CN108764347A - Tellurion National Imagery recognition methods based on convolutional neural networks - Google Patents
Tellurion National Imagery recognition methods based on convolutional neural networks Download PDFInfo
- Publication number
- CN108764347A CN108764347A CN201810537533.1A CN201810537533A CN108764347A CN 108764347 A CN108764347 A CN 108764347A CN 201810537533 A CN201810537533 A CN 201810537533A CN 108764347 A CN108764347 A CN 108764347A
- Authority
- CN
- China
- Prior art keywords
- layer
- image
- neural networks
- convolutional neural
- tellurion
- 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
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T9/00—Image coding
- G06T9/005—Statistical coding, e.g. Huffman, run length coding
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)
- Computational Linguistics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Evolutionary Biology (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The tellurion National Imagery recognition methods based on convolutional neural networks that the invention discloses a kind of.First, a plurality of types of images of every country on common teaching globe are acquired by way of data acquisition and data enhancing to construct tellurion National Imagery data set, wherein each country acquires multiple images from different spatial positions and angle shot under different illumination conditions, different focus conditions.Secondly, the every image concentrated to data carries out compression and pretreatment operation.Then, a kind of new convolutional neural networks model is designed in conjunction with the characteristics of classical convolutional neural networks model M obileNet and DenseNet, and new model is trained on collected data set, make the characteristics of image of every country on model learning to tellurion and then is classified to it.The characteristics of identification model of the method for the present invention design combines two kinds of models of MobileNet and DenseNet, has higher recognition accuracy and lower model complexity.
Description
Technical field
The present invention relates to deep learning, field of image recognition more particularly to a kind of tellurions based on convolutional neural networks
National Imagery recognition methods.
Background technology
Traditional image-recognizing method is mostly used greatly for the research of tellurion country identification at present, it is necessary first to pass through people
Work extracts feature, such as utilizes sift, hog, lbp, haar characteristics of image operator extraction characteristics of image, reuses such as svm,
The graders such as adaboost are classified.It is relatively complicated that characteristic procedure is manually extracted in this method, and is difficult to extract more
Advanced feature, therefore recognition effect is unsatisfactory.
Convolutional neural networks need not carry out image cumbersome manual features extraction, but directly input original image
Into network, low-level feature is extracted in lower level using convolution operation, so combine in the higher layers these low-level features from
And obtain higher level abstract characteristics.The mode of this autonomous learning of convolutional neural networks, and it is not necessarily to manual intervention, it can
The stronger advanced features of ability to express are extracted, there is extraordinary recognition effect.
Invention content
The tellurion National Imagery recognition methods based on convolutional neural networks that the present invention provides a kind of, this method combine
The characteristics of two kinds of models of MobileNet and DenseNet, has higher recognition accuracy and lower model complexity.
A kind of tellurion National Imagery recognition methods based on convolutional neural networks provided by the invention, including following step
Suddenly：
Step S101：Enhance two ways by data acquisition and data and builds common teaching globe National Imagery
Data set；
Step S102：Compression and pretreatment operation are carried out to image；
Step S103：In conjunction with the connection side between layers of depth the decomposable asymmetric choice net operation and DenseNet of MobileNet
Formula builds convolutional neural networks structure；In the network architecture, input layer is connected with one layer of Standard convolution layer, and connection is continuous later
12 layer depth decomposable asymmetric choice net convolutional layers, and by Standard convolution layer respectively with the 4th layer, the 8th layer and the 12nd layer depth decomposable asymmetric choice net convolution
Layer is directly connected to, and is then connected one layer of overall situation and be averaged pond layer and one layer of full articulamentum, finally by classification more than softmax
Device exports final classification results；
Step S104：New convolutional neural networks model is trained on collected National Imagery data set, is learned
The feature of every country image is practised it to be identified classification.
Preferably, the step S101 acquires the multiple types image of every country on common teaching globe, packet
Include the image from different spatial and angle shot under different illumination conditions and focus condition；Reuse translation, rotation, contracting
The data enhancement method put and cut is enhanced and is expanded to the image data manually acquired.
Preferably, the step S102 carries out compression and pretreatment operation to image；Using run-length encoding algorithm to image
It is compressed, using luminance transformation, histogram equalization and normalization pre-process image.
