CN107808132A - A kind of scene image classification method for merging topic model - Google Patents
A kind of scene image classification method for merging topic model Download PDFInfo
- Publication number
- CN107808132A CN107808132A CN201710992656.XA CN201710992656A CN107808132A CN 107808132 A CN107808132 A CN 107808132A CN 201710992656 A CN201710992656 A CN 201710992656A CN 107808132 A CN107808132 A CN 107808132A
- Authority
- CN
- China
- Prior art keywords
- image
- classification
- model
- training
- neural networks
- 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
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/35—Categorising the entire scene, e.g. birthday party or wedding scene
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
A kind of scene image classification method for merging topic model is claimed in the present invention, is related to deep learning and image classification field.This method includes：Data set is pre-processed, expands the quantity for having obtained data set, obtains the image data format for meeting deep learning model treatment；Structure meets the convolutional neural networks model of scene image classification, and pre-training is carried out to the image data set after processing using convolutional neural networks；Use training set, repetitive exercise end to end is carried out to the convolutional neural networks of structure, adjust the parameter in network, the model completed using checking set pair training is verified, the scene image feature with judgement index extracted is modeled, existing hiding theme variable between feature and image is extracted, obtains the image subject distribution of k dimensional vectors representative, k represents theme quantity；Each image can regard a probability distribution graph being made up of multiple themes as, and the classification of scene image is realized using grader.
Description
Technical field
The invention belongs to deep learning and image classification identification technology field, specifically a kind of scene for merging topic model
Image classification method.
Background technology
Scene image is classified, that is, gives one group of scene graph for including multiple target classifications (such as mountain range, river, highway)
Picture, the global semanteme of image is analyzed and understood according to each target class other distribution relation.Scene image is classified not only
There is overall understanding to the classification of entire image, also the contextual information in image between each object and region is carried out
Analysis so as to having had deeper understanding in image, promoted such as target identification, image in machine vision to examine
The development in the fields such as rope, is with a wide range of applications.
With the fast development of intelligent photographing device and computer hardware, the depth convolutional Neural net in deep learning is used
Network substitutes traditional images feature extracting method, and avoiding traditional algorithm for pattern recognition needs to carry out the cumbersome of manual features extraction
Process, there is more powerful feature extraction and feature representation ability.Use the convolutional neural networks mould of deep learning Algorithm for Training
Type also achieves abundant achievement in actual applications from great breakthrough is not only achieved in theory.
Deep learning model belongs to multilayer neural network model, and convolutional neural networks are first depth successfully trained
Learning model.Deep-neural-network model uses the primitive nature figure not Jing Guo any man's activity in the training stage
Picture, it very effective can learn the graphical representation with Invariance features such as rotation, distortions, this is largely reduced
Influence of the manual extraction feature to classification accuracy.The trainable multi-layer rack that convolutional neural networks are made up of many levels
Structure, using processing mode end to end, the process of image preprocessing and feature extraction is regarded as a flight data recorder, passes through deconvolution
Operation or the precision optimizing network parameter of analysis classification results, improve constantly the classification essence of the convolutional neural networks model of training
Degree.
The content of the invention
Present invention seek to address that for image in prior art bottom visual information and people to the high level of image understanding
Semantic gap problem caused by the inconsistency of semantic information between bottom and high level.Propose a kind of effect for obtaining Classification and Identification
The scene image classification method for the fusion topic model that fruit is further enhanced.Technical scheme is as follows：
A kind of scene image classification method for merging topic model, it comprises the following steps：
1) pretreatment operation, is carried out to the image data set of acquisition, this there are two benefits, is to play EDS extended data set first
Effect, add training samples number, secondly by pre-training so that processing after picture format meet deep learning framework
Form, be used as training set using the 70% of treated data set, remaining 30% collects as checking；
2), structure meet scene image classification convolutional neural networks model, using convolutional neural networks to processing after
Training set carries out pre-training, i.e., extracts the position of input picture, context by the convolutional layer and pond layer of convolutional neural networks
Feature, the position of the image learnt by the use of these, contextual feature are used as the basis of image scene classification and foundation；
3), using training set, repetitive exercise end to end is carried out to the convolutional neural networks of structure, adjusted in the training process
Parameter in whole network, the model completed using checking set pair training are verified that the characteristics of image of network model extraction judges
The classification of scene image；
4), the scene image feature with judgement index extracted by pre-training is modeled, extracts feature and figure
The existing hiding theme variable as between, obtain the image subject distribution of k dimensional vectors representative；Each image represents multiple themes
The probability distribution formed, the classification of scene image is realized using grader.
