Spatial shape feature descriptors in classification of engineered objects using high spatial resolution remote sensing data

Abstract

Spatial and spectral features are two important attributes that form the knowledge based database, useful in classification of engineered objects, using remote sensing data. Spectral features alone may be insufficient to identify buildings and roads in urban areas due to spectral homogeneity and similarity exhibited by them. This has led researchers to explore the spatial features described in terms of shape descriptors to improve accuracy of classification of engineered objects. This paper discusses the parameters of spatial shape features and the method for implementing these features for improving the extraction of engineered objects, using the support vector machine (SVM). SVM classified results obtained using spatial shape features is compared with gray level co-occurrence statistical features in which the former has shown better classification accuracy for buildings and roads. The classification accuracy is also calculated using spectral features of buildings and roads by classifiers such as spectral angle mapper and spectral information divergence. The analysis shows that spatial shape features improve the classification results of buildings and roads in urban areas.

https://ift.tt/2EC2nVu