CN109543716A - A kind of K line morphology image-recognizing method based on deep learning - Google Patents
A kind of K line morphology image-recognizing method based on deep learning Download PDFInfo
- Publication number
- CN109543716A CN109543716A CN201811238452.8A CN201811238452A CN109543716A CN 109543716 A CN109543716 A CN 109543716A CN 201811238452 A CN201811238452 A CN 201811238452A CN 109543716 A CN109543716 A CN 109543716A
- Authority
- CN
- China
- Prior art keywords
- network
- line morphology
- input
- area
- area generation
- 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
- G06F18/29—Graphical models, e.g. Bayesian networks
Landscapes
- Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Theoretical Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Artificial Intelligence (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of K line morphology image-recognizing method based on deep learning, comprising steps of 1) be input in the neural network containing multilayer convolutional layer using the financial K line morphology image coordinate corresponding with the form containing identification in need as the input of neural network；2) convolutional layer of step 1) is exported into the input as Area generation network, carries out Area generation network training；3) pond is carried out using the output of the Area generation network of step 2) as area-of-interest；4) using the area-of-interest pond result of step 3) as the input of Faster-RCNN detection network；5) location information and recommendation scores for recommending frame are ultimately generated by the Faster-RCNN detection network of step 4).The method overcomes the problem of financial K line morphology feature that existing finance quantization program analyst beyond expression of words is empirically derived, and can learn the financial K line morphology that analyst wants identification and to contain in the identification of the realtime graphic of financial K line morphology feature.
Description
Technical field
The present invention relates to image identification technical fields, and in particular to a kind of K line morphology image recognition based on deep learning
Method.
Background technique
The identification of the form of K line and index is the pith in finance quantization investment analysis.Its identification accuracy is direct
The winning rate for influencing transaction, determines the feasibility of quantization program, however many forms (such as the bottom w form, the maincenter of opinion is twined,
Trend and consolidation form etc.) for financial analyst be understand by thinking it is inexpressible.Due in the identification of financial K line morphology
Ambiguity, it is necessary to allow form identification be detached from the quantization journey of the fixation being rule of thumb transformed based on time series
Sequence allows identification equally to can be suitably used for the inexpressible morphological feature being difficult to fixed routine Unified Expression.Deep learning is theoretical
The research in field is concentrated mainly on algorithm, applies it to that financial investment field is fewer and fewer, also in explore, creation
Stage lacks independent, system Framework of Theoretical Analysis.Wherein, K line morphology image recognition neural network based is in the literature
Almost without occurring, also rare people's research in actual finance quantization exploitation, but K line morphology image, in K line chart
Different shape, K line chart and other data, the combination there are also trading volume, index of trading etc. is that most of investor makes investment
Important evidence, only the information of time series may be not enough to reflect trading situation, need to accomplish the combination of the time and space, this
When K line morphology image identification with regard to particularly important.
Summary of the invention
In view of the deficiencies of the prior art, it is an object of the present invention to provide a kind of K line morphology image based on deep learning
Recognition methods, the method overcome the financial K line that existing finance quantization program analyst beyond expression of words is empirically derived
The problem of morphological feature, can learn the financial K line morphology that analyst wants identification and linear to contain financial K
In the realtime graphic identification of state feature.
The purpose of the present invention can be achieved through the following technical solutions:
A kind of K line morphology image-recognizing method based on deep learning, the described method comprises the following steps:
1) using the financial K line morphology image coordinate corresponding with the form containing identification in need as the defeated of neural network
Enter, is input in the neural network containing multilayer convolutional layer；
2) convolutional layer of step 1) is exported into the input as Area generation network, carries out Area generation network training；
3) pond is carried out using the output of the Area generation network of step 2) as area-of-interest；
4) using the area-of-interest pond result of step 3) as the input of Faster-RCNN detection network；
5) location information and recommendation scores for recommending frame are ultimately generated by the Faster-RCNN detection network of step 4).
Further, the finance K line morphology image be by K line and its index, other finance data indexs it is a kind of or
Image made of multiple combinations.
