CN109447993A - A kind of sea ice image partition method based on mixing true and false sample strategy - Google Patents
A kind of sea ice image partition method based on mixing true and false sample strategy Download PDFInfo
- Publication number
- CN109447993A CN109447993A CN201811248533.6A CN201811248533A CN109447993A CN 109447993 A CN109447993 A CN 109447993A CN 201811248533 A CN201811248533 A CN 201811248533A CN 109447993 A CN109447993 A CN 109447993A
- Authority
- CN
- China
- Prior art keywords
- sea ice
- true
- sample
- image
- sea
- 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
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/10—Image acquisition modality
- G06T2207/10024—Color image
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- Geometry (AREA)
- General Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
The present invention relates to a kind of sea ice image partition methods based on mixing true and false sample strategy, comprising the following steps: the sea ice landform of emulation is generated using diamond shape-square fractal method；The corresponding colour band information of practical sea ice is analyzed, color assignment is carried out to the sea ice landform of emulation, classification thresholds are set according to simulation sample height value, generate true value figure；It is filtered pretreatment to the sea ice of emulation, and forms training set with the true sea ice sample for the tape label being collected into；It is trained by the full convolutional network of symmetrical coder-decoder, obtains the segmentation network of sea ice image；Image segmentation is carried out to sea ice image to be split.The present invention generates the artificially generated terrain of sea ice using diamond shape-square fractal method, sets to obtain the true value figure of tape label, supplementary training collection sample by threshold value；It effectively improves Small Sample Size to plunge into the commercial sea the segmentation effect of ice atlas picture, is conducive to the practical application of Sea Ice Remote Sensing image Segmentation Technology.
Description
Technical field
It is especially a kind of based on mixing true and false sample (Mixing Real and the present invention relates to a kind of image partition method
Fake Samples, MRFS) strategy sea ice image partition method, belong to Remote Sensing Image Segmentation research field.
Background technique
Gradually accelerated by the sea ice ablation speed of influenced by global warming, Arctic, coverage area constantly reduces, while sea
Ice is one of main Oceanic disasters.Therefore, the sea ice conditions of key area are monitored, are wanted to marine environment correlation is grasped
Prime information, the reliable data information of accumulation, maintenance channel safe are of great significance.To the semantic segmentation skill of Sea Ice Remote Sensing image
Art research is conducive to carry out the open-minded of navigation channel in the sea ice of ablation, and semantic segmentation generallys use depth convolutional network and comes to a large amount of
Sea ice image and corresponding true value figure are trained.
It extracts sea ice image information thus to obtain the ice-covered spatial and temporal distributions data in sea, is sea ice and ecological environment research
In a Basic Problems, remote sensing satellite technology flourish, provide technical support to solve this problem.However by
The weather conditions of polar region restrict, and getting a large amount of Sea Ice Remote Sensing image, there are certain difficulties, while making true value figure needs and disappearing
Consume excessive manpower and material resources.Under this contradiction, study a kind of for solving the Training strategy of the segmentation problem of small sample sea ice
Be conducive to various important remote sensing applications such as sea ice conditions and ship navigation channel setting.
In conclusion the present invention proposes to carry out sea ice image segmentation using the strategy of mixing true and false sample composition training set,
Sea ice simulation sample is wherein generated by diamond shape-square fractal method.
Summary of the invention
For the above-mentioned prior art, the technical problem to be solved in the present invention is to provide one kind can more effectively promote small sample
In the case of sea ice image segmentation precision based on mixing true and false sample strategy sea ice image partition method.
In order to solve the above technical problems, a kind of sea ice image partition method based on mixing true and false sample strategy of the present invention,
The following steps are included:
Step (1): the sea ice landform of emulation is generated using diamond shape-square fractal method；
Step (2): the corresponding colour band information of the practical sea ice of analysis carries out color assignment to the sea ice landform of emulation, according to
Simulation sample height value sets classification thresholds, generates true value figure；
Step (3): pretreatment, and the true sea ice sample with the tape label being collected into are filtered to the sea ice of emulation
Form training set；
Step (4): it is trained by the full convolutional network of symmetrical coder-decoder, obtains the segmentation of sea ice image
Network；
Step (5): image segmentation is carried out to sea ice image to be split.
The invention also includes:
Fractal method generates the height at its square center in step (1) are as follows:
Wherein, HCIt is the height value that square base plane center goes out；H00、H01、H11、H10Respectively four angle points of base plane
The height value at place；D0For the stochastic variable for meeting Gaussian Profile, D0Probability density function be f (x), meet:
Wherein, [0,100] mean μ ∈, variances sigma ∈ [0,1].
