CN110321967A - Image classification innovatory algorithm based on convolutional neural networks - Google Patents
Image classification innovatory algorithm based on convolutional neural networks Download PDFInfo
- Publication number
- CN110321967A CN110321967A CN201910624323.0A CN201910624323A CN110321967A CN 110321967 A CN110321967 A CN 110321967A CN 201910624323 A CN201910624323 A CN 201910624323A CN 110321967 A CN110321967 A CN 110321967A
- Authority
- CN
- China
- Prior art keywords
- image
- neural networks
- convolutional neural
- convolution
- characteristic pattern
- 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
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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/23—Clustering techniques
- G06F18/232—Non-hierarchical techniques
- G06F18/2321—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions
- G06F18/23213—Non-hierarchical techniques using statistics or function optimisation, e.g. modelling of probability density functions with fixed number of clusters, e.g. K-means clustering
-
- 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
-
- 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
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Evolutionary Computation (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Molecular Biology (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of innovatory algorithms of image classification based on convolutional neural networks, use AlexNet network model for basic framework, input picture is first suitably pre-processed and data enhance, to reduce to network to the dependence of sample size, feature extraction is carried out by neural network convolutional layer, main feature is retained by pond layer again, reduce by next layer of parameter and calculation amount simultaneously, using the method for multiple dimensioned convolution, so that network model no longer limits the size of input picture, further dimensionality reduction is further carried out to characteristic pattern using LDA algorithm, finally obtain the prediction classification to picture.Image classification innovatory algorithm based on convolutional neural networks of the invention can reduce network model to the dependence of sample size, number of parameters can be further decreased by using LDA algorithm and using multiple dimensioned convolution, simplify calculation amount, and improves the accuracy rate of picture classification.
Description
Technical field
The invention belongs to deep learnings, field of image processing, are related to image classification identification mission in improved depth
Application under nerual network technique.
Background technique
Since the input layer of convolutional neural networks can directly handle multidimensional data, so convolutional neural networks are in computer
Visual field is widely used.And the development of constantly driving society is digitized, data volume size is not allowed in the past yet, various seas
It measures data to continuously emerge, is a very big challenge for neural network.It is various to be directed in order to accelerate the study of neural network
The optimization algorithm of CNN constantly emerges.Currently, for convolutional neural networks be mainly model depth and width and
The direction of data processing optimizes.2018, the convolutional neural networks model that Gao Nuo et al. is proposed based on LeCun et al., needle
The problems such as to its gradient disperse and slow convergence rate, traditional several activation primitives are combined and are improved, for activating letter
Number Sigmoid and Softplus combines to obtain a kind of new CNN model, and is applied to the identification of digital handwriting body.Make
Convolutional neural networks are obtained to increase for the accuracy rate of the identification of digital handwriting body；Meanwhile reducing the instruction of network after improvement
Practice parameter, neural network structure is made to become simpler, it is more adaptable.The same year Wang Gaihua et al. proposes a kind of using unsupervised
The image classification model for practising algorithm and convolution construction is constituted from input without the identical image block of size is randomly selected in label image
Data set is pre-processed, secondly, pretreated image block is extracted dictionary by K-means clustering algorithm twice, and is adopted
With discrete convolution operation extraction final image feature finally, being classified using characteristics of image of the Softmax classifier to extraction,
Image classification accuracy is improved, training complexity is reduced.As a whole, most optimization design is all the knot in network model
Set about at structure, various pretreatment operations is carried out by using different activation primitives or to image, to improve nerve net
The speed and accuracy of network model.
To solve convolutional neural networks to the limitation of input picture and further decreasing trained complexity and quickening model
Convergence rate, the invention proposes a kind of image classification algorithms based on improved convolutional neural networks, so that the network is no longer
It limits the size of input picture and preferably reduces network parameter amount, reach higher accuracy and speed.
Summary of the invention
Goal of the invention: the invention proposes a kind of image classification innovatory algorithm based on convolutional neural networks, to realize
Precisely, efficiently quickly image is identified and is classified.
