FCUnet: Refined remote sensing image segmentation method based on a fuzzy deep learning conditional random field network

Abstract

Abstract Image segmentation is pivotal for the understanding of high‐resolution remote sensing images (HRRS). However, because of the inherent uncertainties in remote sensing images and the highly complex resolution of HRRS, ambiguity often occurs among some geographic entities in the segmentation process, and the fine segmentation of HRRS is not considered sufficiently for most existing segmentation methods. Therefore, in this paper, the authors propose a new collaborative neural network structure called fuzzy deep learning conditional random field network (FCUnet) to solve the refined segmentation of HRRS. First, the authors design a fuzzy U‐Net classification network to obtain effective feature information, which introduces the fuzzy logic unit into the network to process the ambiguity and uncertainty of HRRS. Then, the authors introduce the conditional random field (CRF) at the end of FCUnet to optimize the image segmentation results. Finally, the authors validated the effectiveness and superiority of their approach on three data sets. The experiment results revealed that FCUnet had better refined segmentation performance and generalization ability than state‐of‐the‐art methods.

Keywords

• fuzzy logic
• image classification
• image segmentation
• neural nets