Advantageous effect：The convolutional neural networks of the present invention need not carry out cumbersome manual features extraction operation to image and
It is that directly original image is input in network, extracts low-level feature in lower level using convolution operation, and then in higher level
Middle these low-level features of combination simultaneously obtain higher level abstract characteristics, and final recognition result is exported finally by grader.
This method is not necessarily to manual intervention from the whole process for being input to output, and network carries out autonomous learning, and the whole for remaining image has
Information is imitated, extraordinary recognition effect can be obtained.MobileNet is based on depth decomposable asymmetric choice net convolution, and this convolution rolls up standard
Product operation splits into the point convolution of depth convolution sum one 1 × 1, can largely reduce calculating and moulded dimension.
Each layer of input is adjusted to the output of all layers of front by DenseNet, utmostly ensure that the information between each layer passes
It is defeated, it can solve the problems, such as gradient disappearance, efficiently utilize feature, training parameter quantity is greatly reduced.
Description of the drawings
Fig. 1 is that the present invention is based on the tellurion National Imagery recognition methods flow charts of convolutional neural networks.
Fig. 2 is that the present invention is based on the specific implementation illustrations of the tellurion National Imagery recognition methods of convolutional neural networks.
Specific implementation mode
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with the embodiment of the present invention
In attached drawing, technical scheme in the embodiment of the invention is clearly and completely described, it is clear that described embodiment is
A part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, those of ordinary skill in the art
The every other embodiment obtained without creative efforts, shall fall within the protection scope of the present invention.
Shown in Fig. 1, the tellurion National Imagery recognition methods proposed by the present invention based on convolutional neural networks is main to wrap
It includes：
Step 101, image data set of the structure for convolutional neural networks model training.It is adopted by way of manually shooting
The multiple types image for collecting every country on common teaching globe, be included under different illumination conditions and focus condition from
The image of different spatial and angle shot.Again by translation, rotation, the data enhancement method that scales and cut to artificial
The data of acquisition are enhanced and are expanded.
Step 102, compression and pretreatment operation are carried out to image.It is by run-length encoding algorithm that size is very big original
Image unifies the image of 224 × 224 pixel of boil down to.Pretreatment operation is carried out to image again：Luminance transformation is carried out first, is adjusted
Then the brightness of image solves the problems, such as that image local is over-exposed, most to a suitable threshold value by histogram equalization
After be normalized, so that all images is kept unified canonical form.
Step 103, it is designed for the convolutional neural networks model of identification.In conjunction with the depth decomposable asymmetric choice net convolution of MobileNet
The connection type between layers of operation and DenseNet, designs the convolutional Neural net of the new country of tellurion for identification
Network model.Its model structure is as described below：Input layer is connected with one layer of Standard convolution layer, and connecting continuous 12 layer depth later can
Convolutional layer is decomposed, and Standard convolution layer is directly connected with the 4th layer, the 8th layer and the 12nd layer depth decomposable asymmetric choice net convolutional layer respectively
It connects, then connects one layer of overall situation and be averaged pond layer and one layer of full articulamentum, exported finally finally by softmax multi-categorizers
Classification results.
Step 104, new convolutional neural networks model is trained on collected National Imagery data set, is learnt
The feature of every country image achievees the purpose that classify to it.
Shown in Fig. 2, the tellurion National Imagery recognition methods method based on convolutional neural networks of the present embodiment, including：
Step 201, data acquisition, acquire every country on common teaching globe by way of manually shooting
Multiple types image is included under different illumination conditions and focus condition the image from different spatial and angle shot.
Step 202, data enhancing, using translation, rotation, scale and cut mode to the data of acquisition carry out enhancing and
Expand, builds a data volume abundance, the various tellurion National Imagery data set of sample.
Step 203, compression of images compress each image that data are concentrated using run-length encoding algorithm, unified
It is compressed into the image of 224 × 224 pixels.
Step 204, image preprocessing carry out luminance transformation to image first, and the brightness for adjusting image is suitable to one
Then threshold value solves the problems, such as that image local is over-exposed by histogram equalization, is finally normalized, makes to own
Image keeps unified canonical form.
Step 205, in conjunction with the characteristics of two kinds of models of MobileNet and DenseNet, design new convolutional neural networks mould
Type.
Step 206, using back propagation learning algorithm and Adam optimization algorithms on collected National Imagery data set
New model is trained, makes model learning to the validity feature of every country, so as to unknown tellurion National Imagery
It is identified.