Further, the step of step 1) carries out pretreatment operation to image data set includes cutting, overturn, Nogata
Figure equalization, adjusts the brightness of image, obtains the data set of more Large Copacity, so can obtain the scene image of each classification
To sufficient training, the generalization ability of network is heightened, on the other hand causes the picture format after processing to meet deep learning framework
Form so that image can by network model read and train.
Further, the convolutional neural networks model for meeting scene image classification, including data input layer, convolution meter
Layer, pond layer, full articulamentum, output layer are calculated, data input layer is the data input of pretreatment to network model；Convolutional calculation
Layer regards a filter as using local connection, each neuron, by doing slide to window, value and filtering in window
The corresponding multiplication of value in device, as next layer of input；The position of pond layer is among continuous convolutional layer, for compressing number
According to the quantity with parameter, over-fitting is reduced；Full articulamentum is located at the afterbody of convolutional neural networks model, and maximum possible utilizes warp
The original input information of a small amount of information reverting retained is crossed behind sliding window and pond；Output layer is a small amount of letter reservation
Breath is input to Softmax normalized functions, and its effect is that the output result of network is normalized into probability distribution, probability highest
One is classification results.
Further, the step 3) uses training set, and iteration instruction end to end is carried out to the convolutional neural networks of structure
White silk specifically includes：The convolution of convolutional neural networks forward process be that the correspondence position quadrature of convolution kernel and input picture is summed again
Process, input of the value tried to achieve as next operation, each convolution kernel mobile position on the image of input, from upper
Arrive down, the output matrix obtained after from left to right covering one time is exactly the characteristic pattern of the input of next operation；In error signal
In back-propagation process, hiding Es-region propagations of the error signal by grader forwardly；Model learns net automatically in the training process
Network parameter, update weight；The model completed using checking set pair training is verified, enables the characteristics of image of network model extraction
Enough classifications for more accurately judging scene image.
Further, in the characteristics of image of network model extraction, some of which be characterized in it is similar or identical, this
There is the concept that similar or identical feature abstraction is the theme a bit；The probability topic model of statistics text classification is copied, will be extracted
To image low layer local feature be quantified as vision word, obtained by counting the frequency that each vision word occurs in entire image
Histogram to image represents, each image is considered as into a word frequency vector, each image represents one be made up of multiple themes
Individual probability distribution, each theme represent the probability distribution that multiple vision words are formed again；The characteristic vector that will have been preserved
It is input in topic model, according to influence of the different theme quantity of experimental analysis to classification accuracy, determines theme quantity
Number.
Further, the convolutional neural networks model is using the deep learning framework TensorFlow to increase income, in the frame
Convolutional neural networks model is built on frame, TensorFlow can be run on all kinds of machines, can be simultaneously in multiple CPU, GPU
Or both mixing machine on run.
