CN109816002A - The single sparse self-encoding encoder detection method of small target migrated certainly based on feature - Google Patents
The single sparse self-encoding encoder detection method of small target migrated certainly based on feature Download PDFInfo
- Publication number
- CN109816002A CN109816002A CN201910028640.6A CN201910028640A CN109816002A CN 109816002 A CN109816002 A CN 109816002A CN 201910028640 A CN201910028640 A CN 201910028640A CN 109816002 A CN109816002 A CN 109816002A
- Authority
- CN
- China
- Prior art keywords
- training
- sample
- sae
- feature
- trained
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Landscapes
- Image Analysis (AREA)
Abstract
The invention discloses a kind of based on feature from the single sparse coding device detection method of small target of migration, and this method is as follows: training sample set, test sample collection, raw data set of the building for the Weak target of training;Training sample set is inputted into training in SAE model, obtains the sparse features of sample, i.e. model parameterWith sparse features training softmax, i.e. input feature vector f (Wm+1x+bm+1) each training is trained to softmax after the completion of, retain positive sample, randomly select negative sample similar in the quantity of quantity and positive sample;Using model parameter as next time trained original model parameter, realizes that the parameter of SAE model updates, repeat above step, when the value of the loss function of SAE model training is identical as the value of a preceding loss function, training terminates;Test sample collection is inputted the softmax that last time training obtains to test, obtains test result.The present invention can accurately in detection image Weak target.
Description
Technical field
The present invention relates to computer vision processing technology fields, more particularly to a kind of list migrated certainly based on feature
One sparse self-encoding encoder detection method of small target.
Background technique
Dim target detection is a difficult point of field of image processing, the faint mesh in natural image especially medical image
Target detection difficulty is very big, and general blur margin is clear in the picture for Weak target, and contrast is low, and exists in most cases
Noise jamming greatly increases detection difficulty.Currently, traditional method and deep learning are for such dim target detection
All have some limitations.Detection for weak target, feature extraction are a highly important job, effective feature
Extract the accuracy that can greatly improve late detection.
Summary of the invention
The present invention detects to solve the problems, such as that the prior art can not carry out high-precision to weak target, provides a kind of base
In feature from the single sparse self-encoding encoder detection method of small target of migration, there is the small and weak mesh accurately detected in image
Mark.
To realize aforementioned present invention purpose, the technical solution adopted is as follows: a kind of single sparse from what is migrated based on feature
Encoder detection method of small target, it is as follows that the method comprising the steps of:
S1: it is the image data of a as training sample set that quantity is chosen from image data base, for constructing training sample
The positive sample and negative sample of concentration;It is the image data of 1-a as test sample collection that quantity is chosen from database, for constructing
The positive sample and negative sample that test sample is concentrated;The positive sample includes aneurysms, and 21* is constructed centered on aneurysms
The block of 21 pixels;The negative sample does not include the pixel of aneurysms, and size is the block of 21*21 pixel;Simultaneously from positive sample,
The green channel in color image, blue channel, the contrast enhancing knot corrected by Gamma are chosen in negative sample respectively
Fruit is as raw data set;
Wherein: a indicates that training sample set accounts for the percentage of image data base, and 0 < a < 1, a are manually set;
S2: being trained training sample set, and training sample set is input in SAE model, and training obtains training sample
The sparse features of collection, i.e. SAE model parameter
Wherein:Indicate the weight and biasing of the SAE model obtained by backpropagation;
S3: sparse features training softmax, i.e. input feature vector f (W are utilizedm+1x+bm+1) softmax is trained, often
After the completion of secondary training, retains positive sample, randomly select negative sample similar in the quantity of quantity and positive sample;
Wherein: f indicates sigmod activation primitiveM indicates the m times training;Wm+1、bm+1Respectively indicate
The weight and biasing of the SAE of m+1 training;
S4: by SAE model parameterAs next time trained original model parameter, the parameter of SAE model is realized
It updates, the feature for completing SAE model migrates certainly;Execute S2;Until obtaining the value of the loss function of SAE model training and preceding primary
When the value of loss function is identical, S5 is executed;
S5: after trained SAE model, by the softmax of test sample collection input to the end, test result is obtained.