Further, the detailed process of Area generation network training is carried out described in step 2) are as follows: by the convolution of step 1)
Input of the layer output as Area generation network, uses the window sliding of fixed size on the last layer characteristic pattern of convolution,
The feature of each window output fixed size dimension, each window carry out 9 candidate recurrence frames to return coordinate and classification,
In order to identify an object under different sizes, different size of stroke of window is carried out using to characteristic pattern, is generated
The process of training data is to cover whether ground truth is more than threshold value referring initially to anchor, more than just by current anchor's
Object classification marker is " presence "；If being all not above threshold value, just select the maximum label of a coating ratio to deposit
"；The wherein loss function of Area generation network is defined as:
Wherein subscript i is the number in small lot training sample, piFor the prediction probability of target, the p if target is positive examplei *
=1, otherwise pi *=0, tiFor the vector that four parameters of the frame of prediction are constituted, ti *For the corresponding parameter of ground truth to
Amount；Specific calculation is as follows:
tx=(x-xa)/wa,ty=(y-ya)/ha,
tw=log (w/wa),th=log (h/ha),
Wherein x, y, w and h indicate the center coordinate for the recommendation frame that motion neural network forecast comes out and the height of the recommendation frame
And width, subscript a and subscript * respectively represent anchor framework and ground truth framework, NclsIt is the size of small lot, Nreg
It is the number of anchor, LclsUsing cross entropy, LregUsing Smooth L1, is defined as:
Wherein x is the difference of target value and regressand value.
Further, pond is carried out using the output of the Area generation network of step 2) as area-of-interest described in step 3)
Change, i.e., generates network (Region Proposal Net, RPN) from candidate region and obtain candidate area-of-interest (Roi) column
Table takes all features by convolutional neural networks, carries out subsequent classification and recurrence.
Compared with the prior art, the invention has the following advantages and beneficial effects:
A kind of K line morphology image-recognizing method based on deep learning provided by the invention, will propose the identification of K line morphology
It is raised to image recognition level, the disk data that the truer general stock investor of simulation sees, so as to very straight
The information that research disk in ground is seen is seen, meanwhile, the Successful utilization of this method will overturn the quantization side of current finance K line morphology
Formula, it is no longer necessary to finance K line morphology is explained with code language, also no longer needs the procedure identification finance K line morphology with fixation, and
It is that can only reach using the picture for containing the form and corresponding coordinate position as the input of neural network to financial K line
The study of form and the purpose of automatic identification, and acquistion can be finished classes and leave school to lower False Rate and omission factor in a small amount of training sample
Financial K line morphology.
Detailed description of the invention
Fig. 1 is a kind of flow chart of the K line morphology image-recognizing method based on deep learning of the embodiment of the present invention.
Fig. 2 is the architecture diagram that Faster-RCNN of the embodiment of the present invention detects network.
Fig. 3 is the flow chart of test phase of the embodiment of the present invention.
Specific embodiment
Present invention will now be described in further detail with reference to the embodiments and the accompanying drawings, but embodiments of the present invention are unlimited
In this.
Embodiment:
Present embodiments provide a kind of K line morphology image-recognizing method based on deep learning, the flow chart of the method
As shown in Figure 1, comprising the following steps:
1) using the financial K line morphology image coordinate corresponding with the form containing identification in need as the defeated of neural network
Enter, is input in the neural network containing multilayer convolutional layer；
2) convolutional layer of step 1) is exported into the input as Area generation network, carries out Area generation network training；
3) pond is carried out using the output of the Area generation network of step 2) as area-of-interest；
4) using the area-of-interest pond result of step 3) as the input of Faster-RCNN detection network；
5) being ultimately generated by the Faster-RCNN detection network of step 4) recommends the location information of frame and recommendation scores (to survey
It is as shown in Figure 3 to try process).
The neural network framework that the present embodiment is used --- Faster-RCNN (as shown in Figure 2), essential core are divided into three
Point, respectively candidate region generates network (Region Proposal Net, RPN) part, Area generation network training and joint
Training.
The effect that candidate region generates network is one image of input, exports a collection of rectangle candidate region.It is mentioned including feature
It takes, candidate region (anchor), window classification and position refine.Feature extraction includes several layers convolutional layer, and the present embodiment is direct
Use ResNet50 residual error neural network as convolutional network layer, anchor is the core of RPN network, and effect is to provide a base
Quasi- window size obtains nine kinds of different size of candidate windows according to multiple and Aspect Ratio.It can be found that conv4_x's is last
The part that output is RPN and area-of-interest pond (Roi pooling) is shared, and conv5_x acts on area-of-interest
Characteristic pattern after pond is most followed by an average pond layer, obtains 2048 dimensional features, is respectively used to classification and frame returns.Its
Middle classified part output is target and non-targeted probability, and frame returns four parameters of part output box, and the center including frame is sat
Mark x and y, box wide w and long h.
The effect of Area generation network training is that the image for belonging to label by cost function screening is learnt, and is minimized
The window's position deviation of error in classification and prospect sample.