Advantages of the present invention and have the beneficial effect that：
The present invention considers the easy availability of existing picture and fast development and the depth learning technology of computer hardware
The achievement of acquirement, select than traditional artificial extraction feature, multiple features to melt using convolutional neural networks model extraction characteristics of image
The feature that the methods of conjunction obtains has more objectivity, with more judgement index.Allow the network to learn by adjusting network parameter
The feature of scene image can largely be differentiated, the feature extracted is preserved；Many features are similar in these features
Or it is similar, in order to avoid useful feature is unintentionally abandoned during classification, these are had similar or phase by we
The concept that the feature extraction of nearly meaning is the theme, this also compensate for the gap problem that low-level feature directly arrives high-level characteristic, reduce
The feature quantity of grader is input to, improves classification effectiveness and the classification degree of accuracy.
Brief description of the drawings
Fig. 1 is that the present invention provides the flow chart that preferred embodiment realizes the method for scene image classification based on deep learning；
The step of Fig. 2 is image preprocessing.
Embodiment
Below in conjunction with the accompanying drawing in the embodiment of the present invention, the technical scheme in the embodiment of the present invention is carried out clear, detailed
Carefully describe.Described embodiment is only the part of the embodiment of the present invention.
The present invention solve above-mentioned technical problem technical scheme be：
Fig. 1 shows the flow chart of the method that scene image classification is realized based on deep learning of the present invention, specific steps
It is as follows：
(1) image data set pre-process, using cutting, overturning, histogram equalization, adjust image
Brightness, EDS extended data set, and adjust picture format be convolutional neural networks model readable format, choose the 70% of data set
As training set, remaining 30% as checking collection；
(2) using the deep learning framework TensorFlow to increase income, convolutional neural networks model is built on the frame.
TensorFlow is google deep learning framework, and the visualization tool TensorBoard carried is one very handy
Network structure visualization tool, it is highly useful for analyzing and training network；TensorFlow can be run on all kinds of machines,
It can be run simultaneously on the machine of multiple CPU, GPU or both mixing, flexibility is good；
Structure meets the convolutional neural networks model of scene image classification, including data input layer, convolutional calculation layer, Chi Hua
Layer, full articulamentum, output layer.Data input layer is the data input of pretreatment to network model；Convolutional layer is connected using local
Connect, each neuron regards a filter as, and by doing slide to window, the value in window is corresponding with the value in filter
It is multiplied, as next layer of input；The position of pond layer is among continuous convolutional layer, for the number of compressed data and parameter
Amount, reduce over-fitting；Full articulamentum be located at the afterbody of convolutional neural networks model, the utilization of maximum possible by sliding window with
The original input information of a small amount of information reverting for retaining behind pond；
(4) training set is used, repetitive exercise end to end is carried out to the convolutional neural networks of structure, adjusted in the training process
Parameter in whole network, the model completed using checking set pair training is verified, enables the characteristics of image of network model extraction
Enough classifications for more accurately judging scene image；In the training process, because the multilayer of network model is calculated using backpropagation
Method learns to the node of hidden layer；
(5) in the feature extracted, some of which is characterized in similar or identical, and we have these similar or identical
The concept that is the theme of feature abstraction；The probability topic model of statistics text classification is copied, the image low layer extracted is local
Characteristic quantification is vision word, and the histogram table of image is obtained by counting the frequency that each vision word occurs in entire image
Show, each image is considered as a word frequency vector, each image represents a probability distribution being made up of multiple themes, Mei Gezhu
Topic represents the probability distribution that multiple vision words are formed again；It is direct low-level feature abstract can be made up by this method
To the gap problem of high-level characteristic mapping, feature that some have judgement index is it also avoid during classification by random something lost
Abandon, improve the classification accuracy of scene image；The characteristic vector preserved is input in topic model, according to experimental analysis not
Influence of the same theme quantity to classification accuracy, determine the number of theme quantity；
(6) each image represents the probability distribution that multiple themes are formed, and scene image is realized using grader
Classification.Input object by extracted theme feature vector as grader, is divided into K class, can set if desired
K output point, the degree of accuracy of scene image classification is judged according to output result.