The invention has the advantages that:
The artificially generated terrain that sea ice is generated using diamond shape-square fractal method, sets to obtain tape label by threshold value
True value figure supplements training set sample；
Small Sample Size is effectively improved by the Training strategy of mixed reality sample and simulation sample to plunge into the commercial sea ice atlas picture
Segmentation effect, be conducive to the practical application of Sea Ice Remote Sensing image Segmentation Technology.
Detailed description of the invention
Fig. 1 uses the flow chart of the sea ice image partition method of mixing true and false sample strategy；
Fig. 2 mixes true and false sample policing algorithm schematic diagram；
Specific embodiment
Illustrate present embodiment with reference to the accompanying drawing, as shown in Figure 1, the step of present embodiment is as follows:
Step 1: generating the sea ice landform matrix of emulation using diamond shape-square fractal method, it is flat to calculate square base
Height at the center of face:
Wherein, HCIt is the height value that square base plane center goes out；H00、H01、H11、H10Respectively four angle points of base plane
The height value at place；D0It is the random offset of center, D0For the stochastic variable for meeting Gaussian Profile, probability density function
It for f (x), can set under normal conditions mean μ ∈ [0,100], variances sigma ∈ [0,1].
In three-dimensional space, to generate with fractal Brown motion (Fractional Brownian Motion, FBM) characteristic
Thick ice landforms, we can generate the third seat of point on a two-dimentional base plane with the thought of midpoint displacement offset
Mark, to complete the modeling process of the thick ice in expectation.Square base plane originally is divided into four smaller squares,
And the height value of four corner points of each square is equally known.That is, in original four own predominant heights
On the basis of value, by operation, and five height values is increased newly, not only base plane have been refined in this way, while also increasing
The density of three-dimensional point element.Above procedure is repeated to each small square, so that it may smaller square net is marked off,
The three-dimensional point element for more having height value is generated simultaneously.In this way, by successive ignition, when the density of three-dimensional point element reaches
It is required that when, we can be obtained by the data model of the geometric object with FBM characteristic.The height value that thick ice is arranged is greater than 0, sea
The height value of water generates the thick ice and seawater landform with certain height value less than 0；
Step 2: the corresponding colour band information of analysis sea ice, studies the corresponding variation range of color of thick ice and thin ice, to imitative
True sample carries out color assignment, thick ice and seawater rgb matrix is respectively set to 15 and 10 random variation range, thick ice color is
(220,220,240) -15, color intensity of sea water are (15,15,20) -10.Classification thresholds are set as 0 according to simulation sample height value, it will
Simulation sample label is thick ice and two class of seawater, generates true value figure；
Step 3: being filtered the true sea ice sample group as pretreatment, with the tape label being collected into simulation sample
At training set: having used four kinds of median filtering, adaptive median filter, gaussian filtering and Wiener filtering filters, simulation sample
Ratio with authentic specimen composition training set is 1:2, and simulation sample 50 is opened, and authentic specimen 100 is opened；
Step 4: being trained by the full convolutional network of symmetrical coder-decoder, the segmentation net of sea ice image is obtained
Network: being trained mixing sample using SegNet web results, passes through the test of true sea ice image and carries out analysis and assessment
Obtain the optimal segmentation network of effect.
Step 5: carrying out image segmentation: the optimum segmentation network obtained using above-mentioned steps to sea ice image to be split
Sea ice image with segmentation is split.
It is as shown in Figure 2 to mix true and false sample policing algorithm schematic diagram.
It is above-mentioned for the present invention it is special for embodiment, be not intended to limit the invention.It is provided by the invention to be based on the mixing true and false
The sea ice image partition method of sample strategy is equally applicable to other small sample sea ice image segmentation networks.This hair is not being departed from
In bright spirit and scope, a little adjustment and optimization can be done, is subject to protection scope of the present invention with claim.
