Remote detection of flowering Somei Yoshino (Prunus x yedoensis) in an urban park using IKONOS imagery: comparison of hard and soft classifiers
Identification of flowering trees in urban areas is challenging due to weak spectral signals and the high heterogeneity of urban landscapes. We hypothesized that a soft classifier, such as mixture tuned matched filtering (MTMF), would be better able to identify pixels including blooming cherry trees...
Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Article |
| Published: |
SPIE
2015
|
| Subjects: | |
| Online Access: | http://eprints.utm.my/id/eprint/54870/ http://dx.doi.org/10.1117/1.JRS.9.096046 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Identification of flowering trees in urban areas is challenging due to weak spectral signals and the high heterogeneity of urban landscapes. We hypothesized that a soft classifier, such as mixture tuned matched filtering (MTMF), would be better able to identify pixels including blooming cherry trees than a hard classifier such as maximum likelihood (ML). To test this hypothesis, we compared the accuracy of MTMF and ML in classifying blossoms of Somei Yoshino cherry trees (Prunus × yedoensis) in an urban park in Tokyo using IKONOS imagery. An accuracy assessment demonstrated that the MTMF classifier (overall accuracy: 62.2%, kappa coefficient: 0.507, and user's accuracy of SY: 48.1%) performed better than ML in identifying flowering SY (overall accuracy 48.7% with kappa accuracy: 0.321 and user's accuracy of blooming SY: 38.9%). Our results suggest that both methods are able to classify cherry blossoms in an urban landscape, but MTMF is more accurate than ML. However, the producer's accuracy of MTMF (72.7%) was slightly lower than ML (77.7%), suggesting that the accuracy of MTMF could decrease due to the limited number of available bands (four for IKONOS) and the existence of endmembers, such as dry grass in this study, with stronger signals than flowers. |
|---|