CN106203496B - Hydrographic curve extracting method based on machine learning - Google Patents
Hydrographic curve extracting method based on machine learning Download PDFInfo
- Publication number
- CN106203496B CN106203496B CN201610520993.4A CN201610520993A CN106203496B CN 106203496 B CN106203496 B CN 106203496B CN 201610520993 A CN201610520993 A CN 201610520993A CN 106203496 B CN106203496 B CN 106203496B
- Authority
- CN
- China
- Prior art keywords
- machine learning
- image
- extracting method
- curve
- method based
- 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.)
- Expired - Fee Related
Links
- 238000000034 method Methods 0.000 title claims abstract description 55
- 238000010801 machine learning Methods 0.000 title claims abstract description 23
- 238000012549 training Methods 0.000 claims abstract description 49
- 239000000284 extract Substances 0.000 claims abstract description 15
- 238000005070 sampling Methods 0.000 claims abstract description 13
- 238000012545 processing Methods 0.000 claims description 4
- 238000013528 artificial neural network Methods 0.000 claims description 3
- 238000012706 support-vector machine Methods 0.000 claims description 3
- 239000000203 mixture Substances 0.000 claims description 2
- 238000000605 extraction Methods 0.000 description 13
- 230000000694 effects Effects 0.000 description 12
- 238000003709 image segmentation Methods 0.000 description 6
- 230000011218 segmentation Effects 0.000 description 6
- 238000004040 coloring Methods 0.000 description 5
- 238000005286 illumination Methods 0.000 description 4
- 238000011160 research Methods 0.000 description 4
- 238000012952 Resampling Methods 0.000 description 3
- 238000004458 analytical method Methods 0.000 description 2
- 238000013145 classification model Methods 0.000 description 2
- 239000003086 colorant Substances 0.000 description 2
- 238000010191 image analysis Methods 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- LGZXYFMMLRYXLK-UHFFFAOYSA-N mercury(2+);sulfide Chemical compound [S-2].[Hg+2] LGZXYFMMLRYXLK-UHFFFAOYSA-N 0.000 description 2
- 238000000638 solvent extraction Methods 0.000 description 2
- 241001269238 Data Species 0.000 description 1
- 230000003044 adaptive effect Effects 0.000 description 1
- 230000032683 aging Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000007797 corrosion Effects 0.000 description 1
- 238000005260 corrosion Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000003708 edge detection Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 125000001475 halogen functional group Chemical group 0.000 description 1
- 238000013508 migration Methods 0.000 description 1
- 230000005012 migration Effects 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 238000004321 preservation Methods 0.000 description 1
- 238000003672 processing method Methods 0.000 description 1
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 description 1
Classifications
-
- 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
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- 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
- G06V10/56—Extraction of image or video features relating to colour
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
The hydrographic curve extracting method based on machine learning that the invention discloses a kind of.When the method for the present invention is to hydrological data image march line drawing, it selects and extracts certain features in image with resolving ability, and the image pixel of certain area is sampled using the sampling window of changeable scale, as sample data, the iconic element with different characteristic is divided by the method for machine learning, and new training sample is added by incremental mode according to classifying quality;And post-processed using chain code following, effectively eliminate the influence of noise generated after classification.Compared with prior art, the present invention solve the hydrographic curve to be extracted it is thinner when especially prominent aim curve disconnection problem, and the problem is difficult to effectively be solved in original hydrographic curve extracting method.
Description
Technical field
The present invention relates to a kind of extraction sides of the hydrographic curve in image extraction method more particularly to hydrological data image
Method belongs to image segmentation field.
Background technique
In the current information-based and digitized epoch, with universal and storage medium the high speed development of computer, respectively
Kind research field all payes attention to the digitlization of data information further.Due to historical reasons, the fields such as hydrology and water conservancy use net mostly
Trrellis diagram paper hourly observation data.However paper material save it is improper etc. due to will cause damage, pollution the problems such as, be easy pair
The information carried causes damages.And paper material takes up space, and is not easy to the exchange and transmitting of information, is more likely to bury
It is may being hidden in massive information, need the knowledge excavated.It is therefore desirable to digitize to these papery datas.Utilize figure
The mode of picture processing by these information collections and establishes database, will avoid a large amount of manual duplication of labour, also can efficiently precisely
To these information carry out typing, have stronger practical application value.
