Decomposition-based multiobjective particle swarm optimization for change detection in SAR images

Abstract

Owing to the immunity to illumination and atmospheric conditions, synthetic aperture radar (SAR) images have been the main source of data for environmental monitoring. However, it is a challenging task for change detection because of the influence of speckle noise. In this paper, we propose an unsupervised multiobjective particle swarm optimization approach based on decomposition for change detection in SAR images. For the change detection task, it can be modeled as a multiobjective optimization problem (MOP), which consists of two contradictory objectives, namely, retaining image change details and removing the speckle noise. We optimize this MOP by using particle swarm optimization, which decomposes it into a set of subproblems by assigning different weights to these two objectives, thus obtaining the optimal trade-off between them. To accurately identify changed regions, the strategy of majority voting is employed to assemble partial good solutions to generate the final change detection result. The impressive experimental results on both real data sets have demonstrated the effectiveness and superiority of the proposed method.