The specific embodiment of the invention further include: the present invention uses following technical solutions:
Step 1: generating the sea ice landform matrix of emulation using diamond shape-square fractal method, it is flat to calculate square base
Height at the center of face:
Wherein, HCIt is the height value that square base plane center goes out；H00、H01、H11、H10Respectively four angle points of base plane
The height value at place；D0The random offset that center is pointed out, be a mean value be 0, variance δ2Gaussian random variable；
Step 2: the corresponding colour band information of analysis sea ice, carries out color assignment to simulation sample, generating has certain elevation
The thick ice and seawater landform of value set classification thresholds according to simulation sample height value, generate true value figure and to its rgb matrix: will
15 and 10 random variation range is respectively set in thick ice and seawater, and it is 0 that setting classification thresholds, which are elevation, obtains the true value of tape label
Figure；
Step 3: being filtered the true sea ice sample group as pretreatment, with the tape label being collected into simulation sample
At training set: having used four kinds of median filtering, adaptive median filter, gaussian filtering and Wiener filtering filters, simulation sample
Ratio with authentic specimen composition training set is 1:2；
Step 4: being trained by the full convolutional network of symmetrical coder-decoder, the segmentation net of sea ice image is obtained
Network: being trained mixing sample using SegNet web results, and it is optimal to obtain effect by the test of true sea ice image
Divide network.
Step 5: carrying out image segmentation: the optimum segmentation network obtained using above-mentioned steps to sea ice image to be split
Sea ice image with segmentation is split.
The specific embodiment of the invention further include: the technical scheme comprises the following steps:
Step (1): the sea ice landform matrix of emulation is generated using diamond shape-square fractal method；
Step (2): the corresponding colour band information of analysis sea ice carries out color assignment to simulation sample, according to simulation sample height
Journey value sets classification thresholds, generates true value figure；
Step (3): the true sea ice sample group as pretreatment, with the tape label being collected into is filtered to simulation sample
At training set；
Step (4): it is trained by the full convolutional network of symmetrical coder-decoder, obtains the segmentation of sea ice image
Network；
Step (5): image segmentation is carried out to sea ice image to be split.
In the case where small sample training, effectively mentioned by the Training strategy that mixed reality and simulation sample form training set
The accuracy of sea ice image segmentation is risen.
Using diamond shape-square fractal method, with simulating thick ice and seawater with certain elevation and roughness
Shape carries out color assignment to thick ice and seawater in conjunction with the corresponding colour band information of true sea ice image, generates true value according to height value
Image adds label to simulation sample.The height at its square center is generated in fractal method are as follows:
Wherein, HCIt is the height value that square base plane center goes out；H00、H01、H11、H10Respectively four angle points of base plane
The height value at place；D0The random offset that center is pointed out, be a mean value be 0, variance δ2Gaussian random variable.
Claims (2)
1. a kind of sea ice image partition method based on mixing true and false sample strategy, which comprises the following steps:
Step (1): the sea ice landform of emulation is generated using diamond shape-square fractal method；
Step (2): the corresponding colour band information of the practical sea ice of analysis carries out color assignment to the sea ice landform of emulation, according to emulation
Sample height value sets classification thresholds, generates true value figure；
Step (3): it is filtered pretreatment to the sea ice of emulation, and is formed with the true sea ice sample for the tape label being collected into
Training set；
Step (4): being trained by the full convolutional network of symmetrical coder-decoder, obtains the segmentation network of sea ice image；
Step (5): image segmentation is carried out to sea ice image to be split.
2. a kind of sea ice image partition method based on mixing true and false sample strategy according to claim 1, feature exist
In: fractal method described in step (1) generates the height at its square center are as follows:
Wherein, HCIt is the height value that square base plane center goes out；H00、H01、H11、H10Respectively base plane four corner points
Height value；D0For the stochastic variable for meeting Gaussian Profile, D0Probability density function be f (x), meet:
Wherein, [0,100] mean μ ∈, variances sigma ∈ [0,1].