Preferably, a value is 0.75, i.e., the image data conduct that quantity is 75% is chosen from image data base
Training sample set, it is 25% image data as test sample collection that quantity is chosen from database.
Preferably, the expression formula of step S2, the Softmax are as follows:
Wherein, ViIt is the output of classifier prime output unit;I indicates classification index, and total classification number is C;SiIt indicates
Be the corresponding feature vector of current training sample index and all sample index and ratio.
Preferably, the formula that the parameter of the SAE model in step S4 updates are as follows:
Wherein: WmIndicate the m times it is trained when SAE weight matrix;α is learning rate;s2Indicate the number of hiding layer unit;
ΔWmPartial derivative matrix for the m times loss function when trained about weight;λ is regularization penalty factor;bmIndicate the m times instruction
The bias matrix of SAE when practicing;For matrix Δ WmAn element in matrix;hW,b(x(i)) it is when to input be x(i)It is corresponding
Output;Layer is hidden for sparse coding device shadow to be averaged activity;Indicate the activation of j-th of neuron of hidden layer;ΔbmFor m
Partial derivative matrix when secondary trained about weight b.
Further, the loss function formula of the S4 are as follows:
Wherein, β is sparse item penalty factor;Referred to as KL divergence, for measuring two probability distribution
Degree of closeness;For the average activity of j-th of neuron of hidden layer;ρ is sparsity parameter;The J (W, b) passes through following formula
It is expressed:
Wherein: n is the number of sample;x(i)Indicate the input of i-th of neuron;It is l i-th of neuron of layer to next
The weight of j-th of neuron of layer.
Beneficial effects of the present invention are as follows: the present invention utilizes training sample by building training sample set, test sample collection
This collection SAE model, more new training sample set and SAE model repeatedly, make the value of the loss function of SAE model with it is previous
When being worth identical, terminate training;It goes to test trained model by test sample collection again, obtains test result, this method can
The accurately Weak target in detection image.
Detailed description of the invention
Fig. 1 is the flow chart of the method for the present invention.
Fig. 2 is the schematic diagram of training process of the present invention.
Fig. 3 is the structural schematic diagram of sparse coding device.
Fig. 4 is the present embodiment test result comparison diagram.
Specific embodiment
The present invention will be described in detail with reference to the accompanying drawings and detailed description.
Embodiment 1
As shown in Figure 1, a kind of single sparse coding device detection method of small target migrated certainly based on feature, this method tool
Steps are as follows for body:
Step S1: respectively from database Retinopathy Online Challenge, DIARETDB1 and E-ophtha
The image data of selection 75% is as training sample set, for constructing the positive sample and negative sample of training sample concentration;From data
25% other image data is chosen in library Retinopathy Online Challenge, DIARETDB1 and E-ophtha
As test sample collection, for constructing the positive sample and negative sample of test sample concentration;The positive sample be comprising aneurysms,
And the block of 21*21 pixel is constructed centered on aneurysms;The negative sample is not comprising aneurysms, and pixel is 21*21's
Block;It chooses the green channel in color image, blue channel respectively from positive sample, negative sample simultaneously, corrected by Gamma
The contrast enhancing result arrived is as raw data set;Shown color image is true color image, the color value of each of which pixel
All determined by tri- numerical value of R, G, B.
The present embodiment is also constructed simultaneously by green channel, blue channel, by the school Gamma when constructing training sample set
The raw data set of the contrast enhancing result composition just obtained.Current embodiment require that the database of selection needs to meet small and weak mesh
Target feature, and the sample standard deviation symbol in database Retinopathy Online Challenge, DIARETDB1 and E-ophtha
Close the feature of Weak target.