Step 207, image recognition carry out any tellurion National Imagery using trained convolutional neural networks model
Identification judges which country it is.
The present embodiment acquires each state on common teaching globe by way of data acquisition and data enhancing
A plurality of types of images of family construct tellurion National Imagery data set, wherein each country acquires in different illumination items
From multiple images of different spatial positions and angle shot under part, different focus conditions, and to every image of data concentration
Compression and pretreatment operation are carried out.In conjunction with classical convolutional neural networks model M obileNet the operation of depth decomposable asymmetric choice net and
The characteristics of connection type between layers of DenseNet, devises a kind of new convolutional neural networks model, and is collecting
Data set on to new model training, make the characteristics of image of every country on model learning to tellurion and then reach
The purpose classified to it.
Finally it should be noted that：The above embodiments are only used to illustrate the technical solution of the present invention., rather than its limitations；To the greatest extent
Present invention has been described in detail with reference to the aforementioned embodiments for pipe, it will be understood by those of ordinary skill in the art that：Its according to
So can with technical scheme described in the above embodiments is modified, either to which part or all technical features into
Row equivalent replacement；And these modifications or replacements, various embodiments of the present invention technology that it does not separate the essence of the corresponding technical solution
The range of scheme.
Claims (3)
1. a kind of tellurion National Imagery recognition methods based on convolutional neural networks, includes the following steps：
Step S101：Enhance two ways by data acquisition and data and builds common teaching globe National Imagery data
Collection；
Step S102：Compression and pretreatment operation are carried out to image；
Step S103：In conjunction with the connection type between layers of depth the decomposable asymmetric choice net operation and DenseNet of MobileNet, structure
Build convolutional neural networks structure；In the network architecture, input layer is connected with one layer of Standard convolution layer, connects continuous 12 layers later
Depth decomposable asymmetric choice net convolutional layer, and Standard convolution layer is carried out with the 4th layer, the 8th layer and the 12nd layer depth decomposable asymmetric choice net convolutional layer respectively
It is directly connected to, then connects one layer of overall situation and be averaged pond layer and one layer of full articulamentum, finally by the output of softmax multi-categorizers
Final classification results；
Step S104：New convolutional neural networks model is trained on collected National Imagery data set, study is each
The feature of a National Imagery is to be identified it classification.
2. the tellurion National Imagery recognition methods based on convolutional neural networks according to claim 1, which is characterized in that institute
Step S101 is stated, the multiple types image of every country on common teaching globe is acquired, is included in different illumination conditions
With under focus condition from the image of different spatial and angle shot；Reuse translation, rotation, the data for scaling and cutting
Enhancement method is enhanced and is expanded to the image data manually acquired.
3. the tellurion National Imagery recognition methods based on convolutional neural networks according to claim 1, which is characterized in that institute
Step S102 is stated, compression and pretreatment operation are carried out to image；Image is compressed using run-length encoding algorithm, uses brightness
Transformation, histogram equalization and normalization pre-process image.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810537533.1A CN108764347B (en) | 2018-05-30 | 2018-05-30 | Globe country image identification method based on convolutional neural network |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810537533.1A CN108764347B (en) | 2018-05-30 | 2018-05-30 | Globe country image identification method based on convolutional neural network |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108764347A true CN108764347A (en) | 2018-11-06 |
| CN108764347B CN108764347B (en) | 2021-09-24 |
Family
ID=64004208
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810537533.1A Expired - Fee Related CN108764347B (en) | 2018-05-30 | 2018-05-30 | Globe country image identification method based on convolutional neural network |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108764347B (en) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583333A (en) * | 2018-11-16 | 2019-04-05 | 中证信用增进股份有限公司 | Image-recognizing method based on water logging method and convolutional neural networks |
| CN109801224A (en) * | 2018-12-04 | 2019-05-24 | 北京奇艺世纪科技有限公司 | A kind of image processing method, device, server and storage medium |
