Research on multisource remote sensing image classification algorithms based on image fusion and the EM-HMRF

Abstract

Aiming at classifying multisource remote sensing images, we first introduce a Markov Random Field (MRF) to build prior probability models for multiple object classes. The Expectation Maximization-Hierarchical Markov Random Field (EM-HMRF) algorithm is then introduced to take advantage of the equivalence relation between the EM-HMRF and the fuzzy classification method. Second, this paper focused on exploiting self-adaptivity for selecting the prior distribution model parameter β automatically, and then two fusion schemes (centralized-based and distributed-based fusion) are introduced to achieve better classification results. A new algorithm is derived for supporting multisource remote sensing image classification by using image fusion and the EM-HMRF. The experimental results on synthetic images and real remote sensing images indicate that our proposed algorithm with two fusion schemes can not only greatly improve the accuracy of image classification but also strengthen the anti-interference of noise, thereby providing good evidence to support the effectiveness and superiority of our proposed algorithm in solving multisource remote sensing image classification problems. Our proposed algorithm for image classification with a fusion scheme should have great potential value for multisource remote sensing image classification strategies.