Semantic Classification Of Urban Trees Using High Resolution Satellite Imagery

In order to manage green resources and framing policies for sustainable development, municipalities need exact and updated inventories of urban vegetation. Automatic tree detection in urban areas using traditional classification techniques remains a very difficult task. To classify urban trees as park, roadside, and institutional-residential trees a novel three-level (pixel-object-patch) framework for semantic classification of urban trees has been proposed. The classification strategy should exploit object features, spectral response, texture, size, geometry which are used to differentiate between vegetation types based on patterns that they possess. Semantic classification is carried out by extracting green channel initially, to distinguish between vegetation and non-vegetation at the pixel level. Next, vegetation-type classification determines ground vegetation and tree vegetation at the object level by performing feature extraction. Lastly at the patch level, considering difference between spectral and textural features, and then analyzing with trained satellite images using SVM and KNN will lead to urban tree classification as park, roadside, and institutional-residential trees. The satellite images of Bengaluru city had been considered to demonstrate the applicability and effectiveness of the proposed method. The results reveal that the proposed method can achieve a satisfactory performance, with the overall accuracy reaching 94%