Papery hydrological data is usually the hydrographic curve for the bluish violet drawn on the coordinate net ruled paper of Chinese red, in number
During change, just need to obtain each intersection point of hydrographic curve and grid lines when obtaining the information in drawing, as each moment
Observation.This is crossed range request and is split to image, relates to grid lines segmentation and divides with hydrographic curve.
Image segmentation is exactly that image is divided into several regions specific, with unique properties according to certain standard
And therefrom extract the technology and process of interesting target.Image segmentation is the key precondition of image analysis, the quality of segmentation
Superiority and inferiority is largely fixed the effect of subsequent image analysis.Image segmentation can be divided into gray level image segmentation and color image point
It cuts.Compared with gray level image, color image not only includes luminance information, has further included various colors information, partitioning scheme is more
For multiplicity, but corresponding segmentation difficulty is also bigger.So far, researcher both domestic and external is in color images field
A large amount of research is had been carried out, and proposes many partitioning algorithms, and for the segmentation strategy of specific image, mainly includes base
In histogram thresholding method, based on region method, edge detection method, Segmentation by Fuzzy Clustering method and neural network etc..
In research before, what is generallyd use to the segmentation of hydrological data image is the threshold value based on color histogram point
Analysis method, it is also considered that gradient information merges use with colouring information.The adaptive completion of such method energy is under normal circumstances
Image segmentation, and can be reduced camera shooting be uneven illumination influence.But discovery is extracted and is obtained when actually using such method
The hydrographic curve obtained is easy to produce broken string in some special cases, and it is very serious usually to break, it is difficult to expanding method solution
Certainly.
Summary of the invention
Goal of the invention: for the digitlization of papery hydrological data, providing a kind of hydrographic curve extracting method, can be accurate
Hydrographic curve therein is extracted, curve disconnection problem is effectively evaded.
Hydrographic curve extracting method of the invention, related hydrological data image is by shooting papery hydrological data
It arrives.
The present invention is specifically solved the above problems using following technical scheme.
A kind of hydrographic curve extracting method based on machine learning, comprising the following steps:
Step A, the scale of selected sampling window and the target signature that need to be sampled, and representative training is acquired accordingly
Sample set;The scale of the window is scalable.The selection of window size determines the data volume for classification, also directly affects
The scale of calculation amount.
Step B, using the method for machine learning, training generates classification prediction model from training sample;
Step C, to each pixel in image to be processed, target signature is collected as to be sorted according to sampling window
Sample is classified using the classification prediction model that step B training obtains;
Step D, judge whether the classification results of each pixel in image to be processed are preferable, so that curve extracts complete and do not have
There is the apparent region of other classification errors.If so, entering step F;Otherwise, E is entered step;
Step E, representative sample point is chosen from curve broken string region and the apparent region of classification error, to it
It is added in training sample set after sampling, and repeats step B-D;
Step F, to treated, image is post-processed, and removes noise that may be present.
Preferably, training sample set collected should include at least " hydrographic curve ", " grid lines ", " other in step A
The other sample of three type of background ".
Preferably, post processing of image method used in step F is combined using chain code following with expansion process.Wherein chain
Before code tracking hydrographic curve, first grid lines is tracked, determines that grid lines is corresponding and makees graph region;This step can mitigate with
Processing intensity when track hydrographic curve.
Preferably, the connected domain that size is less than specific threshold is regarded as into noise after chain code following, and eliminates image.It should
Threshold value value is 10000.
Preferably, the size of connected domain is indicated using the area of the minimum circumscribed rectangle of the connected domain.
Compared with prior art, the invention has the following advantages:
One, the disconnection problem being easy to produce when the present invention preferably can solve to extract filament;
Two, the present invention is based on the classification to sample mode, as long as selecting sufficient training sample, do not need to consider illumination
The problems such as influence;
Three, it using off-line learning, does not need to resurvey sample training to each image.
Detailed description of the invention
Fig. 1, Fig. 2 and Fig. 3 are the hydrological data image that three width are shot.
Fig. 4 a and Fig. 4 b are the result that existing method extracts hydrographic curve to Fig. 1 and Fig. 2.