Fig. 2 shows the step of image is pre-processed, specific as follows：
(1) image is zoomed in and out, overturn：Image scaling conversion be according to certain scaling or downscaled images,
Upset change is that the mirror image for making picture material occur along horizontal or vertical direction changes.There is extensive energy in order to train one
The stronger network model of power, it would be desirable to expand data set, it is more conventional that change is zoomed in and out and overturn to image
Two kinds of image enchancing methods, a width is trained in a network in the form of different images content, improve network
Generalization ability and classification accuracy.
(2) histogram equalization is carried out to image：Histogram equalization is most bright between most dark-part in image by improving
Contrast, find out nuance of the luminance difference away from smaller content in picture material；The image of a more high-contrast is created in addition
It is added in training set, reduces degree of dependence of the image to brightness.
(3) saturation degree of image is adjusted：The saturation degree of adjustment image is the adjustment to pattern colour chroma, and increase saturation degree is just
The purity of image color can be increased, the purity of image color will be reduced by reducing saturation degree, when the saturation degree of image is reduced to 0
When, image is changing to grey.
The above embodiment is interpreted as being merely to illustrate the present invention rather than limited the scope of the invention.
After the content for having read the record of the present invention, technical staff can make various changes or modifications to the present invention, these equivalent changes
Change and modification equally falls into the scope of the claims in the present invention.
Claims (6)
1. a kind of scene image classification method for merging topic model, it is characterised in that comprise the following steps：
1) pretreatment operation, is carried out to the image data set of acquisition, pretreatment one is to have expanded data set, after two are so that processing
Picture format meet the form of deep learning framework, using the 70% of treated data set as training set, remaining 30%
Collect as checking；
2), structure meets the convolutional neural networks model of scene image classification, using convolutional neural networks to the training after processing
Collection carries out pre-training, i.e., extracts the position of input picture, contextual feature using the convolutional layer and pond layer of convolutional neural networks,
The position of the image learnt by the use of these, contextual feature are used as the basis of image scene classification and foundation；
3), using training set, repetitive exercise end to end is carried out to the convolutional neural networks of structure, adjusts net in the training process
Parameter in network, the model completed using checking set pair training are verified that the characteristics of image of network model extraction judges scene
The classification of image；
4), the scene image feature with judgement index extracted by pre-training is modeled, and is extracted between feature and image
Existing hiding theme variable, obtain the image subject distribution of k dimensional vectors representative；Each image represents multiple themes and formed
A probability distribution, the classification of scene image is realized using grader.
2. the scene image classification method of fusion topic model according to claim 1, it is characterised in that the step 1)
The step of carrying out pretreatment operation to image data set includes cutting, overturn, and histogram equalization, adjusts the brightness of image, obtains
To the data set of more Large Copacity, so the scene image of each classification can be made sufficiently to be trained, heighten the general of network
Change ability, the picture format after processing is on the other hand caused to meet the form of deep learning framework so that image can be by network
Model read and training.
3. the scene image classification method of fusion topic model according to claim 1 or 2, it is characterised in that the symbol
Close the convolutional neural networks model of scene image classification, including it is data input layer, convolutional calculation layer, pond layer, full articulamentum, defeated
Go out layer, data input layer is the data input of pretreatment to network model；Convolutional calculation layer is connected using local, each nerve
Member regards a filter as, and by doing slide to window, value in window is corresponding with the value in filter to be multiplied, as under
One layer of input；The position of pond layer is among continuous convolutional layer, for compressed data and the quantity of parameter, reduces plan
Close；Full articulamentum is located at the afterbody of convolutional neural networks model, and the utilization of maximum possible retains after sliding window and pond
The original input information of a small amount of information reverting；Output layer is that a small amount of information of reservation is input to Softmax normalization
Function, it is that the output result of network is normalized into the as classification results of probability distribution probability highest one that it, which is acted on,.
