Tree Detection from Urban Developed Areas in High-Resolution Satellite Images

Abstract

Preserving trees is a challenging area which indeed an automated method to analyze the percentage of trees area in respect of total land area. In this regard, a good level of extraction approach is required for finding trees area. Initially, three image segmentation approaches have implemented for detection of tree areas in urban developed regions, basic color thresholding, automatic thresholding, and region growing segmentation methods. A semi-automatic approach is proposed for detecting tree areas from high-resolution satellite images (HRSI) of urban developed area in this paper. Initially, a pixel-level classifier will train to assign into two-class label {tree and non-tree} to each pixel in a HRSI and later as pixels group. The pixel-level classification is then refined by region growing method in an image to accurately segmentation of tree and non-tree regions. Therefore, this refined segmentation results will show the tree crowns with natural shape. The proposed approach will be trained on an aerial image of different urban developed area. Finally, the outcomes show tree detection results as well as good scalability of this approach.