Object-based imagery analysis for automatic urban tree species detection using high resolution satellite image

Sustainable management and monitoring the urban forest is an important activity in an urbanized world and subsequently operational approaches requires information about the status to determine the best strategic. In spite of availability of some traditional methods which imposes difficulties for tree species identification in larger urban areas, there is a demand for a fast, sensitive, that is expected to facilitating improvement of monitoring involve remote sensing technologies and image analysis techniques for urban forest inventory, urban tree species detection and ecology management. The main goal of this research is to build generic rule from World View-2 satellite imagery in conjunction with spectral, spatial, color and textural information, which is extracted from available training data for tree species detection. After segmentation, the most important step was feature selection, which is used for dimensionality reduction and discrimination between different attributes. The attribute evaluator method, which performed in this study, was CfsSubsetEval. Result of attribute selection indicates that 26 attributes were extracted from 56 attributes of the WorldView-2 image. In this research, most of satisfactory results achieved from the generic model and proves it can be easily performed to different WorldView-2 images from different areas and provided the high accuracy through algorithms for tree species detection namely, Mesua Ferrea, Samanea Saman, and Casuarina Sumatrana without using any training data. This study also explores the use and comparison of object-based classification, and two common pixel-based classification methods namely, maximum likelihood and support vector machines based on WorldView-2 satellite imagery to evaluate the potential of the object-based in compare to pixel-based to detect urban tree species. The method of maximum likelihood classification and support vector machines leads to the lowest classification accuracy since these algorithms extract only the spectral information of each pixel and consequently fail to utilize spatial, color and textural information.

Preview Text FK 2016 42 - IR.pdf Download (1MB) | Preview

Actions (login required)

View Item