A Novel Contextual Speckle Reduction Method of Polsar Images: Evaluation of Speckle Reduction Effects on Sea Ice Classification

Polarimetric Synthetic Aperture Radar (PolSAR) sensors can transmit and receive electromagnetic waves in different polarization states. As a result, PolSAR images are complex multi-dimensional data which can be extensively applied in sea ice monitoring and classification. The discrimination capability of PolSAR data makes them a unique source of information with a significant contribution to make in tackling problems concerning environmental applications, especially in the Arctic regions. The increasing temperature of the Earth due to climate change, combined with the degradation of ice thickness in the Arctic region increases the navigation of shipping routes more than ever before. Thus, oil and gas resources at high latitudes are more accessible, which will increase economic activities in the Arctic. Therefore, sea ice monitoring is one of the most important applications of remote sensing tools, since it is time and cost efficient in comparison with extensive field campaigns. Speckle reduction is one of the most important pre-processing steps in Polarimetric Synthetic Aperture Radar (PolSAR) image processing. This paper demonstrated the importance of the de-speckling method for sea ice classification using PolSAR image. A novel contextual speckle reduction method was proposed based on the integration of a Markovian energy function as a contextual analysis, and the Gaussian model as a pixel-wise analysis which maintained the mean value of original data in the homogenous areas and preserved the edges of features in the heterogeneous regions. Our proposed method is applied to a full polarimetric L-band dataset, ALOS-1 PALSAR acquired from Baffin Bay. The efficiency of the proposed method was compared with other well-known de-speckling methods, such as the Kuan method, the enhanced Lee method, and the Nonlocal Means and Refined Lee method. Different de-speckling indices, including the Mean Square Error (MSE), the Signal to Noise Ratio (SNR), the Edge Saving Index (ESI), and Equivalent Number of Looks (ENL) were extracted for filtered images from different methods. Moreover, the capability of all de-speckled PolSAR images was compared in terms of overall sea ice classification results. The results demonstrated that the proposed method was outperformed by other well-known methods in terms of both de-speckled indices as well as classification results.