Fig. 5 a and Fig. 5 b are result of the different phase to Fig. 2 classification prediction of train classification models in the method for the present invention.
Fig. 6 a and Fig. 6 b are result of the different phase to Fig. 3 classification prediction of train classification models in the method for the present invention.
Fig. 7 a and Fig. 7 b are the result that the method for the present invention extracts hydrographic curve to Fig. 2 and Fig. 3.
Fig. 8 is flow chart of the invention.
Specific embodiment
Technical solution of the present invention is described in detail with reference to the accompanying drawing: Fig. 1 and Fig. 2 respectively illustrates the two width hydrology
Data image carries out digitized key to it and is to hydrographic curve therein (bluish violet) and coordinate net ruling (Chinese red)
Extraction.As can be seen from the figure since the holding time is long, preservation condition is not ideal enough, in addition to damaged, papery of grinding away in image
Outside aging, there is also coloring halo of many colours dye, fade the problems such as, or even on a drawing not expect field color deep or light different.And
The colouring information of illumination effect when due to shooting, partial region weakens or loses original feature, so that the problem of extraction becomes
It obtains increasingly complex.
Threshold Analysis method based on color histogram used by research before, has merged gradient information and color is believed
Breath, can preferably solve the above problem, to processing result such as Fig. 4 a and Fig. 4 b of Fig. 1 and Fig. 2.It can see image
Extraction result it is preferable, the identification and extraction to object are completed in big enable, but the hydrographic curve extracted sometimes can have broken string,
Belong to special case, at the broken string in Fig. 4 b.To careful observation is carried out at these broken strings, it is found that the main reason for breaking is Tu Zhongshui
Literary curve is too thin, (usually curve with being overlapped when grid lines less parallel) curve so that when having a large amount of be overlapped with grid lines
Mass colour do not cover the color of grid lines thoroughly, the color finally shown by for both superposition.The superposition of this color is led
The migration for having caused colouring information, the pixel at this no longer meet universal hydrographic curve color characteristic, when extracting curve at this
It is easy for generating broken string.What is more, when a certain section of curve with grid lines less parallel, it will generates a large amount of coincidence area
Domain, often broken string is serious at this, it is difficult to pass through the methods of expansion and carry out completion.Due to no longer meeting threshold trait, original processing
Method is no longer applicable in, and noise spot substantial increase will be will lead to by adjusting threshold value, and image zooming-out is unstable.Then it is considered as new
Thinking extracts hydrographic curve.
Result after being migrated in view of the colouring information of overlapping region not with curvilinear characteristic or grid lines feature etc.
Existing information is identical, and target is that the characteristic information after changing these is still identified as aim curve.The invention proposes be based on
The hydrographic curve extracting method of machine learning obtains certain amount by carrying out the sampling of multiple features fusion to pixel in image
Tape label training sample, and using the method for machine learning training obtain classification prediction model.It can be to figure using model
Middle pixel carries out classification prediction, and the pixel extraction for being predicted as hydrographic curve is come out.And by classification error pixel region
Resampling is carried out, training sample set can be made to close more complete, training generates more healthy and strong classification prediction model.In order not to make
The influence of noise curve generated in treatment process extracts as a result, will also be using the method for chain code following to the hydrographic curve extracted
Image is post-processed.
Specifically, the present invention comprises the steps of:
Step A, window size and sampling feature combination are determined, training sample is acquired;
The feature that the scale of sampling window and needs sample first is determined when acquiring training sample.Sampling window is with current
Pixel is window center point, combines the information that other pixels of current pixel point are surrounded in window, i.e. local message.By
The information having in single pixel is fairly limited, can be with when other local messages around pixel are accounted for together
So that sample dimension is higher, it is more advantageous to careful classification.A variety of scales, such as 3*3,5*5,7*7, scale can be used in window
More big, the information for including in sample is more, facilitate it is more careful accurately classify, but calculate the time it is also longer;Scale is smaller then
Information in sample is fewer, corresponding, and it is also shorter to calculate the time.Specific scale selection should by the demand of practical application and
It is fixed.
