An adaptive texture selection framework for ultra-high resolution UAV imagery

The capacity for additional textural derivatives to compensate for the lack of broader spectral sensitivity of consumer grade digitial cameras is established within a UAV context. A texture selection framework utilising random forest machine learning, was developed for application with ultra-high spatial resolution UAV imagery limited to the visible spectrum. The framework represents an adaptive approach, providing a rapid assessment of different texture measures relative to a specific user-defined application. This framework is illustrated within the context of UAV salt marsh mapping. This study highlights the importance of texture selection for improving classification of UAV imagery exhibiting high local spatial variance.