Land use/cover classification of small areas by conventional digital camcorder imagery: A comparative performance of traditional and advanced methods
This study aims to investigate the performance of digital camcorder datasets for land cover classification. The chosen study area was the Universiti Sains Malaysia campus in Penang, Peninsular Malaysia. We encountered difficulties in obtaining cloud-free scenes because Malaysia is an equatorial region. This problem can be overcome by using airborne images. Digital images were taken from a low-altitude light aircraft (Cessna 172Q) at an average altitude of 2.44 km above sea level. The color image was separated into three bands (i.e., red, green, and blue) for multispectral analysis. We compared the performance of traditional methods (i.e., minimum distance and maximum likelihood) and advanced methods (i.e., frequency-based contextual and neural network (NN) techniques). The classified land cover map was geometrically corrected to provide a geocode map. This study presents preliminary findings vis-à-vis the potential application of an ordinary digital camcorder in local urban studies. The NN classifier produced the best result among the tested methods. A high degree of accuracy was achieved by the NN technique.