Used local message is also required to select when on the other hand, to window sample, these information are by some spies
Value indicative composition, including color characteristic (RGB, Lab or HSI etc.), Gradient Features, textural characteristics (LBP or Gabor etc.), SIFT are special
Sign etc..The selection of specific feature combination is considered as influence of these features to hydrographic curve extraction effect, chooses wherein appropriate
Several groups of features be combined.Excessive feature selecting can bring the redundancy of information and the load of calculating, and very few feature then may be used
Can classifying quality be declined.Feature combination choice relation classification effect and calculating when load.
When carrying out above-mentioned sampling to pixel, should according to the window size arranged in advance, feature combination is by certain
Sequence obtains each characteristic value and is integrated into orderly sample vector.Acquire color characteristic when, should according to from left to right, on to
Under sequence successively each pixel in window is acquired.In addition, in addition to add one-dimensional characteristic to training sample, as working as
The class label of preceding training sample.
When acquiring training sample, the selection of sampled point is particularly important, it should however be noted that 1, sampling when should take into account each difference
Target category, should each obtain enough sample points;2, in the identical pixel of target category, Yao Jinliang, which is covered, to be had
The pixel of different local features;3, it should be selected in the similar pixel of and local feature identical in classification several with typicalness
Pixel is sampled.Wherein, training sample set collected should include at least " hydrographic curve ", " grid lines ", " other back
The other sample of three type of scape ".
Step B, classification prediction model is generated using machine learning method training;
Machine learning method includes the study and unsupervised study of supervision.Since class object is bright in current problem
Really, it only wants to extract hydrographic curve, therefore uses the learning method for having supervision, utilize the training sample with class label of acquisition
Obtain classification prediction model.Such learning method includes decision tree, Bayes classifier, k nearest neighbor, BP neural network, perceptron
And support vector machines etc..Different machine learning methods has the characteristics that different, should select according to actual needs.Engineering
The choice relation of learning method curve extract effect and efficiency.
Different hydrological data image discoveries is analyzed, the color and structure feature between each image are much like, from feature
The angle in space can also substantially be classified with same group of interface on feature space, be extracted even if being different image.
Then the method for determining to use off-line learning, target is the classification prediction model that training generates an excellent effect, for institute
Pixel in image to be handled is classified, is extracted, rather than generates a model for the training of each image.
Step C, classification prediction, and supplementary training sample set are carried out to image to be processed;
According to the mode arranged in above-mentioned steps A, feature samples corresponding to the extraction of image pixel by pixel to be processed,
And prediction classification is carried out via classification prediction model obtained as input.Extract the picture for being wherein predicted as " hydrographic curve "
Element, the result extracted as this curve.
If curve extracts, result is complete, and effect is satisfactory, can enter next step;But it generally can not be immediately obtained
It is satisfactory to extract as a result, the curve extracted is often relatively rough and will appear broken string, it can also extract many noise spots.Solution
Certainly method is constantly to obtain new sample with incremental mode training sample set is added, so that training is improved further and is good for
Strong classification prediction model.Previous prediction classification results both are from as the new training sample of increment every time, i.e., are therefrom looked for
Out at curve broken string and other sort out the biggish region of error rate, with the picture of typicalness on local feature in selection region
Vegetarian refreshments is sampled.The method is intended to carry out past error prediction study and makes up, the resampling near classification interface, by
Omission and vacancy on this compensation sample space, obtain more complete training set, to obtain more fine accurately boundary
Face and more healthy and strong classification prediction model.
Using the training sample set re -training classification prediction model after addition increment and the image is divided again
Class prediction, if curve extraction effect is satisfactory, the image is by currently processed, into next step;Otherwise it repeats above-mentioned
The process of increment addition training sample.
The training process of classification prediction model be not it is stranghtforward, need repeatedly to modify and increase new sample and instructed again
Practice;Simultaneously nor in some entirely different " training stage " middle completion, but the classifying quality in certain image is bad
When just start " retraining ";I.e., it has no one explicit and limited " training stage ".In addition, to training sample set
Increase and need manually-operated intervention, by the selected sampled point newly increased by hand.But due to the initial accumulated rank of training sample
Duan Tongchang can be quickly completed, and obtain the preferable model of effect, and only just may require that the existing mould of retraining under special circumstances
Type, actually manually-operated workload very little.
