A/Prof Samsung Lim of UNSW Sydney

Classification of hyperspectral imagery (HSI) is a promising technique for the understanding of image scenes. However, there exist some challenges. For example, high spectral dimensionality, high variability of available training samples among different classes and the presence of mixed pixels are well-known problems in the HSI classification. Now with the advances of image processing technology, not only spectral information but also spatial information has been applied to the HSI classification. In this talk, an integrated framework for the HSI classification will be presented. Firstly, two convolutional neural networks (CNNs) are developed for the extraction of representative features. The two neural networks are able to predict the class membership probabilities of test pixels. Secondly, two probabilistic relaxation methods are integrated into the framework in order to refine the probabilistic results from the CNNs. This framework enhances the classification performance by exploiting the contextual information available from neighbor pixels. This is particularly advantageous when only a limited number of training samples are available. The proposed framework was tested on both simulated and real-world data sets. The experimental results suggest that the proposed method outperforms CNN-based methods.

Biography:

Associate Professor Samsung Lim teaches GIS in the School of Civil and Environmental Engineering, UNSW, Sydney, Australia. Samsung’s main research interests are threefold: GIS, lidar and remote sensing. Samsung’s publications in GIS include spatial and temporal fire occurrence patterns using landscape attributes such as climate, terrain and land use to develop a better understanding of wildfire occurrence across Australia, identifying the most significant attributes linked to fire occurrence, effective flood emergency plans and response measures, integration of spatial analysis and an agent-based model for shelter assignment and routing in flood scenarios, and risk analysis methods to flag epidemics for urgent intervention. Samsung’s publications in lidar and remote sensing include deep learning from multispectral/hyperspectral images, data fusion of lidar point clouds and multispectral/hyperspectral images, spatial information extraction from lidar data, UAV imagery and aerial/satellite imagery for data segmentation and classification, digital elevation modeling, feature extraction, building edge detection, change detection, and monitoring environmental changes. Samsung is a national delegate of Commission 3, International Federation of Surveyors (FIG) and a senior member of Institute of Electrical and Electronics Engineers (IEEE).

Email:

Address:UNSW Sydney, , Sydney, Australia