Technical solution: a kind of image classification innovatory algorithm based on convolutional neural networks specifically includes the following steps:
Step 1: carrying out image enhancement, filtering and noise reduction pretreatment operation to input picture, with special to image after reduction
Levy the influence extracted；
Step 2: the image obtained after pretreatment carries out convolution operation, extracts characteristics of image, and use maximum Chi Huafang
Method carries out pond, extracts the maximum pixel of acceptance region intermediate value, gives up rest of pixels point, obtained characteristic pattern is in size reduction
The key message of image is also retained simultaneously, reduces convolution kernel size, then carries out the operation of convolution pondization again, and output is more taken out
The characteristic pattern of elephant；
Step 3: the characteristic pattern obtained by pond convolution operation that step 2 is obtained passes through multiple continuous convolutional layers
Convolution is carried out, the feature in different channels is adequately merged, and this feature figure is sent to pyramid pond layer and carries out pond
Change；
Step 4: characteristic pattern is subjected to convolution operation by multiple dimensioned convolutional layer, thus obtains fixed size
Characteristic pattern；
Step 5: dimensionality reduction and classification further are carried out to characteristic pattern using LDA method, is projected, will be thrown using LDA
The sample characteristics information of movie queen, which is brought into probability density function, to be calculated, and obtains probability distribution information, output calculated result and pre-
Survey classification.
Further, in step 1, gaussian filtering is carried out for input picture, to inhibit noise, smoothed image, and simultaneously
Image is overturn, color, saturation and contrast are adjusted.
Further, four layers of continuous convolutional layer has been used to fill the feature in the different channels of image in step 3 altogether
Divide fusion.
Further, in step 4, characteristic pattern is mapped using three kinds of different scales, be respectively adopted three kinds it is different
Convolution operation carries out convolution, no matter so that input picture size, can finally obtain fixed-size characteristic pattern.
Further, in step 5, dimensionality reduction, global Scatter Matrix S are carried out to eigenmatrix using LDAtIs defined as:
Wherein m is total number of samples, xiFor i-th of sample vector, μ is the mean vector of all samples, and T is to ask in matrix theory
The mathematic sign of transposed matrix.
Within class scatter matrix SωIs defined as:
Wherein N is the total classification number of sample, XiFor the i-th class sample matrix, x is the vector of i-th class each sample, μiIt is i-th
The mean vector of all samples of class.
Inter _ class relationship matrix SbIs defined as:
Sb=St-Sω
Therefore optimization aim is defined as:
Wherein W ∈ Rd×(N-1), it is that the matrix of N-1 feature vector composition is calculated one group by optimization aim formula
The projection matrix that optimal discriminant vector is constituted, the matrix project N-dimensional feature space, and the low-dimensional feature of output N-1 dimension is empty
Between.
Further, present networks have overlapping region between adjacent pool window using the maximum pond of overlapping, can be with
Lifting feature it is rich, avoid over-fitting.
Further, in neural network in data be sent in activation primitive and calculate, used activation primitive is to repair
Linear positive unit Leaky ReLU, which can suitably retain the information of negative axis, so that characteristic information is calculated by the function
Later, the information between minus zone will not be lost completely, and obtained information is more complete.
The invention adopts the above technical scheme compared with prior art, has following technical effect that
1, the present invention no longer limits the size of input image size, is made by the multiple dimensioned convolutional layer structure in present networks
Present networks can input the picture of arbitrary dimension size, and image feature information has obtained significantly retaining, and final is accurate
Rate is also improved.
2, invention further reduces the parameter amount of model, the arithmetic speed of network is accelerated.Present invention employs LDA
Algorithm carries out the study of similarity, enhances the discriminating power of feature.It also has the function of dimensionality reduction simultaneously, and image is main
Characteristic information is simultaneously not affected by destruction, so that the network is more efficiently quick.
3, the image classification innovatory algorithm of the invention based on convolutional neural networks can reduce network model to sample number
The dependence of amount can further decrease number of parameters by using LDA algorithm and using multiple dimensioned convolution, simplify and calculate
Amount, and improve the accuracy rate of picture classification.
Detailed description of the invention
Fig. 1 is improved multiple dimensioned convolutional layer structure chart proposed by the present invention.
Fig. 2 is the improved image classification algorithms network structure of the present invention.
Specific embodiment
Further explanation is done to the present invention with reference to the accompanying drawing.
As shown in Figure 1, a kind of image classification innovatory algorithm based on convolutional neural networks, what is proposed is improved multiple dimensioned
Convolutional layer structure maps characteristic pattern using three kinds of different scales, then respectively 8 × 8,6 × 6,4 × 4 are respectively adopted
Three kinds of various sizes of convolution kernels carry out convolution to characteristic pattern, and step-length is respectively S2, S1, S1, and wherein convolution kernel size is respectively 2
× 2,3 × 3,1 × 1, convolution nuclear volume is 256,256,256, Leaky ReLU be activation primitive, so that no matter input
How is image size, can finally obtain fixed-size characteristic pattern.
