Leaf-Less-Tree feature for semantic labeling applications on Google Earth Engine

摘要

Semantic labeling methods classify the earth surface into predefined classes. Very High Resolution images are popular in recent researches. Although they contains huge amount of data, obtaining required information is challenging. Semantic labeling methods utilize spectral and spatial features to define feature of classes. Tree class is typically considered in semantic labeling methods. Spectral information is insufficient in leaf less trees for feature extraction as there is no leaf to utilize vegetation indices. On the other side, tree branches are visible that could be used for spatial feature. A new spatial feature named Leaf-Less-Tree (LLT) is developed and presented in this work to improve leaf less tree detection in semantic labeling applications. Evaluation indicates accuracy improvement in test images.