Step S2: being trained training sample set, as shown in Fig. 2, training sample set is input in SAE model, instructs
Get the sparse features of sample, i.e. model parameter
Wherein:Indicate the weight and biasing of the SAE model obtained by backpropagation;
Step S3: sparse features training softmax, i.e. input feature vector f (W are utilizedm+1x+bm+1) softmax is instructed
Practice;Every time after the completion of training, retains positive sample, randomly select negative sample similar in the quantity of quantity and positive sample;
Wherein: f indicates sigmod activation primitiveM indicates the m times training;Wm+1、bm+1Respectively indicate
The weight and biasing of the SAE of m+1 training;
One deep neural network model being made of the sparse self-encoding encoder of multilayer of SAE model described in the present embodiment,
Input of the output of preceding layer self-encoding encoder as its later layer self-encoding encoder, the last layer are (logistic points of a classifier
Class device or softmax classifier)
As shown in figure 3, the sparse self-encoding encoder is a kind of unsupervised machine learning algorithm, sparsity refers to when one
When the output of a neuron is 1, then it is assumed that this neuron is activation;When output is 0, then it is assumed that this mind
It is to inhibit through member.Neuron is set then to be referred to as sparsity limitation in holddown within the most of the time.The present embodiment is in reality
In the training process of border, it is desirable to which machine oneself can learn some important features into sample, by applying to hidden layer
Limitation and sparsity limitation enable the machine to learn under rugged environment the feature for preferably expressing sample, and can be effective
Dimensionality reduction is carried out to sample.But in the actual operation process, it can not correctly judge which neuron needs which is activated need to press down
System.Therefore the concept for needing to introduce average activity, is usedIt indicates, formula is as follows:
Wherein: s2Indicate the number of hidden layer neuron;Indicate that this is hidden when network is endowed specific input x
The activation of unit;Parameter ρ, referred to as sparsity parameter are introduced simultaneously in calculating process, and is made as far as possible
Softmax described in the present embodiment has very extensive application in machine learning, and Softmax calculates simple, effect
Fruit is significant, and especially in processing (C > 2) problems of classifying, the last output unit of classifier needs Softmax function to carry out numerical value more
Processing.It is defined as follows about Softmax function:
Wherein, ViIt is the output of classifier prime output unit;I indicates classification index, and total classification number is C;SiIt indicates
Be currentElement index and all elements index and ratio.
Polytypic output numerical value is converted relative probability by Softmax, and numerical value is made to be easier to understand and compare.
Step S4: by model parameterAs next time trained original model parameter, the parameter of SAE model is realized
It updates, the feature for completing SAE model migrates certainly;After feature migration, the parameters of SAE model are updated;Repetition step S2,
Step S3;Until obtain the value of loss function of SAE model training it is identical as previous value when, training terminate.
The present embodiment updates SAE model parameter using backpropagation, and it is leading biography that the backpropagation, which is with error,
Movement is broadcast, by asking local derviation progressive updating weight and biasing to weight and biasing in back-propagation process.Then parameter update can
To be obtained by following formula:
Wherein: WmIndicate the m times it is trained when SAE weight matrix;α is learning rate;s2Indicate the number of hiding layer unit;
ΔWmPartial derivative matrix for the m times loss function when trained about weight;λ is regularization penalty factor;bmIndicate the m times instruction
The bias matrix of SAE when practicing;For matrix Δ WmAn element in matrix;hW,b(x(i)) it is when to input be x(i)It is corresponding
Output;Layer is hidden for sparse coding device shadow to be averaged activity;Indicate the activation of j-th of neuron of hidden layer;ΔbmFor m
Partial derivative matrix when secondary trained about weight b.
The value of this implementation loss function is calculated by loss function, the loss function formula are as follows:
Wherein, β is sparse item penalty factor;Referred to as KL divergence, for measuring two probability distribution
Degree of closeness;For the average activity of j-th of neuron of hidden layer;ρ is sparsity parameter;The J (W, b) passes through following formula
It is expressed:
Wherein: n is the number of sample;x(i)Indicate the input of i-th of neuron;It is l i-th of neuron of layer to next
The weight of j-th of neuron of layer.
In the present embodiment every time training when training sample set all can include raw data set, raw data set it is main
Effect is in training process, in order to improve the accuracy of final mask.
Step S5: it after trained SAE model, is tested using the softmax that last time training obtains;It will test
Sample set is input in softmax, obtains test result.
The present embodiment test result mainly passes through two measurement standards, specific Sensitivity and accuracy rate
Accuracy.The present embodiment is obtained by the single sparse coding device detection method of small target migrated certainly based on feature
It arrives that test results are shown in figure 4, obtains the accuracy rate and specificity of each database.It may indicate that and be based on by test result
The detection method of small target of the feature of sparse coding device and softmax migration, can be extracted well by sparse coding device
The sparse features of sample improve the classification capacity of softmax by progressive training method step by step, significant to improve
The accuracy and specificity of target detection.
