Unsupervised SAR image segmentation based on quantum-inspired evolutionary gaussian mixture model

In this paper, an unsupervised SAR image segmentation algorithm (QEAGMM) based on quantum-inspired evolutionary Gaussian Mixture Models (GMM) is proposed. The method first divides the original image into small blocks. Then, the heterogeneous and homogeneous blocks are obtained using FCM clustering. Finally, the parameters of gaussian mixture model are trained by expectation-maximization (EM) algorithm using a part of homogeneous samples. However, the EM algorithm is apt to fall into a local optimum and the result is sensitive to initialization. So we embed the EM algorithm in quantum evolutionary algorithm (QEA) and propose a quantum-inspired-based EM algorithm (QEA-EM) to train the gaussian mixture model. This method not only improves the accuracy of parameters estimation but also performs better than immune clonal selection EM algorithm (ICSEM) on computational complexity. The experimental results show that compared to gaussian mixture model clustering algorithm (GMMC), the proposed method is successfully applied to texture mosaic images and SAR images, and shows overall improvement in performance.