| CN110489584A (en) * | 2019-07-19 | 2019-11-22 | 长沙理工大学 | Image classification method and system based on the MobileNets model intensively connected |
| CN112949796A (en) * | 2021-03-23 | 2021-06-11 | 青岛农业大学 | Hybrid pest and disease identification device and method based on deep learning |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106780543A (en) * | 2017-01-13 | 2017-05-31 | 深圳市唯特视科技有限公司 | A kind of double framework estimating depths and movement technique based on convolutional neural networks |
| US20180137406A1 (en) * | 2016-11-15 | 2018-05-17 | Google Inc. | Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs |
-
2018
- 2018-05-30 CN CN201810537533.1A patent/CN108764347B/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180137406A1 (en) * | 2016-11-15 | 2018-05-17 | Google Inc. | Efficient Convolutional Neural Networks and Techniques to Reduce Associated Computational Costs |
| CN106780543A (en) * | 2017-01-13 | 2017-05-31 | 深圳市唯特视科技有限公司 | A kind of double framework estimating depths and movement technique based on convolutional neural networks |
Non-Patent Citations (2)
| Title |
|---|
| GUOKUN LAI ET AL.: "Learning Depthwise Graph Convolution from Data Manifold", 《HTTPS://ARXIV.ORG/ABS/1710.11577V2》 * |
| 胡挺 等: "面向移动平台的轻量级卷积神经网络架构", 《计算机工程》 * |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109583333A (en) * | 2018-11-16 | 2019-04-05 | 中证信用增进股份有限公司 | Image-recognizing method based on water logging method and convolutional neural networks |
| CN109583333B (en) * | 2018-11-16 | 2020-12-11 | 中证信用增进股份有限公司 | Image identification method based on flooding method and convolutional neural network |
| CN109801224A (en) * | 2018-12-04 | 2019-05-24 | 北京奇艺世纪科技有限公司 | A kind of image processing method, device, server and storage medium |
| CN110489584A (en) * | 2019-07-19 | 2019-11-22 | 长沙理工大学 | Image classification method and system based on the MobileNets model intensively connected |
| CN110489584B (en) * | 2019-07-19 | 2021-10-01 | 长沙理工大学 | Image classification method and system based on dense connection MobileNet model |
| CN112949796A (en) * | 2021-03-23 | 2021-06-11 | 青岛农业大学 | Hybrid pest and disease identification device and method based on deep learning |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108764347B (en) | 2021-09-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108764347A (en) | Tellurion National Imagery recognition methods based on convolutional neural networks | |
| CN106778902B (en) | Dairy cow individual identification method based on deep convolutional neural network | |
| CN109145872B (en) | CFAR and Fast-RCNN fusion-based SAR image ship target detection method | |
| CN105701480B (en) | A kind of Video Semantic Analysis method | |
| CN109886161B (en) | Road traffic identification recognition method based on likelihood clustering and convolutional neural network | |
| CN109241817B (en) | Crop image recognition method shot by unmanned aerial vehicle | |
| CN106815604B (en) | Method for viewing points detecting based on fusion of multi-layer information | |
| CN110479636B (en) | Method and device for automatically sorting tobacco leaves based on neural network | |
| EP3455782A1 (en) | System and method for detecting plant diseases | |
| Zheng et al. | A local-coloring method for night-vision colorization utilizing image analysis and fusion | |
| CN109063737A (en) | Image processing method, device, storage medium and mobile terminal | |
| CN109543697A (en) | A kind of RGBD images steganalysis method based on deep learning | |
| CN107563389A (en) | A kind of corps diseases recognition methods based on deep learning | |
| CN108446699A (en) | Identity card pictorial information identifying system under a kind of complex scene | |
| CN104715259A (en) | Nuclear self-adaptive optimizing and classifying method of X-ray mammary gland images | |
| CN110232670A (en) | A method of the image visual effect enhancing based on low-and high-frequency separation | |
| CN111008647B (en) | Sample extraction and image classification method based on void convolution and residual linkage | |
| CN110310238A (en) | A kind of removing rain based on single image method based on the compression rewards and punishments neural network for reusing raw information | |
| CN103119625A (en) | Video character separation method and device | |
| CN106570503A (en) | Method and system for identifying vehicle body color | |
| CN107516083A (en) | A kind of remote facial image Enhancement Method towards identification | |
| CN113221913A (en) | Agriculture and forestry disease and pest fine-grained identification method and device based on Gaussian probability decision-level fusion | |
| CN113554568A (en) | Unsupervised circulating rain removal network method based on self-supervision constraint and unpaired data | |
| CN116596891A (en) | Wood floor color classification and defect detection method based on semi-supervised multitasking detection | |
| Lin et al. | Deep learning for intelligent substation device infrared fault image analysis |
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 | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20210924 |