As shown in Fig. 2, a kind of image classification innovatory algorithm based on convolutional neural networks, described based on convolutional Neural net
The image classification innovatory algorithm of network, which is first passed through, carries out image pretreatment operation to image, obtains characteristic information distinctness, interference information
Less image.Next the image again obtained pretreatment operation is input in the neural network, passes through two layers of convolution pond
Layer group carries out feature extraction to image, and used convolution kernel size is larger, for extracting the more apparent edge feature of image
Then obtained characteristic pattern is had the convolutional layer of identical convolution kernel size by information by continuous four-ply, for more extracting
The characteristic information in image difference channel, while multi-channel information sufficiently being merged, it is more abstracted, is more representative
Characteristic pattern.In the process, using the convolution kernel size of setting and suitable step sizes, so that picture size size is in a system
It does not change in column convolution process.After the pyramid pond layer in the present invention program, all different sizes
Input picture is all converted to the characteristic pattern of fixed size, carries out computing and sorting using LDA algorithm again on this basis, obtains
The final prediction tag along sort of input picture.Specific implementation process is as follows:
Step 1: the series of preprocessing such as image enhancement, filtering and noise reduction are carried out to input picture and are operated, after reducing
Influence to image characteristics extraction；
Step 2: the image obtained after pretreatment carries out convolution operation, extracts characteristics of image, and use maximum Chi Huafang
Method carries out pond, reduces convolution kernel size after obtaining characteristic pattern, carries out the operation of convolution pondization, the feature being more abstracted again
Figure, then carries out the Fusion Features of next step again；
Step 3: the characteristic pattern that step 2 is obtained carries out convolution by multiple continuous convolutional layers, by different channels
Feature is adequately merged, so that the characteristic pattern finally obtained is more abstract, more representative, and this feature figure is sent to gold
Word tower basin layer carries out pond；
Step 4: after characteristic pattern is mapped according to different scale, various sizes of convolution kernel is respectively adopted and carries out
Convolution operation thus obtains the characteristic pattern of fixed size；
Step 5: dimensionality reduction and classification further are carried out to characteristic pattern using LDA method, is projected, will be thrown using LDA
The sample characteristics information of movie queen is brought into probability density function, its probability for belonging to some classification is calculated, maximum general
Rate is the prediction classification of the image.
For multiple dimensioned convolutional layer in network structure there are many levels, specific structure is as follows:
Characteristic pattern is mapped using three kinds of different scales, respectively 8 × 8,6 × 6,4 × 4.Next three kinds are used
Different convolution kernels carries out convolution operation to corresponding mappings characteristics figure respectively, and step-length is respectively S2, S1, S1, wherein convolution kernel
Size is respectively 2 × 2,3 × 3,1 × 1, convolution nuclear volume is respectively 256,256,256, Leaky ReLU be activation primitive.It is logical
Such structure is crossed, no matter so that input picture size, can finally obtain fixed-size characteristic pattern, to no longer limit
The size of input picture processed.
The above is only a preferred embodiment of the present invention, it is noted that for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (7)
1. a kind of image classification innovatory algorithm based on convolutional neural networks, it is characterised in that: specifically include the following steps:
Step 1: carrying out image enhancement, filtering and noise reduction pretreatment operation to input picture, to mention after reduction to characteristics of image
The influence taken；
Step 2: obtained image carries out convolution operation after pretreatment, extracts characteristics of image, and using maximum pond method into
Row pond extracts the maximum pixel of acceptance region intermediate value, gives up rest of pixels point, obtained characteristic pattern is while size reduction
The key message of image is also retained, convolution kernel size is reduced, then carries out the operation of convolution pondization again, what output was more abstracted
Characteristic pattern；
Step 3: the characteristic pattern of step 2 output obtained by pond convolution operation is carried out by multiple continuous convolutional layers
Convolution is adequately merged the feature in different channels, and this feature figure is sent to pyramid pond layer and carries out pond；
Step 4: characteristic pattern is subjected to convolution operation by multiple dimensioned convolutional layer, thus obtains the feature of fixed size
Figure；
Step 5: dimensionality reduction and classification further are carried out to characteristic pattern using LDA method, projected using LDA, after projection
Sample characteristics information be brought into probability density function and calculate, obtain probability distribution information, output calculated result and prediction class
Not.
2. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: step
In rapid one, gaussian filtering is carried out for input picture, to inhibit noise, smoothed image, and image is overturn simultaneously, color
Color, saturation and contrast are adjusted.
3. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: step
Four layers of continuous convolutional layer have been used sufficiently to merge the feature in the different channels of image in rapid three altogether.
4. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: step
In rapid four, characteristic pattern is mapped using three kinds of different scales, three kinds of different convolution operations are respectively adopted and carry out convolution, make
No matter obtaining input picture size, fixed-size characteristic pattern can be finally obtained.
5. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: step
In rapid five, dimensionality reduction, global Scatter Matrix S are carried out to eigenmatrix using LDAtIs defined as:
Wherein m is total number of samples, xiFor i-th of sample vector, μ is the mean vector of all samples, and T is to seek transposition in matrix theory
The mathematic sign of matrix.
Within class scatter matrix SωIs defined as:
Wherein N is the total classification number of sample, XiFor the i-th class sample matrix, x is the vector of i-th class each sample, μiFor the i-th class institute
There is the mean vector of sample.
Inter _ class relationship matrix SbIs defined as:
Sb=St-Sω
Therefore optimization aim is defined as:
Wherein W ∈ Rd×(N-1), be the matrix that N-1 feature vector forms, by optimization aim formula, be calculated one group it is optimal
The projection matrix that discriminant vector is constituted, the matrix project N-dimensional feature space, the low-dimensional feature space of output N-1 dimension.
6. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: this
Network is had overlapping region between adjacent pool window, can be kept away with the rich of lifting feature using the maximum pond of overlapping
Exempt from over-fitting.
7. a kind of image classification innovatory algorithm based on convolutional neural networks according to claim 1, it is characterised in that: mind
It is sent in activation primitive and calculates through the data in network, used activation primitive is amendment linear unit Leaky
ReLU。
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624323.0A CN110321967B (en) | 2019-07-11 | 2019-07-11 | Image classification improvement method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910624323.0A CN110321967B (en) | 2019-07-11 | 2019-07-11 | Image classification improvement method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110321967A true CN110321967A (en) | 2019-10-11 |
| CN110321967B CN110321967B (en) | 2021-06-01 |
Family
ID=68121994
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910624323.0A Active CN110321967B (en) | 2019-07-11 | 2019-07-11 | Image classification improvement method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110321967B (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
| CN111178507A (en) * | 2019-12-26 | 2020-05-19 | 集奥聚合(北京)人工智能科技有限公司 | Atlas convolution neural network data processing method and device |
| CN111325149A (en) * | 2020-02-20 | 2020-06-23 | 中山大学 | Video action identification method based on voting time sequence correlation model |
| CN112132145A (en) * | 2020-08-03 | 2020-12-25 | 深圳大学 | Image classification method and system based on model extended convolutional neural network |
| CN112731410A (en) * | 2020-12-25 | 2021-04-30 | 上海大学 | Underwater target sonar detection method based on CNN |
| CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113205111A (en) * | 2021-04-07 | 2021-08-03 | 零氪智慧医疗科技(天津)有限公司 | Identification method and device suitable for liver tumor and electronic equipment |
| CN113435389A (en) * | 2021-07-09 | 2021-09-24 | 大连海洋大学 | Chlorella and chrysophyceae classification and identification method based on image feature deep learning |
| CN114387467A (en) * | 2021-12-09 | 2022-04-22 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN115239928A (en) * | 2022-09-05 | 2022-10-25 | 四川蜀天信息技术有限公司 | 3D data large-screen visualization system based on GIS |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096605A (en) * | 2016-06-02 | 2016-11-09 | 史方 | A kind of image obscuring area detection method based on degree of depth study and device |
| CN106897739A (en) * | 2017-02-15 | 2017-06-27 | 国网江苏省电力公司电力科学研究院 | A kind of grid equipment sorting technique based on convolutional neural networks |
| JP2017220222A (en) * | 2016-06-06 | 2017-12-14 | 富士通株式会社 | Method, program and apparatus for comparing data graphs |
-
2019
- 2019-07-11 CN CN201910624323.0A patent/CN110321967B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106096605A (en) * | 2016-06-02 | 2016-11-09 | 史方 | A kind of image obscuring area detection method based on degree of depth study and device |
| JP2017220222A (en) * | 2016-06-06 | 2017-12-14 | 富士通株式会社 | Method, program and apparatus for comparing data graphs |
| CN106897739A (en) * | 2017-02-15 | 2017-06-27 | 国网江苏省电力公司电力科学研究院 | A kind of grid equipment sorting technique based on convolutional neural networks |
Cited By (17)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110956611A (en) * | 2019-11-01 | 2020-04-03 | 武汉纺织大学 | Smoke detection method integrated with convolutional neural network |