4. the scene image classification method of fusion topic model according to claim 3, it is characterised in that
The step 3) uses training set, and repetitive exercise specifically includes the convolutional neural networks progress to structure end to end：'s
The convolution of convolutional neural networks forward process is the process that the correspondence position quadrature of convolution kernel and input picture is summed again, is tried to achieve
Input of the value as next operation, each convolution kernel mobile position on the image of input, from top to bottom, from left to right
The output matrix obtained after covering one time is exactly the characteristic pattern of the input of next operation；In error signal back-propagation process
In, hiding Es-region propagations of the error signal by grader forwardly；The automatic learning network parameter of model in the training process, renewal power
Weight；The model completed using checking set pair training is verified, the characteristics of image that network model extracts more accurately is sentenced
The classification of disconnected scene image.
5. the scene image classification method of fusion topic model according to claim 4, it is characterised in that the network mould
In the characteristics of image of type extraction, some of which is characterized in similar or identical, these is had similar or identical feature abstraction
The concept being the theme；The probability topic model of statistics text classification is copied, the image low layer local feature extracted is quantified as
Vision word, the histogram that image is obtained by counting the frequency that each vision word occurs in entire image represents, by every width
Image is considered as a word frequency vector, and each image represents a probability distribution being made up of multiple themes, and each theme represents again
The probability distribution that multiple vision words are formed；The characteristic vector preserved is input in topic model, according to experiment
Influence of the different theme quantity to classification accuracy is analyzed, determines the number of theme quantity.
6. the scene image classification method of fusion topic model according to claim 1, it is characterised in that the convolution god
The deep learning framework TensorFlow to increase income is used through network model, builds convolutional neural networks model on the frame,
TensorFlow can be run on all kinds of machines, can be run simultaneously on the machine of multiple CPU, GPU or both mixing.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710992656.XA CN107808132A (en) | 2017-10-23 | 2017-10-23 | A kind of scene image classification method for merging topic model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710992656.XA CN107808132A (en) | 2017-10-23 | 2017-10-23 | A kind of scene image classification method for merging topic model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN107808132A true CN107808132A (en) | 2018-03-16 |
Family
ID=61584638
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710992656.XA Pending CN107808132A (en) | 2017-10-23 | 2017-10-23 | A kind of scene image classification method for merging topic model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107808132A (en) |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108510574A (en) * | 2018-04-17 | 2018-09-07 | 福州大学 | A kind of example-based learning and the 3D rendering method of cutting out for enhancing visual quality |
| CN108537277A (en) * | 2018-04-10 | 2018-09-14 | 湖北工业大学 | A kind of image classification knowledge method for distinguishing |
| CN109165682A (en) * | 2018-08-10 | 2019-01-08 | 中国地质大学（武汉） | A kind of remote sensing images scene classification method merging depth characteristic and significant characteristics |
| CN109166130A (en) * | 2018-08-06 | 2019-01-08 | 北京市商汤科技开发有限公司 | A kind of image processing method and image processing apparatus |