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.Any modification done within the spirit and principles of the present invention and changes equivalent replacement
Into etc., it should all be included in the scope of protection of the claims of the present invention.
Claims (5)
1. a kind of single sparse coding device detection method of small target migrated certainly based on feature, it is characterised in that: this method packet
Include that steps are as follows:
S1: it is the image data of a as training sample set that quantity is chosen from image data base, is concentrated for constructing training sample
Positive sample and negative sample;It is the image data of 1-a as test sample collection that quantity is chosen from database, for constructing test
Positive sample and negative sample in sample set;The positive sample includes aneurysms, and 21*21 picture is constructed centered on aneurysms
The block of element;The negative sample does not include the pixel of aneurysms, and size is the block of 21*21 pixel;Simultaneously from positive sample, negative sample
The green channel in color image, blue channel, the contrast enhancing result corrected by Gamma are chosen in this respectively to make
For raw data set;
Wherein: a indicates that training sample set accounts for the percentage of image data base, and 0 < a < 1, a are manually set;
S2: being trained training sample set, and training sample set is input in SAE model, and training obtains training sample set
Sparse features, i.e. SAE model parameter
Wherein:Indicate the weight and biasing of the SAE model obtained by backpropagation;
S3: sparse features training softmax, i.e. input feature vector f (W are utilizedm+1x+bm+1) softmax is trained, it instructs every time
After the completion of white silk, retains positive sample, randomly select negative sample similar in the quantity of quantity and positive sample;
Wherein: f indicates sigmod activation primitiveM indicates the m times training;Wm+1、bm+1It respectively indicates the m+1 times
The weight and biasing of trained SAE;
S4: by SAE model parameterAs next time trained original model parameter, realize that the parameter of SAE model updates,
The feature for completing SAE model migrates certainly;Execute S2;Until obtaining the value and preceding primary loss of the loss function of SAE model training
When the value of function is identical, S5 is executed;
S5: after trained SAE model, by the softmax of test sample collection input to the end, test result is obtained.
2. the single sparse coding device detection method of small target according to claim 1 migrated certainly based on feature, special
Sign is: a value is 0.75, i.e., it is 75% image data as training sample that quantity is chosen from image data base
Collection, it is 25% image data as test sample collection that quantity is chosen from database.
3. the single sparse coding device detection method of small target according to claim 1 migrated certainly based on feature, special
Sign is: step S2, the expression formula of the Softmax are as follows:
Wherein, ViIt is the output of classifier prime output unit;I indicates classification index, and total classification number is C;SiIndicate be
The index of the current corresponding feature vector of training sample and all sample index and ratio.
4. the single sparse coding device detection method of small target according to claim 1 migrated certainly based on feature, special
Sign is: the formula that the parameter of the SAE model in step S4 updates are as follows:
Wherein: WmIndicate the m times it is trained when SAE weight matrix;α is learning rate;s2Indicate the number of hiding layer unit;ΔWm
Partial derivative matrix for the m times loss function when trained about weight;λ is regularization penalty factor;bmIndicate the m times it is trained when
The bias matrix of SAE;For matrix Δ WmAn element in matrix;hW,b(x(i)) it is when to input be x(i)Corresponding output;Layer is hidden for sparse coding device shadow to be averaged activity;Indicate the activation of j-th of neuron of hidden layer;ΔbmFor the m times training
When partial derivative matrix about weight b.