Step D, chain code following is post-processed.
Since original shooting image often has much noise point, the curve of above-mentioned steps, which extracts still to exist in result, to be permitted
It is difficult to the ambient noise eliminated, they are that misalignment leads to classification error and introduces mostly more.Due to the main needle of the present invention
Extraction to thinner hydrographic curve image, and if removing noise using common corrosion or filtering method, often
Curve becomes very thin and even seriously breaks again, can not obtain satisfactory result.Ideal target is to remove noise spot
It removes, and any variation does not occur for hydrographic curve, the mode of chain code following can be taken to be post-processed in order to reach this target.
Chain code following method can track in a manner of chain code and record the information of each connected domain in figure, as each pixel
Its affiliated connected domain is marked, and records the size and bezel locations of each connected domain.The size of connected domain is not with wherein pixel
Subject to number, and it is subject to the area of its minimum circumscribed rectangle.
Result images after extracting to classification, tracking are wherein predicted as the pixel of " aim curve ", obtain its connected domain
Information;That is, image to be carried out to the binaryzation of " aim curve/non-targeted curve ", and above-mentioned chain code following is carried out to it.Its with
Track result will include true hydrographic curve target area and noise spot region, and the connected domain where the former is usually very big, and
The latter is in contrast smaller.Then specific threshold can be arranged to the size of connected domain, to exclude those lesser, noise institutes
Connected domain.
It is curve in resulting image after category of model extracts in order to prevent so do not obtain maximum connected domain directly
There are still broken strings.It is this broken string it is often subtleer, be easy solve, can after removing noise spot in addition to aim curve into
Row expansive working several times.
In order to improve treatment effect, the advanced row of Cheng Qian can also be crossed once to image copy at above-mentioned " curve tracing "
" grid lines tracking " to determine grid lines region, and carries out above-mentioned " curve tracing " in the area.That is, to image pair
The binaryzation of this progress " grid lines/non-grid line ", and chain code following is carried out to it, the bezel locations in largest connected domain are obtained,
Border line as grid lines.The purpose of the operation, which is to remove outside grid lines, extracts unrelated all pixels with curve, reduces
Complexity when curve tracing.
In order to verify effect of the invention, chooses several colored hydrology source map pictures and tested, it is carried out above-mentioned
Classification extraction process.Arranging selected window size size is 7*7, and the feature group of institute's sampled point is combined into the RGB of each pixel
Color value, HSI color value and Lab color value 9 characteristic values in total;That is, the dimension of related sample is 7*7*9=
441.And arranging selected machine learning method is support vector machines.By taking Fig. 2 as an example, training sample set is combined into when initial
Sky carries out initial samples to Fig. 2 first, obtains training after enough training samples and generates SVM classifier, and for Fig. 2 into
Row classification prediction, result such as Fig. 5 a, wherein black color dots are the point that prediction result is " hydrographic curve ", and Grey Point is " grid
Line ".As it can be seen that at this time to the classifying quality of Fig. 2 and unsatisfactory, it is biggish there are at many broken strings, especially occurring two
Broken position.Several times to resampling at these broken strings, the SVM classifier effect generated after a few wheel re -trainings is mentioned
Height, to being resolved at the classification results interrupt line of Fig. 2, such as Fig. 5 b.
Classification prediction is carried out to Fig. 3 using the trial of current SVM classifier, as a result such as Fig. 6 a, the curve obtained at this time is not deposited
In broken string phenomenon, but there are also too many noise, classifying quality is not good.Again to these noise point samplings, a new round is trained
Classifier, again to result such as Fig. 6 b of Fig. 3 classification.Classifying quality is preferable at this time, it is believed that current class device can have been completed pair
The classificating requirement of this two figures.If it is desired, also aforesaid operations can be being repeated to other images.
The classification results of Fig. 5 b and Fig. 6 b are continued to post-process, remove unwanted " grid lines " Grey Point, are utilized
The mode cancelling noise point of chain code following, and expansive working several times in addition is carried out to curve, it obtains curve and extracts result such as Fig. 7 a
With Fig. 7 b.As it can be seen that extraction of the present invention to completing to hydrographic curve thinner in Fig. 2 and Fig. 3.
