Integration of texture and landscape features into object-based classification for delineating Torreya using IKONOS imagery

Abstract

Torreya (Torreya grandis ‘Merrillii’), an evergreen conifer, is a tree species of high economic value, which grows mainly in the mountainous areas of southern China. This research demonstrates the use of spatial information in object-based classification for improving the identification of Torreya from IKONOS images. Spatial information, including texture derived from the calculation of local indicators of spatial association (LISA) measures, and landscape features were used to improve the classification accuracy. Landscape-level features were calculated for an extensive mosaic of patches, which were defined as spatial units. A landscape feature called ‘ratio of effective mesh size’ () is proposed in this research. is determined by optimizing the ‘effective mesh size’. Both the spatial unit and the fragmentation geometry layers were readily created by utilizing an object-based approach to generate a hierarchy of objects and geographic information system (GIS)-ready vector layers. Among the landscape metrics, provided the largest statistically significant difference between the landscapes. Although both LISA measures and increased the accuracy of object-based classification, it was found that the has the best potential to be integrated into the object-based classification when utilizing spatial information of an extensive mosaic of patches.

Acknowledgements

This research was supported by the National Natural Science Foundation of China with Grant No. 30671212. We thank Dr Longbin Huang from the University of Queensland and Dr Nathan Moore from Michigan State University for revising and improving the English. We also thank the anonymous reviewers for their constructive comments and suggestions.

Notes