5. the single sparse coding device detection method of small target according to claim 4 migrated certainly based on feature, special
Sign is: the loss function formula of the S4 are as follows:
Wherein, β is sparse item penalty factor;Referred to as KL divergence, for measure two probability distribution close to journey
Degree;For the average activity of j-th of neuron of hidden layer;ρ is sparsity parameter;The J (W, b) carries out table by following formula
It reaches:
Wherein: n is the number of sample;x(i)Indicate the input of i-th of neuron;It is i-th of neuron of l layer to next layer
The weight of j neuron.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910028640.6A CN109816002B (en) | 2019-01-11 | 2019-01-11 | Single sparse self-encoder weak and small target detection method based on feature self-migration |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910028640.6A CN109816002B (en) | 2019-01-11 | 2019-01-11 | Single sparse self-encoder weak and small target detection method based on feature self-migration |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109816002A true CN109816002A (en) | 2019-05-28 |
CN109816002B CN109816002B (en) | 2022-09-06 |
Family
ID=66603394
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910028640.6A Active CN109816002B (en) | 2019-01-11 | 2019-01-11 | Single sparse self-encoder weak and small target detection method based on feature self-migration |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109816002B (en) |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472667A (en) * | 2019-07-19 | 2019-11-19 | 广东工业大学 | Small object classification method based on deconvolution neural network |
CN110930409A (en) * | 2019-10-18 | 2020-03-27 | 电子科技大学 | Salt body semantic segmentation method based on deep learning and semantic segmentation model |
CN110972174A (en) * | 2019-12-02 | 2020-04-07 | 东南大学 | Wireless network interruption detection method based on sparse self-encoder |
CN111462817A (en) * | 2020-03-25 | 2020-07-28 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Classification model construction method and device, classification model and classification method |
CN112465042A (en) * | 2020-12-02 | 2021-03-09 | 中国联合网络通信集团有限公司 | Generation method and device of classification network model |
Citations (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070002275A1 (en) * | 2005-07-01 | 2007-01-04 | Siemens Corporate Research Inc. | Method and System For Local Adaptive Detection Of Microaneurysms In Digital Fundus Images |
CN104156736A (en) * | 2014-09-05 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method on basis of SAE and IDL |
CN104166859A (en) * | 2014-08-13 | 2014-11-26 | 西安电子科技大学 | Polarization SAR image classification based on SSAE and FSALS-SVM |
US20150379708A1 (en) * | 2010-12-07 | 2015-12-31 | University Of Iowa Research Foundation | Methods and systems for vessel bifurcation detection |
CN105224943A (en) * | 2015-09-08 | 2016-01-06 | 西安交通大学 | Based on the image swift nature method for expressing of multi thread normalization non-negative sparse coding device |
AU2014271202A1 (en) * | 2013-05-19 | 2016-01-07 | Commonwealth Scientific And Industrial Research Organisation | A system and method for remote medical diagnosis |
CN105320965A (en) * | 2015-10-23 | 2016-02-10 | 西北工业大学 | Hyperspectral image classification method based on spectral-spatial cooperation of deep convolutional neural network |
CN105654117A (en) * | 2015-12-25 | 2016-06-08 | 西北工业大学 | Hyperspectral image spectral-spatial cooperative classification method based on SAE depth network |
CN105787517A (en) * | 2016-03-11 | 2016-07-20 | 西安电子科技大学 | Polarized SAR image classification method base on wavelet sparse auto encoder |
US20160292856A1 (en) * | 2015-04-06 | 2016-10-06 | IDx, LLC | Systems and methods for feature detection in retinal images |
CN106096652A (en) * | 2016-06-12 | 2016-11-09 | 西安电子科技大学 | Based on sparse coding and the Classification of Polarimetric SAR Image method of small echo own coding device |
CN106651899A (en) * | 2016-12-09 | 2017-05-10 | 东北大学 | Fundus image micro-aneurysm detection system based on Adaboost |
CN106815601A (en) * | 2017-01-10 | 2017-06-09 | 西安电子科技大学 | Hyperspectral image classification method based on recurrent neural network |
CN107341511A (en) * | 2017-07-05 | 2017-11-10 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on super-pixel Yu sparse self-encoding encoder |
