Post Accuracy Assessment Classification
- 1. Rahul Rakshit Robert Gilmore Pontius Jr. PhD Candidate Asst. Professor Clark University Clark University Objectives 1. To use the high quality sampled information that accuracy assessment reveals for creating a soft classified residential lawns map. 2. To incorporate supplemental variables for aiding the segregation of residential lawns from fine- green (grassy) areas. 3. To use virtual fieldwork for validation. holmes, Graduate School of Geography, Clark University 1
- 2. Traditional image Satellite Image/ Objectives processing Aerial Photo methodology Image Classification 1. To use the high quality sampled information Accuracy that accuracy assessment reveals for creating Assessment a soft classified residential lawns map. Hard Classified Map Our Contribution Supplemental 2. To incorporate supplemental Virtual Variables variables for aiding the segregation of residential 3. To use virtual Fieldwork for lawns from fine- green fieldwork for validation. Accuracy (grassy) areas. Assessment Soft Classified Map holmes, Graduate School of Geography, Clark University 2
- 3. Aerial Photos •4 Bands •Orthorectified •0.45 m Resolution Image Courtesy: Google Earth holmes, Graduate School of Geography, Clark University 2
- 4. holmes, Graduate School of Geography, Clark University 3
- 5. holmes, Graduate School of Geography, Clark University 4
- 6. Image Courtesy: Google Earth holmes, Graduate School of Geography, Clark University 5
- 7. holmes, Graduate School of Geography, Clark University 6
- 8. Supplemental variables are selected based on the likelihood of them containing residential lawns. holmes, Graduate School of Geography, Clark University 7
- 9. holmes, Graduate School of Geography, Clark University 8
- 10. Hero Map, Graduate School of Geography, Clark University 9
- 11. holmes, Graduate School of Geography, Clark University 10
- 12. holmes, Graduate School of Geography, Clark University 11
- 13. holmes, Graduate School of Geography, Clark University 12
- 14. holmes, Graduate School of Geography, Clark University 13
- 15. holmes, Graduate School of Geography, Clark University 14
- 16. holmes, Graduate School of Geography, Clark University 15
- 17. holmes, Graduate School of Geography, Clark University 16
- 18. Image Courtesy: Google Earth holmes, Graduate School of Geography, Clark University 17
- 19. Coniferous Fine-Green Fine-Green Impervious Impervious Deciduous Images Courtesy: Google Earth holmes, Graduate School of Geography, Clark University 18
- 20. Images Courtesy: Google Earth and MS Virtual Earth Street View Google Earth Virtual Earth 1 Virtual Earth 2 Virtual Earth 3 Virtual Earth 4 holmes, Graduate School of Geography, Clark University 19
- 21. Percentage of Upper Percentage Lower Stratum Fine-Green Near Buildings Zoned Res Res -1999 Study Area Bound of Lawn Bound 1 TRUE TRUE TRUE TRUE 5 64% 76 88% 2 TRUE TRUE TRUE FALSE 6 24% 38 52% 3 TRUE TRUE FALSE UN -USED 1 1% 6 13% 4 TRUE FALSE UN -USED UN -USED 6 0% 0 0% 5 FALSE TRUE TRUE TRUE 12 1% 10 19% 6 FALSE UNUSED UN -USED UN -USED 70 1% 2 6% Total 100 5% 8 12% holmes, Graduate School of Geography, Clark University 20
- 22. holmes, Graduate School of Geography, Clark University 21
- 23. Figure of Merit: The rate at which the classification is entirely correct Error of omission Correctly classified Error of comission Figure of Merit Stratum 1U2U3U4 29 Stratum 1U2U3 41 Stratum 1U2 44 Stratum 1 37 0 5 10 15 20 holmes, Graduate School of Geography, Clark University 22
- 24. 1. To use the high quality sampled information that accuracy assessment reveals for creating a soft classified residential lawns map. 2. To incorporate supplemental variables for aiding the segregation of residential lawns from fine- green (grassy) areas. 3. To use virtual fieldwork for validation. holmes, Graduate School of Geography, Clark University 23
- 25. This material is based upon work supported by the National Science Foundation under Grant No. 0709685 Any opinions, findings, & conclusions or recommendations expressed in this material are those of the author(s) & do not necessarily reflect the views of the National Science Foundation. holmes, Graduate School of Geography, Clark University 24