Multiscale remote sensing data segmentation and post-segmentation change detection based on logical modeling: Theoretical exposition and experimental results for forestland cover change analysis

Quantification of forestland cover extents, changes and causes thereof are currently of regional and global research priority. Remote sensing data (RSD) play a significant role in this exercise. However, supervised classification-based forest mapping from RSD are limited by lack of ground-truth- and spectral-only-based methods. In this paper, first results of a methodology to detect change/no change based on unsupervised multiresolution image transformation are presented. The technique combines directional wavelet transformation texture and multispectral imagery in an anisotropic diffusion aggregation or segmentation algorithm. The segmentation algorithm was implemented in unsupervised self-organizing feature map neural network. Using Landsat TM (1986) and ETM+ (2001), logical-operations-based change detection results for part of Mau forest in Kenya are presented. An overall accuracy for change detection of 88.4%, corresponding to kappa of 0.8265, was obtained. The methodology is able to predict the change information a-posteriori as opposed to the conventional methods that require land cover classes a priori for change detection. Most importantly, the approach can be used to predict the existence, location and extent of disturbances within natural environmental systems.