Development of Object-based Forest Type Classification Methodology

Abstract

In this paper, we proposed a new method to generate forest type classification maps accurately and efficiently using high-resolution imagery, such as aerial photographs. We introduced hierarchical image segmentation techniques which not only make use of high-resolution images, but also combine low-resolution images created from the high resolution images themselves. One of the major challenges in object-based classification is the selection of the right texture features for the accurate classification of forest types. Here we proposed three new types of structural texture features, namely, the pattern of illuminated / shadowed pixels, the pattern of the boundary lines of illuminated / shadowed crown areas, and local binary pattern textures to describe the morphological characteristics of forests. Regarding the choice of spectral features, we utilized normalized spectral features of the illuminated areas of tree crowns. For each object, spectral and textural features were calculated, and forest classification was performed using the Nearest Neighbor classification method. To assess the effectiveness of the proposed method, the classified map was verified by comparing the entire map to a visual interpretation map. Furthermore, we also compared the accuracy and processing time of the proposed object-based method with those of the conventional object-based classification method. Our evaluation confirmed that by utilizing the proposed method we could achieve classification results that were at a practical level close to the results of visual interpretation.