Hydrographic curve extracting method based on machine learning method of the invention, based on the classification to sample mode, as long as
Sufficient training sample is selected, the influence for the problems such as considering illumination is not needed;Using off-line learning, do not need to each figure
As resurveying sample training;Training sample is selected and added with incremental mode, adapts to the new classificating requirement constantly to arrive.
The disconnection problem that the present invention is easy to produce when preferably can solve to extract filament has good researching value.
Claims (9)
1. a kind of hydrographic curve extracting method based on machine learning, which comprises the following steps:
Step A, the scale of selected sampling window and the target signature that need to be sampled, and representative training sample is acquired accordingly
Set;
Step B, using machine learning method, training generates classification prediction model from training sample;
Step C, to each pixel in image to be processed, target signature is collected as sample to be sorted according to sampling window
This, is classified using the classification prediction model;
Step D, judge whether the classification results of each pixel in image to be processed reach expected, curve whether extract it is complete and
With the presence or absence of the apparent region of other classification errors;If classification results reach expected, F is entered step;Otherwise, E is entered step;
Step E, representative sample point is chosen from curve broken string region and the apparent region of classification error, it is sampled
After be added in training sample set, and repeat step B-D;
Step F, to treated, image is post-processed.
2. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that in step A, the window
The scale of mouth is scalable.
3. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that in step A, to part
Using the combination of various types of feature, the feature includes color characteristic, Gradient Features, textural characteristics for the selection of feature
And SIFT feature.
4. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that in step B, the machine
Device learning method includes support vector machines, neural network method and combinations thereof.
5. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that in step C, traverse institute
There is the pixel that current sampling window can be used to extract feature, corresponding each local feature value is obtained according to window and composition waits for point
Class sample vector.
6. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that step E is further are as follows:
The sample point of selection should be added in original training set as training sample with incremental form and re -training generates prediction mould
Type, to carry out directive adjustment to original model.
7. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that in step F, after described
Processing mode is that chain code following is combined with expansion process.
8. the hydrographic curve extracting method based on machine learning as described in claim 1, which is characterized in that also need to add sample
Classification " grid lines ", to being positioned as graph region on figure when to handle image.
9. the hydrographic curve extracting method based on machine learning as claimed in claim 7, which is characterized in that the chain code following is used
The connected domain present in tracking image, calculates, records the size of these connected domains, thus excludes the ruler being made of noise
Very little lesser connected domain.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610520993.4A CN106203496B (en) | 2016-07-01 | 2016-07-01 | Hydrographic curve extracting method based on machine learning |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201610520993.4A CN106203496B (en) | 2016-07-01 | 2016-07-01 | Hydrographic curve extracting method based on machine learning |
Publications (2)
Publication Number | Publication Date |
---|---|
CN106203496A CN106203496A (en) | 2016-12-07 |
CN106203496B true CN106203496B (en) | 2019-07-12 |
Family
ID=57466160
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201610520993.4A Expired - Fee Related CN106203496B (en) | 2016-07-01 | 2016-07-01 | Hydrographic curve extracting method based on machine learning |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN106203496B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109753971B (en) * | 2017-11-06 | 2023-04-28 | 阿里巴巴集团控股有限公司 | Correction method and device for distorted text lines, character recognition method and device |
CN108090686B (en) * | 2017-12-29 | 2022-01-25 | 北京大学 | Medical event risk assessment analysis method and system |
CN111539587B (en) * | 2020-03-06 | 2023-11-24 | 武汉极善信息技术有限公司 | Hydrologic forecasting method |
CN113359216A (en) * | 2021-06-03 | 2021-09-07 | 山东捷瑞数字科技股份有限公司 | Method, system and storage medium for identification of recorded data of tipping-bucket rain gauge |
