Classification of Post-earthquake High Resolution Image Using Adaptive Dynamic Region Merging and Gravitational Self-Organizing Maps

Abstract

Post-earthquake high resolution image classification has opened up the possibility for rapid damage mapping, which is crucial for damage assessments and emergency rescue. However, the classification accuracy is challenged by the diversity of disaster types as well as the lack of uniform statistical characteristics in post-earthquake high resolution images. In this paper, combining adaptive dynamic region merging (ADRM) and gravitational self-organizing map (gSOM), we propose a novel object-based classification framework. This approach consists of two parts: the segmentation by ADRM and the classification by gSOM. The ADRM produces the homogeneous regions by integrating an adaptive spectral-texture descriptor with a dynamic merging strategy. The gSOM regards the regions as basic unit and characterized them explicitly by fractal texture to adapt to various disaster types. Subsequently, these regions are represented by neurons in a self-organizing map and clustered by adjacency gravitation. By moving the neurons around the gravitational space and merging them according to the gravitation, the gSOM is able to find arbitrary shape and determine the class number automatically. To confirm the validity of the presented approach, three aerial seismic images in Wenchuan covering several disaster types are utilized. The obtained quantitative and qualitative experimental results demonstrated the feasibility and accuracy of the proposed seismic image classification method.