This website is best viewed in a browser that supports web standards.
Very-high resolution aerial imagery, while providing a very high level of detail of the land surface, also introduces new challenges for land use / land cover classification. Higher spatial resolution increases the spectral heterogeneity in the land cover features being classified, and markedly increases the shadows in the image resulting in very large numbers of confused pixels which can exceed 55% of an aerial photo image. This paper describes a methodology that resulted in a LULC classification accuracy of 89% with a Kappa of 85% (five classes: coniferous, deciduous, bare ground, water, and roads) by overcoming the intense pixel confusion caused by very heavy tree shadows and the use of very-high resolution (30cm) multispectral imagery (Green, Red, NIR) in a forested, non-urban area. ISODATA classification using a large number of clusters with subsequent iterative majority filtering handled the heterogeneity well. Shadowed areas were identified, removed, and underwent a separate ISODATA re-classification to effectively assign those pixels to the base LULC classes to overcome the confusion caused by shadows. A process of elimination, using a hierarchical class model, was a simple, effective way to overcome the challenges of having to deal with the spectrally complex, highly confused deciduous class.