Joint training includes four steps: 1) individually training RPN network, network parameter are loaded by pre-training model；2) it individually instructs
Practice Faster-RCNN network, using the output candidate region of first step RPN as the input of detection network；3) RPN is trained again,
The parameter of fixed network common portion at this time only updates the parameter of the exclusive part RPN；4) it is finely tuned again by the result of RPN
Faster-RCNN network, the parameter of fixed network common portion only update the parameter of the exclusive part Faster-RCNN.
Identification for K line morphology image, we are divided into the identification of single K line morphology and the knowledge of compound K line morphology
Not.The identification of single K line morphology refers to that, only using K line chart piece as input, identification is special by the form that more K lines are composed
Sign；The identification of compound K line morphology refer to using K line and indicator combination at picture as input, identification by more K lines and
Morphological feature made of indicator combination.The two methods that are identified by of compound K line morphology are labelled and are learnt, one is
K line and index are learnt together as a label, one is K line and index are beaten different labels respectively
It practises.
The above, only the invention patent preferred embodiment, but the scope of protection of the patent of the present invention is not limited to
This, anyone skilled in the art is in the range disclosed in the invention patent, according to the present invention the skill of patent
Art scheme and its patent of invention design are subject to equivalent substitution or change, belong to the scope of protection of the patent of the present invention.
Claims (4)
1. a kind of K line morphology image-recognizing method based on deep learning, which is characterized in that the described method comprises the following steps:
1) defeated using the financial K line morphology image coordinate corresponding with the form containing identification in need as the input of neural network
Enter into the neural network containing multilayer convolutional layer；
2) convolutional layer of step 1) is exported into the input as Area generation network, carries out Area generation network training；
3) pond is carried out using the output of the Area generation network of step 2) as area-of-interest；
4) using the area-of-interest pond result of step 3) as the input of Faster-RCNN detection network；
5) location information and recommendation scores for recommending frame are ultimately generated by the Faster-RCNN detection network of step 4).
2. a kind of K line morphology image-recognizing method based on deep learning according to claim 1, it is characterised in that: institute
Stating financial K line morphology image is the image being composed of K line and its index, other finance data index one or more.
3. a kind of K line morphology image-recognizing method based on deep learning according to claim 1, which is characterized in that step
It is rapid 2) described in carry out the detailed process of Area generation network training are as follows: regard the output of the convolutional layer of step 1) as Area generation net
The input of network, uses the window sliding of fixed size on the last layer characteristic pattern of convolution, and each window exports fixed size
The feature of dimension, each window carry out recurrence coordinate and classification to 9 candidate recurrences frames, in order to can be by an object
Identified under different sizes, carry out different size of stroke of window using to characteristic pattern, generate training data process be referring initially to
Anchor covers whether ground truth is more than threshold value, more than just by the object classification marker of current anchor " to deposit
"；If being all not above threshold value, the maximum label of a coating ratio is selected just to exist "；Wherein Area generation network
Loss function is defined as:
Wherein subscript i is the number in small lot training sample, piFor the prediction probability of target, the p if target is positive examplei *=1,
Otherwise pi *=0, tiFor the vector that four parameters of the frame of prediction are constituted, ti *For the corresponding parameter vector of ground truth；
Specific calculation is as follows:
tx=(x-xa)/wa,ty=(y-ya)/ha,
tw=log (w/wa),th=log (h/ha),
Wherein x, y, w and h indicate the center coordinate for the recommendation frame that motion neural network forecast comes out and the height and width of the recommendation frame
Degree, subscript a and subscript * respectively represent anchor framework and ground truth framework, NclsIt is the size of small lot, NregIt is
The number of anchor, LclsUsing cross entropy, LregUsing Smooth L1, is defined as:
Wherein x is the difference of target value and regressand value.