| CN109341703A (en) * | 2018-09-18 | 2019-02-15 | 北京航空航天大学 | A kind of complete period uses the vision SLAM algorithm of CNNs feature detection |
| CN109447096A (en) * | 2018-04-13 | 2019-03-08 | 西安电子科技大学 | A kind of pan path prediction technique and device based on machine learning |
| CN109447092A (en) * | 2018-10-25 | 2019-03-08 | 哈尔滨工程大学 | Access extracting method between ice based on sea ice scene classification |
| CN109614488A (en) * | 2018-12-04 | 2019-04-12 | 广西大学 | Distribution network live line work condition distinguishing method based on text classification and image recognition |
| CN109657602A (en) * | 2018-12-17 | 2019-04-19 | 中国地质大学（武汉） | Automatic functional region of city method and system based on streetscape data and transfer learning |
| CN109948733A (en) * | 2019-04-01 | 2019-06-28 | 深圳大学 | More classification methods, sorter and the storage medium of alimentary tract endoscope image |
| CN110059597A (en) * | 2019-04-04 | 2019-07-26 | 南京理工大学 | Scene recognition method based on depth camera |
| CN110413769A (en) * | 2018-04-25 | 2019-11-05 | 北京京东尚科信息技术有限公司 | Scene classification method, device, storage medium and its electronic equipment |
| CN110598746A (en) * | 2019-08-13 | 2019-12-20 | 武汉大学 | Adaptive scene classification method based on ODE solver |
| CN110648153A (en) * | 2018-06-11 | 2020-01-03 | 中国移动通信集团浙江有限公司 | Change tendency prediction method and device, electronic equipment and storage medium |
| CN111613225A (en) * | 2020-04-27 | 2020-09-01 | 深圳壹账通智能科技有限公司 | Method and system for automatically reporting road violation based on voice and image processing |
| CN111709889A (en) * | 2020-06-09 | 2020-09-25 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN111858349A (en) * | 2020-07-23 | 2020-10-30 | 深圳慕智科技有限公司 | Semantic amplification method based on image semantic cutting and merging |
| CN112088380A (en) * | 2018-06-07 | 2020-12-15 | 法弗人工智能有限公司 | Image segmentation |
| CN112597311A (en) * | 2020-12-28 | 2021-04-02 | 东方红卫星移动通信有限公司 | Terminal information classification method and system based on low-earth-orbit satellite communication |
| CN113269224A (en) * | 2021-03-24 | 2021-08-17 | 华南理工大学 | Scene image classification method, system and storage medium |
| CN113268914A (en) * | 2020-02-14 | 2021-08-17 | 聚积科技股份有限公司 | Method for establishing LED screen adjustment standard judgment model |
| CN114020922A (en) * | 2022-01-06 | 2022-02-08 | 智者四海（北京）技术有限公司 | Text classification method, device and storage medium |
| CN116597237A (en) * | 2023-07-14 | 2023-08-15 | 江西小马机器人有限公司 | Belt start detection method, system, storage medium and computer |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2624149A2 (en) * | 2012-02-02 | 2013-08-07 | Xerox Corporation | Document processing employing probabilistic topic modeling of documents represented as text words transformed to a continuous space |
| CN104268546A (en) * | 2014-05-28 | 2015-01-07 | 苏州大学 | Dynamic scene classification method based on topic model |
| CN104376326A (en) * | 2014-11-02 | 2015-02-25 | 吉林大学 | Feature extraction method for image scene recognition |
| US20170032251A1 (en) * | 2015-07-31 | 2017-02-02 | Intuit Inc. | Method and system for applying probabilistic topic models to content in a tax environment to improve user satisfaction with a question and answer customer support system |
| CN106547880A (en) * | 2016-10-26 | 2017-03-29 | 重庆邮电大学 | A kind of various dimensions geographic scenes recognition methodss of fusion geographic area knowledge |
| CN107239529A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of public sentiment hot category classification method based on deep learning |
-
2017
- 2017-10-23 CN CN201710992656.XA patent/CN107808132A/en active Pending
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2624149A2 (en) * | 2012-02-02 | 2013-08-07 | Xerox Corporation | Document processing employing probabilistic topic modeling of documents represented as text words transformed to a continuous space |
| CN104268546A (en) * | 2014-05-28 | 2015-01-07 | 苏州大学 | Dynamic scene classification method based on topic model |
