-identification method based on positive sample balance constraint
CN107886123B (en) synthetic aperture radar target identification method based on auxiliary judgment update learning
CN113033309B (en) Fault diagnosis method based on signal downsampling and one-dimensional convolutional neural network
CN108846413B (en) Zero sample learning method based on global semantic consensus network
CN107944410B (en) Cross-domain facial feature analysis method based on convolutional neural network
CN113128620B (en) Semi-supervised domain self-adaptive picture classification method based on hierarchical relationship
CN111400494B (en) Emotion analysis method based on GCN-Attention
CN108171318A (en) One kind is based on the convolutional neural networks integrated approach of simulated annealing-Gaussian function
CN107528824B (en) Deep belief network intrusion detection method based on two-dimensional sparsification
CN113011487B (en) Open set image classification method based on joint learning and knowledge migration
CN113723238B (en) Face lightweight network model construction method and face recognition method
CN115601583A (en) Deep convolution network target identification method of double-channel attention mechanism
CN108596044B (en) Pedestrian detection method based on deep convolutional neural network
CN116152554A (en) Knowledge-guided small sample image recognition system
Dongmei et al. Classification and identification of citrus pests based on InceptionV3 convolutional neural network and migration learning
CN115062727A (en) Graph node classification method and system based on multi-order hypergraph convolutional network
CN115408603A (en) Online question-answer community expert recommendation method based on multi-head self-attention mechanism
CN112200262B (en) Small sample classification training method and device supporting multitasking and cross-tasking

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
WD01 Invention patent application deemed withdrawn after publication
WD01 Invention patent application deemed withdrawn after publication

Application publication date: 20170517