Advantages of Hyperspectral over RGB image on Land Cover Classification

Authors

  • Jajang Jaenudin Politeknik Caltex Riau
  • Cheng Hao Ko National Taiwan University of Science and Technology
  • Vincentius Christian Bintang National Taiwan University of Science and Technology
  • Jih-Run Tsai National Space Organization (NSPO) Taiwan
  • Shin Fan Lin National Space Organization (NSPO) Taiwan
  • Jiun-Kai Tseng National Taiwan University of Science and Technology

Keywords:

RGB Image, Hyperspectral Image, Separability, Classification, Accuracy

Abstract

Crop identification and land cover estimation are essential for farming and land management practices in the precision agriculture field. Conventional measurements are expensive and time-consuming and thus cannot be treated as appropriate for large areas. An automatic crop or land classification should be applied to overcome these problems. Therefore, high-quality data availability is required to feed the classification tools. To fulfill the needs, we have used an airborne system for collecting in the Taiwan agriculture area. A VNIR hyperspectral image has been proven to significantly increasing accuracy compared to an RGB image. With simple discriminant algorithm LD and QD, the classification accuracy of VNIR images reaches 88.14 % and 92.02%, respectively. Meanwhile, RGB images attain 52.73% and 52.27%.

Downloads

Published

2021-08-25