STUDIES ON GARLIC FARMLAND EXTRACTION BY REMOTELY SENSED DATA

摘要

Garlic plant is an economic crop in Taiwan, and it’s almost the most important seasoning food for everyday Chinese food cooking. The shortage of garlic production always causes social, economical or even political problems and its price could be very expensive. In that case, the Government might be forced to open the market for foreign garlic, and so the early knowledge of the planting area of garlic is important for the balancing and control of garlic farming and market. In this paper the growth spectra of garlic were measured with GER spectroradiometer and compared to corn and other seven food -crop spectra. An airborne multi-spectral scanner was also used for images acquisition to reckon the garlic parcels NDVI thresholds. And it’s found that when the spectral reflectance curves of garlic were normalized with growth calendar, the correlation between ground and airborne measured NDVI curves was 0.76. It’s also conjectured that most food plants may have their specific growth curves, and multi-temporal remotely sensed data have the potential for garlic parcels discrimination. Furthermore, we found out the specific cultivation behavior of paving straws on garlic field during the early growth period for keeping soil moisture can induce distinctive spectrum for identification, which paves the way for the feasibility of earlier estimation of garlic area using remotely sensed data. GIS polygons of the experiment farmland were used for assisting the segmentation of garlic parcels, which included a 1x1.2 km2 area or about 202 parcels. Finally the threshold technique to recognize garlic plant site from AMSS images was developed. The results show that the accuracy is about 78% when one AMSS image is applied, while the accuracy rises to 87% when three AMSS images are applied.