| CN104376326A (en) * | 2014-11-02 | 2015-02-25 | 吉林大学 | Feature extraction method for image scene recognition |
| US20170032251A1 (en) * | 2015-07-31 | 2017-02-02 | Intuit Inc. | Method and system for applying probabilistic topic models to content in a tax environment to improve user satisfaction with a question and answer customer support system |
| CN106547880A (en) * | 2016-10-26 | 2017-03-29 | 重庆邮电大学 | A kind of various dimensions geographic scenes recognition methodss of fusion geographic area knowledge |
| CN107239529A (en) * | 2017-05-27 | 2017-10-10 | 中国矿业大学 | A kind of public sentiment hot category classification method based on deep learning |
Non-Patent Citations (2)
| Title |
|---|
| 陈培新等: ""融合潜在主题信息和卷积语义特征的文本主题分类"", 《信号处理》 * |
| 顾广华等: ""图像场景语义分类研究进展综述"", 《系统工程与电子技术》 * |
Cited By (30)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108537277A (en) * | 2018-04-10 | 2018-09-14 | 湖北工业大学 | A kind of image classification knowledge method for distinguishing |
| CN109447096A (en) * | 2018-04-13 | 2019-03-08 | 西安电子科技大学 | A kind of pan path prediction technique and device based on machine learning |
| CN108510574A (en) * | 2018-04-17 | 2018-09-07 | 福州大学 | A kind of example-based learning and the 3D rendering method of cutting out for enhancing visual quality |
| CN110413769A (en) * | 2018-04-25 | 2019-11-05 | 北京京东尚科信息技术有限公司 | Scene classification method, device, storage medium and its electronic equipment |
| CN112088380A (en) * | 2018-06-07 | 2020-12-15 | 法弗人工智能有限公司 | Image segmentation |
| CN110648153A (en) * | 2018-06-11 | 2020-01-03 | 中国移动通信集团浙江有限公司 | Change tendency prediction method and device, electronic equipment and storage medium |
| CN109166130A (en) * | 2018-08-06 | 2019-01-08 | 北京市商汤科技开发有限公司 | A kind of image processing method and image processing apparatus |
| CN109166130B (en) * | 2018-08-06 | 2021-06-22 | 北京市商汤科技开发有限公司 | Image processing method and image processing device |
| CN109165682A (en) * | 2018-08-10 | 2019-01-08 | 中国地质大学（武汉） | A kind of remote sensing images scene classification method merging depth characteristic and significant characteristics |
| CN109165682B (en) * | 2018-08-10 | 2020-06-16 | 中国地质大学（武汉） | Remote sensing image scene classification method integrating depth features and saliency features |
| CN109341703A (en) * | 2018-09-18 | 2019-02-15 | 北京航空航天大学 | A kind of complete period uses the vision SLAM algorithm of CNNs feature detection |
| CN109447092A (en) * | 2018-10-25 | 2019-03-08 | 哈尔滨工程大学 | Access extracting method between ice based on sea ice scene classification |
| CN109614488A (en) * | 2018-12-04 | 2019-04-12 | 广西大学 | Distribution network live line work condition distinguishing method based on text classification and image recognition |
| CN109657602A (en) * | 2018-12-17 | 2019-04-19 | 中国地质大学（武汉） | Automatic functional region of city method and system based on streetscape data and transfer learning |
| CN109948733A (en) * | 2019-04-01 | 2019-06-28 | 深圳大学 | More classification methods, sorter and the storage medium of alimentary tract endoscope image |
| CN110059597A (en) * | 2019-04-04 | 2019-07-26 | 南京理工大学 | Scene recognition method based on depth camera |
| CN110598746A (en) * | 2019-08-13 | 2019-12-20 | 武汉大学 | Adaptive scene classification method based on ODE solver |
| CN113268914A (en) * | 2020-02-14 | 2021-08-17 | 聚积科技股份有限公司 | Method for establishing LED screen adjustment standard judgment model |
| CN111613225A (en) * | 2020-04-27 | 2020-09-01 | 深圳壹账通智能科技有限公司 | Method and system for automatically reporting road violation based on voice and image processing |
| CN111709889A (en) * | 2020-06-09 | 2020-09-25 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN111709889B (en) * | 2020-06-09 | 2023-04-07 | 中联云港数据科技股份有限公司 | Image dynamic real-time analysis system and method based on artificial intelligence |
| CN111858349A (en) * | 2020-07-23 | 2020-10-30 | 深圳慕智科技有限公司 | Semantic amplification method based on image semantic cutting and merging |