CN114648570B (en) * | 2022-03-28 | 2024-03-26 | 杭州电子科技大学 | Curve extraction method for differentiated background grid based on deep learning |
CN115409825B (en) * | 2022-09-06 | 2023-09-12 | 重庆众仁科技有限公司 | Temperature, humidity and pressure trace identification method based on image identification |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103020971A (en) * | 2012-12-28 | 2013-04-03 | 青岛爱维互动信息技术有限公司 | Method for automatically segmenting target objects from images |
CN103971367A (en) * | 2014-04-28 | 2014-08-06 | 河海大学 | Hydrologic data image segmenting method |
CN104573731A (en) * | 2015-02-06 | 2015-04-29 | 厦门大学 | Rapid target detection method based on convolutional neural network |
US9053551B2 (en) * | 2012-05-23 | 2015-06-09 | International Business Machines Corporation | Vessel identification using shape and motion mapping for coronary angiogram sequences |
CN105184265A (en) * | 2015-09-14 | 2015-12-23 | 哈尔滨工业大学 | Self-learning-based handwritten form numeric character string rapid recognition method |
Family Cites Families (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8934685B2 (en) * | 2010-09-21 | 2015-01-13 | General Electric Company | System and method for analyzing and visualizing local clinical features |
-
2016
- 2016-07-01 CN CN201610520993.4A patent/CN106203496B/en not_active Expired - Fee Related
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US9053551B2 (en) * | 2012-05-23 | 2015-06-09 | International Business Machines Corporation | Vessel identification using shape and motion mapping for coronary angiogram sequences |
CN103020971A (en) * | 2012-12-28 | 2013-04-03 | 青岛爱维互动信息技术有限公司 | Method for automatically segmenting target objects from images |
CN103971367A (en) * | 2014-04-28 | 2014-08-06 | 河海大学 | Hydrologic data image segmenting method |
CN104573731A (en) * | 2015-02-06 | 2015-04-29 | 厦门大学 | Rapid target detection method based on convolutional neural network |
CN105184265A (en) * | 2015-09-14 | 2015-12-23 | 哈尔滨工业大学 | Self-learning-based handwritten form numeric character string rapid recognition method |
Also Published As
Publication number | Publication date |
---|---|
CN106203496A (en) | 2016-12-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN106203496B (en) | Hydrographic curve extracting method based on machine learning | |
Zhao et al. | Cloud shape classification system based on multi-channel cnn and improved fdm | |
Ping et al. | A deep learning approach for street pothole detection | |
CN109902806A (en) | Method is determined based on the noise image object boundary frame of convolutional neural networks | |
CN104408449B (en) | Intelligent mobile terminal scene literal processing method | |
CN105389581B (en) | A kind of rice germ plumule integrity degree intelligent identifying system and its recognition methods | |
CN108805018A (en) | Road signs detection recognition method, electronic equipment, storage medium and system | |
CN110991439B (en) | Handwriting character extraction method based on pixel-level multi-feature joint classification | |
CN110458821A (en) | A kind of sperm morphology analysis method based on deep neural network model | |
CN112686902B (en) | Two-stage calculation method for brain glioma identification and segmentation in nuclear magnetic resonance image | |
CN106803248B (en) | Fuzzy license plate image blur evaluation method | |
CN109598681B (en) | No-reference quality evaluation method for image after repairing of symmetrical Thangka | |
CN105138975B (en) | A kind of area of skin color of human body dividing method based on degree of depth conviction network | |
CN109766823A (en) | A kind of high-definition remote sensing ship detecting method based on deep layer convolutional neural networks | |
CN107016680B (en) | A kind of pest image background minimizing technology based on conspicuousness detection | |
CN113158977B (en) | Image character editing method for improving FANnet generation network | |
CN109086772A (en) | A kind of recognition methods and system distorting adhesion character picture validation code | |
CN110533068A (en) | A kind of image object recognition methods based on classification convolutional neural networks | |
CN110956167A (en) | Classification discrimination and strengthened separation method based on positioning characters | |
Tang et al. | Leaf extraction from complicated background | |
CN108733749A (en) | A kind of image search method based on sketch | |
CN109815957A (en) | A kind of character recognition method based on color image under complex background | |
CN109741351A (en) | A kind of classification responsive type edge detection method based on deep learning | |
CN103778431B (en) | Medical image characteristic extracting and identifying method based on two-directional grid complexity measurement | |
CN116311387B (en) | Cross-modal pedestrian re-identification method based on feature intersection |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
C06 | Publication | ||
PB01 | Publication | ||
C10 | Entry into 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 |
Granted publication date: 20190712 |