CN107590515A (en) * | 2017-09-14 | 2018-01-16 | 西安电子科技大学 | The hyperspectral image classification method of self-encoding encoder based on entropy rate super-pixel segmentation |
CN107798349A (en) * | 2017-11-03 | 2018-03-13 | 合肥工业大学 | A kind of transfer learning method based on the sparse self-editing ink recorder of depth |
CN108537233A (en) * | 2018-03-15 | 2018-09-14 | 南京师范大学 | A kind of pathology brain image sorting technique based on the sparse self-encoding encoder of depth stack |
CN108921233A (en) * | 2018-07-31 | 2018-11-30 | 武汉大学 | A kind of Raman spectrum data classification method based on autoencoder network |
CN109033952A (en) * | 2018-06-12 | 2018-12-18 | 杭州电子科技大学 | M-sequence recognition methods based on sparse self-encoding encoder |
CN109102019A (en) * | 2018-08-09 | 2018-12-28 | 成都信息工程大学 | Image classification method based on HP-Net convolutional neural networks |
CN109145832A (en) * | 2018-08-27 | 2019-01-04 | 大连理工大学 | Polarimetric SAR image semisupervised classification method based on DSFNN Yu non local decision |
-
2019
- 2019-01-11 CN CN201910028640.6A patent/CN109816002B/en active Active
Patent Citations (21)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20070002275A1 (en) * | 2005-07-01 | 2007-01-04 | Siemens Corporate Research Inc. | Method and System For Local Adaptive Detection Of Microaneurysms In Digital Fundus Images |
US20150379708A1 (en) * | 2010-12-07 | 2015-12-31 | University Of Iowa Research Foundation | Methods and systems for vessel bifurcation detection |
AU2014271202A1 (en) * | 2013-05-19 | 2016-01-07 | Commonwealth Scientific And Industrial Research Organisation | A system and method for remote medical diagnosis |
CN104166859A (en) * | 2014-08-13 | 2014-11-26 | 西安电子科技大学 | Polarization SAR image classification based on SSAE and FSALS-SVM |
CN104156736A (en) * | 2014-09-05 | 2014-11-19 | 西安电子科技大学 | Polarized SAR image classification method on basis of SAE and IDL |
US20160292856A1 (en) * | 2015-04-06 | 2016-10-06 | IDx, LLC | Systems and methods for feature detection in retinal images |
CN105224943A (en) * | 2015-09-08 | 2016-01-06 | 西安交通大学 | Based on the image swift nature method for expressing of multi thread normalization non-negative sparse coding device |
CN105320965A (en) * | 2015-10-23 | 2016-02-10 | 西北工业大学 | Hyperspectral image classification method based on spectral-spatial cooperation of deep convolutional neural network |
CN105654117A (en) * | 2015-12-25 | 2016-06-08 | 西北工业大学 | Hyperspectral image spectral-spatial cooperative classification method based on SAE depth network |
CN105787517A (en) * | 2016-03-11 | 2016-07-20 | 西安电子科技大学 | Polarized SAR image classification method base on wavelet sparse auto encoder |
CN106096652A (en) * | 2016-06-12 | 2016-11-09 | 西安电子科技大学 | Based on sparse coding and the Classification of Polarimetric SAR Image method of small echo own coding device |
CN106651899A (en) * | 2016-12-09 | 2017-05-10 | 东北大学 | Fundus image micro-aneurysm detection system based on Adaboost |
CN106815601A (en) * | 2017-01-10 | 2017-06-09 | 西安电子科技大学 | Hyperspectral image classification method based on recurrent neural network |
CN107341511A (en) * | 2017-07-05 | 2017-11-10 | 西安电子科技大学 | Classification of Polarimetric SAR Image method based on super-pixel Yu sparse self-encoding encoder |
CN107590515A (en) * | 2017-09-14 | 2018-01-16 | 西安电子科技大学 | The hyperspectral image classification method of self-encoding encoder based on entropy rate super-pixel segmentation |
CN107798349A (en) * | 2017-11-03 | 2018-03-13 | 合肥工业大学 | A kind of transfer learning method based on the sparse self-editing ink recorder of depth |
CN108537233A (en) * | 2018-03-15 | 2018-09-14 | 南京师范大学 | A kind of pathology brain image sorting technique based on the sparse self-encoding encoder of depth stack |
CN109033952A (en) * | 2018-06-12 | 2018-12-18 | 杭州电子科技大学 | M-sequence recognition methods based on sparse self-encoding encoder |
CN108921233A (en) * | 2018-07-31 | 2018-11-30 | 武汉大学 | A kind of Raman spectrum data classification method based on autoencoder network |
CN109102019A (en) * | 2018-08-09 | 2018-12-28 | 成都信息工程大学 | Image classification method based on HP-Net convolutional neural networks |
CN109145832A (en) * | 2018-08-27 | 2019-01-04 | 大连理工大学 | Polarimetric SAR image semisupervised classification method based on DSFNN Yu non local decision |
Non-Patent Citations (1)
Title |
---|