| CN111178507A (en) * | 2019-12-26 | 2020-05-19 | 集奥聚合(北京)人工智能科技有限公司 | Atlas convolution neural network data processing method and device |
| CN111325149A (en) * | 2020-02-20 | 2020-06-23 | 中山大学 | Video action identification method based on voting time sequence correlation model |
| CN111325149B (en) * | 2020-02-20 | 2023-05-26 | 中山大学 | Video action recognition method based on time sequence association model of voting |
| CN112132145A (en) * | 2020-08-03 | 2020-12-25 | 深圳大学 | Image classification method and system based on model extended convolutional neural network |
| CN112132145B (en) * | 2020-08-03 | 2023-08-01 | 深圳大学 | Image classification method and system based on model extended convolutional neural network |
| CN112731410B (en) * | 2020-12-25 | 2021-11-05 | 上海大学 | Underwater target sonar detection method based on CNN |
| CN112731410A (en) * | 2020-12-25 | 2021-04-30 | 上海大学 | Underwater target sonar detection method based on CNN |
| CN113052189B (en) * | 2021-03-30 | 2022-04-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113052189A (en) * | 2021-03-30 | 2021-06-29 | 电子科技大学 | Improved MobileNet V3 feature extraction network |
| CN113205111A (en) * | 2021-04-07 | 2021-08-03 | 零氪智慧医疗科技(天津)有限公司 | Identification method and device suitable for liver tumor and electronic equipment |
| CN113435389A (en) * | 2021-07-09 | 2021-09-24 | 大连海洋大学 | Chlorella and chrysophyceae classification and identification method based on image feature deep learning |
| CN113435389B (en) * | 2021-07-09 | 2024-03-01 | 大连海洋大学 | Chlorella and golden algae classification and identification method based on image feature deep learning |
| CN114387467A (en) * | 2021-12-09 | 2022-04-22 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN114387467B (en) * | 2021-12-09 | 2022-07-29 | 哈工大（张家口）工业技术研究院 | Medical image classification method based on multi-module convolution feature fusion |
| CN115239928A (en) * | 2022-09-05 | 2022-10-25 | 四川蜀天信息技术有限公司 | 3D data large-screen visualization system based on GIS |
| CN115239928B (en) * | 2022-09-05 | 2022-12-20 | 四川蜀天信息技术有限公司 | 3D data large-screen visualization system based on GIS |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110321967B (en) | 2021-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110321967A (en) | Image classification innovatory algorithm based on convolutional neural networks | |
| CN104537393B (en) | A kind of traffic sign recognition method based on multiresolution convolutional neural networks | |
| Taylor et al. | Improving deep learning with generic data augmentation | |
| CN112308158A (en) | Multi-source field self-adaptive model and method based on partial feature alignment | |
| CN105426919B (en) | The image classification method of non-supervisory feature learning is instructed based on conspicuousness | |
| CN108304826A (en) | Facial expression recognizing method based on convolutional neural networks | |
| CN111832546B (en) | Lightweight natural scene text recognition method | |
| CN106960206A (en) | Character identifying method and character recognition system | |
| CN113887610A (en) | Pollen image classification method based on cross attention distillation transducer | |
| CN107085704A (en) | Fast face expression recognition method based on ELM own coding algorithms | |
| CN107609638A (en) | A kind of method based on line decoder and interpolation sampling optimization convolutional neural networks | |
| CN107451545A (en) | The face identification method of Non-negative Matrix Factorization is differentiated based on multichannel under soft label | |
| CN108681706A (en) | A kind of double source remotely-sensed data semantic segmentation method | |
| CN106503616A (en) | A kind of Mental imagery Method of EEG signals classification of the learning machine that transfinited based on layering | |
| CN110599502A (en) | Skin lesion segmentation method based on deep learning | |
| CN113378949A (en) | Dual-generation confrontation learning method based on capsule network and mixed attention | |
| Lin et al. | Lateral refinement network for contour detection | |
| Sethy et al. | Off-line Odia handwritten numeral recognition using neural network: a comparative analysis | |
| Liu et al. | Image retrieval using CNN and low-level feature fusion for crime scene investigation image database | |
| Tereikovskyi et al. | The method of semantic image segmentation using neural networks | |
| CN114882278A (en) | Tire pattern classification method and device based on attention mechanism and transfer learning | |
| CN111310820A (en) | Foundation meteorological cloud chart classification method based on cross validation depth CNN feature integration | |
| CN103336830B (en) | Image search method based on structure semantic histogram | |
| CN114241458A (en) | Driver behavior recognition method based on attitude estimation feature fusion | |
| Cui et al. | Hand gesture segmentation against complex background based on improved atrous spatial pyramid pooling |
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 |