| CN112597311A (en) * | 2020-12-28 | 2021-04-02 | 东方红卫星移动通信有限公司 | Terminal information classification method and system based on low-earth-orbit satellite communication |
| CN112597311B (en) * | 2020-12-28 | 2023-07-11 | 东方红卫星移动通信有限公司 | Terminal information classification method and system based on low-orbit satellite communication |
| CN113269224A (en) * | 2021-03-24 | 2021-08-17 | 华南理工大学 | Scene image classification method, system and storage medium |
| CN113269224B (en) * | 2021-03-24 | 2023-10-31 | 华南理工大学 | Scene image classification method, system and storage medium |
| CN114020922A (en) * | 2022-01-06 | 2022-02-08 | 智者四海（北京）技术有限公司 | Text classification method, device and storage medium |
| CN114020922B (en) * | 2022-01-06 | 2022-03-22 | 智者四海（北京）技术有限公司 | Text classification method, device and storage medium |
| CN116597237A (en) * | 2023-07-14 | 2023-08-15 | 江西小马机器人有限公司 | Belt start detection method, system, storage medium and computer |
| CN116597237B (en) * | 2023-07-14 | 2023-11-21 | 江西小马机器人有限公司 | Belt start detection method, system, storage medium and computer |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107808132A (en) | A kind of scene image classification method for merging topic model | |
| CN110378985B (en) | Animation drawing auxiliary creation method based on GAN | |
| CN107844795B (en) | Convolutional neural network feature extraction method based on principal component analysis | |
| CN110399821B (en) | Customer satisfaction acquisition method based on facial expression recognition | |
| CN108830252A (en) | A kind of convolutional neural networks human motion recognition method of amalgamation of global space-time characteristic | |
| CN106372648A (en) | Multi-feature-fusion-convolutional-neural-network-based plankton image classification method | |
| CN109583425A (en) | A kind of integrated recognition methods of the remote sensing images ship based on deep learning | |
| CN110472627A (en) | One kind SAR image recognition methods end to end, device and storage medium | |
| CN110210551A (en) | A kind of visual target tracking method based on adaptive main body sensitivity | |
| CN106504064A (en) | Clothes classification based on depth convolutional neural networks recommends method and system with collocation | |
| CN107657257A (en) | A kind of semantic image dividing method based on multichannel convolutive neutral net | |
| CN108961245A (en) | Picture quality classification method based on binary channels depth parallel-convolution network | |
| CN108416266A (en) | A kind of video behavior method for quickly identifying extracting moving target using light stream | |
| CN107563345A (en) | A kind of human body behavior analysis method based on time and space significance region detection | |
| CN107292250A (en) | A kind of gait recognition method based on deep neural network | |
| CN104182772A (en) | Gesture recognition method based on deep learning | |
| CN113762138B (en) | Identification method, device, computer equipment and storage medium for fake face pictures | |
| CN108717524A (en) | It is a kind of based on double gesture recognition systems and method for taking the photograph mobile phone and artificial intelligence system | |
| CN110276402A (en) | A kind of salt body recognition methods based on the enhancing of deep learning semanteme boundary | |
| CN107909081A (en) | The quick obtaining and quick calibrating method of image data set in a kind of deep learning | |
| CN109657582A (en) | Recognition methods, device, computer equipment and the storage medium of face mood | |
| Zhang et al. | Inkthetics: a comprehensive computational model for aesthetic evaluation of Chinese ink paintings | |
| CN108122003A (en) | A kind of Weak target recognition methods based on deep neural network | |
| CN109360179A (en) | A kind of image interfusion method, device and readable storage medium storing program for executing | |
| CN107958219A (en) | Image scene classification method based on multi-model and Analysis On Multi-scale Features |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20180316 |