SHAN JUAN ET AL.: "A Deep Learning Method for Microaneurysm Detection in Fundus Images", 《2016 IEEE FIRST INTERNATIONAL CONFERENCE ON CONNECTED HEALTH: APPLICATIONS, SYSTEMS AND ENGINEERING TECHNOLOGIES》 * |
Cited By (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110472667A (en) * | 2019-07-19 | 2019-11-19 | 广东工业大学 | Small object classification method based on deconvolution neural network |
CN110472667B (en) * | 2019-07-19 | 2024-01-09 | 广东工业大学 | Small target classification method based on deconvolution neural network |
CN110930409A (en) * | 2019-10-18 | 2020-03-27 | 电子科技大学 | Salt body semantic segmentation method based on deep learning and semantic segmentation model |
CN110930409B (en) * | 2019-10-18 | 2022-10-14 | 电子科技大学 | Salt body semantic segmentation method and semantic segmentation system based on deep learning |
CN110972174A (en) * | 2019-12-02 | 2020-04-07 | 东南大学 | Wireless network interruption detection method based on sparse self-encoder |
CN110972174B (en) * | 2019-12-02 | 2022-12-30 | 东南大学 | Wireless network interruption detection method based on sparse self-encoder |
CN111462817A (en) * | 2020-03-25 | 2020-07-28 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Classification model construction method and device, classification model and classification method |
CN111462817B (en) * | 2020-03-25 | 2023-06-20 | 哈尔滨工业大学(深圳)(哈尔滨工业大学深圳科技创新研究院) | Classification model construction method and device, classification model and classification method |
CN112465042A (en) * | 2020-12-02 | 2021-03-09 | 中国联合网络通信集团有限公司 | Generation method and device of classification network model |
CN112465042B (en) * | 2020-12-02 | 2023-10-24 | 中国联合网络通信集团有限公司 | Method and device for generating classified network model |
Also Published As
Publication number | Publication date |
---|---|
CN109816002B (en) | 2022-09-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109816002A (en) | The single sparse self-encoding encoder detection method of small target migrated certainly based on feature | |
CN107145908B (en) | A kind of small target detecting method based on R-FCN | |
CN105825511B (en) | A kind of picture background clarity detection method based on deep learning | |
Chen et al. | Assessing four neural networks on handwritten digit recognition dataset (MNIST) | |
CN107391703B (en) | The method for building up and system of image library, image library and image classification method | |
CN111858989B (en) | Pulse convolution neural network image classification method based on attention mechanism | |
CN107909101A (en) | Semi-supervised transfer learning character identifying method and system based on convolutional neural networks | |
CN110020682A (en) | A kind of attention mechanism relationship comparison net model methodology based on small-sample learning | |
CN107563999A (en) | A kind of chip defect recognition methods based on convolutional neural networks | |
CN108961245A (en) | Picture quality classification method based on binary channels depth parallel-convolution network | |
CN108052984A (en) | Method of counting and device | |
CN112101328A (en) | Method for identifying and processing label noise in deep learning | |
CN105957086A (en) | Remote sensing image change detection method based on optimized neural network model | |
CN110443367A (en) | A kind of method of strength neural network model robust performance | |
CN107016415A (en) | A kind of coloured image Color Semantic sorting technique based on full convolutional network | |
CN108647718A (en) | A kind of different materials metallographic structure is classified the method for grading automatically | |
CN106203625A (en) | A kind of deep-neural-network training method based on multiple pre-training | |
CN106780546B (en) | The personal identification method of motion blur encoded point based on convolutional neural networks | |
CN106777402B (en) | A kind of image retrieval text method based on sparse neural network | |
CN108596274A (en) | Image classification method based on convolutional neural networks | |
CN111582397A (en) | CNN-RNN image emotion analysis method based on attention mechanism | |
CN109146873A (en) | A kind of display screen defect intelligent detecting method and device based on study | |
CN107506350A (en) | A kind of method and apparatus of identification information | |
CN110263174A (en) | - subject categories the analysis method based on focus | |
CN109740656A (en) | A kind of ore method for separating based on convolutional neural networks |
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 |