an image classification sampling methodology based on the integration of ip/gis ... ·...
TRANSCRIPT
![Page 1: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/1.jpg)
An Image ClassificationSampling Methodology
Based on the Integration ofIP/GIS Capabilities
ByKen Stumpf
Geographic Resource SolutionsArcata, CA
![Page 2: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/2.jpg)
Techniques for Developinga Comprehensive and
Cost EffectiveLand Cover Mapping
Training Data Set
![Page 3: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/3.jpg)
I will briefly present ...
� The need for this methodology� The sampling methodology
– Image Stratification–Candidate Site Database
Development–Candidate Site Refinement–Sample Plan Development and
Administration
� Benefits of this methodology
![Page 4: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/4.jpg)
Training Data are ExtremelyImportant
Image processing training datacollection issues:–Foundation for accurate and detailed
land cover mapping� represent diversity of land cover� represent area of interest
–Significant cost component� large number of “types”� travel time and equipment� number of samples
– Limited field sampling opportunity
![Page 5: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/5.jpg)
Control Effort and Cost
Reduce the number of data collectionsites while still describing thediversity and geographic range ofthe project area–perform illumination correction–apply classification training site
sampling methodology–use existing data
![Page 6: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/6.jpg)
Illumination Correction
![Page 7: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/7.jpg)
Reduce Sampling Effort
� Eliminate collection of erroneous data� Eliminate collection of redundant data
Let’s preprocess the data so we only collect data only at ‘valid’
training sites.
![Page 8: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/8.jpg)
Characteristics of ‘Valid’Training Areas
� Spectrally homogeneous andnormally distributed
� Accessible� Sufficient size
– for an adequate sample of pixels– to locate in the field– to distinguish from neighboring types
![Page 9: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/9.jpg)
The Norm isOpportunistic Sampling
� Overview project area� Visually select sites� Visit and collect some data� Build the training set as you go
![Page 10: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/10.jpg)
A different approach ….
� Let’s use the data and our IP/GIStools to guide and direct our trainingdata collection efforts -
avoid invalid sites and focus onthose sites necessary to build adetailed training data set thataccurately represents the range ofland cover characteristics over theentire project area.
![Page 11: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/11.jpg)
GRS’s Training DataDevelopment Methodology
� Stratify Project Area/Imagery� Build Candidate Site Database� Refine Candidate Sites� Develop and Administer Training
Set Sample Plan
![Page 12: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/12.jpg)
Image Stratification
� Use unsupervised classification togenerate spectrally homogeneousclasses– Identify diversity of the project area–Estimate area/frequency and relative
magnitude of ‘types’–Break project area into sub regions or
ecotypes to increase diversity ofclasses
![Page 13: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/13.jpg)
![Page 14: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/14.jpg)
Determine Class Area andRelative Magnitude
![Page 15: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/15.jpg)
Unique Area Isodata ClassDatabase
![Page 16: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/16.jpg)
Candidate Training SiteDatabase Refinement
� Apply minimum size limit of 60pixels or 13 acres to the arealisting and create a new set ofcandidate training site locations
select id, iso_class from grid_valwhere pix_count >= 60
Reduced 8.6 million unique areas to 36,833 candidate areas
![Page 17: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/17.jpg)
Characterize Candidate Sites -Frequency by Class
![Page 18: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/18.jpg)
Sample Site Identification
� Determine missing or rare classesselect distinct iso_class from grid_val where iso_class not in
(select distinct iso_class from candidate_trsite)
– identified 0 missing isodata classesselect iso_class,count(*) from candidate_trsite group by
iso_classhaving count(*) < 5 order by iso_class
– identified 42 scarce isodata classes
� Add additional candidate areas tosupplement scarce classes by loweringminimum size limit to 45 pixels or 10 acres–added 305 sites to these 42 classes
![Page 19: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/19.jpg)
Generate GIS Database
� Reclass all areas (pixels) with sizeless than the specified minimumsize(s) to a value of 0
� Vectorize the remaining areas andidentify by unique area number
![Page 20: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/20.jpg)
Reclass Areas ‘Too Small’to ‘0’
![Page 21: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/21.jpg)
Vectorize and LabelCandidate Areas
![Page 22: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/22.jpg)
Candidate Training SiteDatabase Contains ...
� Isodata class value� Area - number of pixels� X,Y coordinates� Slope, aspect, and elevation� Scene indicator� Scarcity indicator� Training group number
![Page 23: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/23.jpg)
Generate Field Maps
![Page 24: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/24.jpg)
Candidate SiteSelection Criteria
� Access� Distance traveled� Scarce isodata classes� Proximity of candidate training
sites to each other� Areas of scene overlap
![Page 25: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/25.jpg)
Constraints - ‘No-Fly’ Zones
![Page 26: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/26.jpg)
Areas of Interest/Fuel Dumps
![Page 27: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/27.jpg)
‘No-Fly’ Zones AND AOIs
![Page 28: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/28.jpg)
Sampling Plan Developmentand Administration
� Daily plan development– sample scarce isodata classes– fulfill daily plan requirements– fulfill overall plan requirements–obtain multi-scene samples
� Prepare field maps� Monitor progress
![Page 29: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/29.jpg)
Area Candidate Site Report
![Page 30: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/30.jpg)
Daily Plan Schedule
![Page 31: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/31.jpg)
Plots and Field Maps
![Page 32: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/32.jpg)
Sampling Status by Area
![Page 33: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/33.jpg)
Sampling Status - Overall
![Page 34: An Image Classification Sampling Methodology Based on the Integration of IP/GIS ... · 2016-04-15 · An Image Classification Sampling Methodology Based on the Integration of IP/GIS](https://reader030.vdocuments.us/reader030/viewer/2022041100/5ed72053c30795314c173eac/html5/thumbnails/34.jpg)
Results and Benefits
� Better signatures - less confusion� Fewer rejected areas� Project area has been sampled
–significant types– scarce types–project area variation
� Less speculation/seat-of-pantsjudgement
� Lower cost and/or less time