際際滷

際際滷Share a Scribd company logo

E Cognition User Summit2009 Jo Neil Dunne Uvm Mapping Green Infrastructure

T
Trimble Geospatial Munich

1) Jarlath O'Neil-Dunne from the University of Vermont Spatial Analysis Laboratory presented on using object-based image analysis (OBIA) to map green infrastructure. 2) Existing data and traditional pixel-based classification are insufficient to meet decision-maker needs, while OBIA can extract meaningful information by combining data with advanced analysis. 3) The presentation demonstrated OBIA techniques like multi-sensor data fusion, contextual operations, and object hierarchy to map features like tree canopy and impervious surfaces at fine scales, going beyond human cognition.

1 of 38
Mapping the Green InfrastructureJarlath ONeil-DunneUniversity of VermontSpatial Analysis LaboratoryE-mail: joneildu@uvm.eduBlog: Letters from the SAL
DisclaimersThank you to the German education systemEverything in my presentation is obsolete thanks to eCognition 8Todd Taylor forgot to order the smoke machine I requestedWhat you will see at the end of this presentation will revolutionize OBIA
We Have ProblemsSource: Chesapeake Bay Program
There Are SolutionsSource: Chesapeake Bay Program
Actionable IntelligenceMayor Ellen Moyer agrees to a 50% Urban Tree Canopy in Annapolis by 2036
Needs Not Met by Existing Data
Data Information
Data + OBIA = Information
The Problem with 80%All that is clear is that the present acceptance of 80% accuracy is resulting in poorer data being delivered todecision makers than was routinely delivered, somewhat more slowly, by human interpreters in the five decadessince the end of World War II.C.E. Olson, Jr.Is 80% accuracy good enough?Proceedings from Pecora 17, 2008
OBIA SystemsRequirementsComponents
Jefferson County
Multi-Sensor Data Fusion
Overlapping Tiles10%10%
Customized Import
Batch Processing
Replicating the Human
The Default ApproachSegmentationSegmentationObjectsObjectsClassificationClassification
eCognundrum
Fine Scale Segmentation
Coarse Scale Segmentation
Cognitive ApproachSegmentationFusionMorphologyClassificationObjects
Iterative Approach
Class Hierarchy
Rule Set
Incorporating Vector GIS DataOriginal buildingsShrink based on nDSMFind tallGrow into tall
Image Processing
Image Processing
Stepwise ApproachSegmentation / classificationReclassify based on sizeRefinement on largeResult
Morphology (Tree Canopy)Original objectsActive & target classesFocus AreaGrow outGrow inReshaped objects
Morphology (Buildings)Original objectsTarget, active, & focus areaGrow & shrinkFinal objects
Incorporating Vector GIS DataChessboard segmentationFocus areaGrow loopFinal objects
Object HierarchyClassification to this pointObject hierarchyUrban areaImpervious within urban
Multi-Sensor Data Fusion
Contextual OperationsOriginal classificationGrowingEdge detectionCompetitive growing
Object Fusion
Consistent Output
Going Beyond Human CognitionS
P

More Related Content

E Cognition User Summit2009 Jo Neil Dunne Uvm Mapping Green Infrastructure

Editor's Notes

  • #12: Point cloud with classification
  • #25: 34 pages