GIS morphology knowledge aided texture windows determination for RS image classification
Zeying Lan, Yang Liu · 2017 2nd International Conference on Frontiers of Sensors Technologies (ICFST) · 2017
Texture is an important feature in RS image classification of land-use, and its precision mainly depends on the scale parameters, which are strongly correlated with the geometry characteristics of the classified objects. However, there is no a recognized reliable method for texture scale extraction. So this paper proposes an new approach to indirectly extract them with the assistance of domain GIS database knowledge. According to the human cognition characteristics, we firstly obtain a set of polygons with representative shapes from existing GIS database, by the spatial partition, the image segmentation and the space-time system evaluation. Secondly, MER algorithm is employed to describe the consistent morphology knowledge of each category's polygon sets, so that the multi texture windows can be correspondingly determined. The experimental results show that: compare with traditional methods, the proposed one calculate the results by integrating computation and inspired calculations, which can extract high-precision scale parameters with the similar performance of enumeration method, meanwhile obtain greatly improvement in algorithm efficiency. Furthermore, under the multi-scale, texture descriptors can evidently represent the different categories, thereby increase the total number of classification system, to meet the requirements of thematic cartography better.