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811248533.6A CN109447993A (en) | 2018-10-25 | 2018-10-25 | A kind of sea ice image partition method based on mixing true and false sample strategy |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811248533.6A CN109447993A (en) | 2018-10-25 | 2018-10-25 | A kind of sea ice image partition method based on mixing true and false sample strategy |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109447993A true CN109447993A (en) | 2019-03-08 |
Family
ID=65548569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811248533.6A Pending CN109447993A (en) | 2018-10-25 | 2018-10-25 | A kind of sea ice image partition method based on mixing true and false sample strategy |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109447993A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211146A (en) * | 2019-05-16 | 2019-09-06 | 中国人民解放军陆军工程大学 | Video foreground segmentation method and device for cross-view simulation |
| CN111046885A (en) * | 2019-12-12 | 2020-04-21 | 厦门大学 | Sea ice mapping method based on sentinel I synthetic aperture radar image |
| CN110849311B (en) * | 2019-11-19 | 2021-03-26 | 中国科学院海洋研究所 | Estimation method for sea ice output area flux of polar region key channel |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150160006A1 (en) * | 2013-12-11 | 2015-06-11 | Conocophillips Company | Derivation of sea ice thickness using isostacy and upward looking sonar profiles |
| CN106156744A (en) * | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106296663A (en) * | 2016-08-01 | 2017-01-04 | 辽宁工程技术大学 | A kind of SAR sea ice image partition method and system |
| CN108009629A (en) * | 2017-11-20 | 2018-05-08 | 天津大学 | A kind of station symbol dividing method based on full convolution station symbol segmentation network |
-
2018
- 2018-10-25 CN CN201811248533.6A patent/CN109447993A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20150160006A1 (en) * | 2013-12-11 | 2015-06-11 | Conocophillips Company | Derivation of sea ice thickness using isostacy and upward looking sonar profiles |
| CN106156744A (en) * | 2016-07-11 | 2016-11-23 | 西安电子科技大学 | SAR target detection method based on CFAR detection with degree of depth study |
| CN106296663A (en) * | 2016-08-01 | 2017-01-04 | 辽宁工程技术大学 | A kind of SAR sea ice image partition method and system |
| CN108009629A (en) * | 2017-11-20 | 2018-05-08 | 天津大学 | A kind of station symbol dividing method based on full convolution station symbol segmentation network |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110211146A (en) * | 2019-05-16 | 2019-09-06 | 中国人民解放军陆军工程大学 | Video foreground segmentation method and device for cross-view simulation |
| CN110849311B (en) * | 2019-11-19 | 2021-03-26 | 中国科学院海洋研究所 | Estimation method for sea ice output area flux of polar region key channel |
| CN111046885A (en) * | 2019-12-12 | 2020-04-21 | 厦门大学 | Sea ice mapping method based on sentinel I synthetic aperture radar image |
| CN111046885B (en) * | 2019-12-12 | 2023-04-07 | 厦门大学 | Sea ice mapping method based on sentinel I synthetic aperture radar image |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Falcieri et al. | Po River plume pattern variability investigated from model data | |
| Gharari et al. | Hydrological landscape classification: investigating the performance of HAND based landscape classifications in a central European meso-scale catchment | |
| Alvioli et al. | Parameter-free delineation of slope units and terrain subdivision of Italy | |
| CN108227041B (en) | Horizontal visibility forecasting method based on site measured data and mode result | |
| CN109447993A (en) | A kind of sea ice image partition method based on mixing true and false sample strategy | |
| CN110208880B (en) | Sea fog detection method based on deep learning and satellite remote sensing technology | |
| CN108052876B (en) | Regional development assessment method and device based on image recognition | |
| CN101976429A (en) | Cruise image based imaging method of water-surface aerial view | |
| CN103632361A (en) | An image segmentation method and a system | |
| CN110222586A (en) | A kind of calculating of depth of building and the method for building up of urban morphology parameter database | |
| CN110414509A (en) | Stop Ship Detection in harbour based on the segmentation of extra large land and feature pyramid network | |
| CN114186491A (en) | Fine particulate matter concentration space-time characteristic distribution method based on improved LUR model | |
| Kraus et al. | Factors favouring phytoplankton blooms in the northern Adriatic: towards the northern Adriatic empirical ecological model | |
| CN1975762A (en) | Skin detecting method | |
| Li et al. | Geomorphological classification of aeolian-fluvial interactions in the desert region of north China | |
| Ashtekar et al. | Digital mapping of soil properties and associated uncertainties in the Llanos Orientales, South America | |
| Rosier et al. | Predicting ocean-induced ice-shelf melt rates using deep learning | |
| CN109360231A (en) | Based on the Sea Ice Remote Sensing image simulation method for dividing shape depth convolution to generate confrontation network | |
| Geyer et al. | Drifter and dye tracks reveal dispersal processes that can affect phytoplankton distributions in shallow estuarine environments | |
| Kokkinos et al. | Assessment of coastal vulnerability for present and future climate conditions in coastal areas of the Aegean Sea | |
| Moreno et al. | Geomorphometric analysis of raster image data to detect terrain ruggedness and drainage density | |
| Sankhua et al. | Hypsometric Analysis of Brahmani–Baitarani Basin Using ArcGIS | |
| Xu et al. | Hierarchical pattern recognition of landform elements considering scale adaptation | |
| Gharari et al. | Land classification based on hydrological landscape units. | |
| Arrell | Predicting glacier accumulation area distributions |
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: 20190308 |