4. a kind of K line morphology image-recognizing method based on deep learning according to claim 1, it is characterised in that: step
It is rapid 3) described in using the output of the Area generation network of step 2) as area-of-interest carry out pond, i.e., from candidate region generate
Network obtains candidate region of interest domain list, and all features are taken by convolutional neural networks, carry out subsequent classification and
It returns.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811238452.8A CN109543716B (en) | 2018-10-23 | 2018-10-23 | K-line form image identification method based on deep learning |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811238452.8A CN109543716B (en) | 2018-10-23 | 2018-10-23 | K-line form image identification method based on deep learning |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109543716A true CN109543716A (en) | 2019-03-29 |
| CN109543716B CN109543716B (en) | 2021-12-21 |
Family
ID=65844367
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811238452.8A Active CN109543716B (en) | 2018-10-23 | 2018-10-23 | K-line form image identification method based on deep learning |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109543716B (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110136126A (en) * | 2019-05-17 | 2019-08-16 | 东南大学 | Cloth textured flaw detection method based on full convolutional neural networks |
| CN110263843A (en) * | 2019-06-18 | 2019-09-20 | 苏州梧桐汇智软件科技有限责任公司 | Stock K line recognition methods based on deep neural network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150154700A1 (en) * | 2013-12-03 | 2015-06-04 | Michael Steven Hackett | Method of Tracking and Displaying Stocks Information Utilizing Candlestick Charts |
| CN105701450A (en) * | 2015-12-31 | 2016-06-22 | 上海银天下科技有限公司 | K line form identification method and device |
| CN106056244A (en) * | 2016-05-30 | 2016-10-26 | 重庆大学 | Stock price optimization prediction method |
| CN106355500A (en) * | 2016-11-10 | 2017-01-25 | 洪志令 | Stock prediction method based on positive and negative related trend matching |
| CN107832897A (en) * | 2017-11-30 | 2018-03-23 | 浙江工业大学 | A kind of Stock Price Forecasting method based on deep learning |
-
2018
- 2018-10-23 CN CN201811238452.8A patent/CN109543716B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150154700A1 (en) * | 2013-12-03 | 2015-06-04 | Michael Steven Hackett | Method of Tracking and Displaying Stocks Information Utilizing Candlestick Charts |
| CN105701450A (en) * | 2015-12-31 | 2016-06-22 | 上海银天下科技有限公司 | K line form identification method and device |
| CN106056244A (en) * | 2016-05-30 | 2016-10-26 | 重庆大学 | Stock price optimization prediction method |
| CN106355500A (en) * | 2016-11-10 | 2017-01-25 | 洪志令 | Stock prediction method based on positive and negative related trend matching |
| CN107832897A (en) * | 2017-11-30 | 2018-03-23 | 浙江工业大学 | A kind of Stock Price Forecasting method based on deep learning |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110136126A (en) * | 2019-05-17 | 2019-08-16 | 东南大学 | Cloth textured flaw detection method based on full convolutional neural networks |
| CN110263843A (en) * | 2019-06-18 | 2019-09-20 | 苏州梧桐汇智软件科技有限责任公司 | Stock K line recognition methods based on deep neural network |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109543716B (en) | 2021-12-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Qi et al. | StagNet: An attentive semantic RNN for group activity and individual action recognition | |
| CN110428428B (en) | Image semantic segmentation method, electronic equipment and readable storage medium | |
| CN113378632B (en) | Pseudo-label optimization-based unsupervised domain adaptive pedestrian re-identification method | |
| CN112966684B (en) | Cooperative learning character recognition method under attention mechanism | |
| CN110287960A (en) | The detection recognition method of curve text in natural scene image | |
| CN106682633B (en) | The classifying identification method of stool examination image visible component based on machine vision | |
| CN105869178B (en) | A kind of complex target dynamic scene non-formaldehyde finishing method based on the convex optimization of Multiscale combination feature | |
| Fang et al. | Crop leaf disease grade identification based on an improved convolutional neural network | |
| CN109583425A (en) | A kind of integrated recognition methods of the remote sensing images ship based on deep learning | |
| CN106845499A (en) | A kind of image object detection method semantic based on natural language | |
| CN109613006A (en) | A kind of fabric defect detection method based on end-to-end neural network | |
| CN112734775A (en) | Image annotation, image semantic segmentation and model training method and device | |
| CN107633226A (en) | A kind of human action Tracking Recognition method and system | |
| CN108108751A (en) | A kind of scene recognition method based on convolution multiple features and depth random forest | |
| CN108629367A (en) | A method of clothes Attribute Recognition precision is enhanced based on depth network | |
| CN106408030A (en) | SAR image classification method based on middle lamella semantic attribute and convolution neural network | |
| CN110119726A (en) | A kind of vehicle brand multi-angle recognition methods based on YOLOv3 model | |
| CN106023159B (en) | Facilities vegetable leaf portion scab image partition method and system | |
| CN112381030B (en) | Satellite optical remote sensing image target detection method based on feature fusion | |
| CN113657414B (en) | Object identification method | |
| CN110956158A (en) | Pedestrian shielding re-identification method based on teacher and student learning frame | |
| CN112950780A (en) | Intelligent network map generation method and system based on remote sensing image | |
| CN110008900A (en) | A kind of visible remote sensing image candidate target extracting method by region to target | |
| CN116721414A (en) | Medical image cell segmentation and tracking method | |
| CN105631405B (en) | Traffic video intelligent recognition background modeling method based on Multilevel Block |
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 |