Semantic Segmentation for Mangrove Using Spectral Indices and Self-Attention Mechanism

Abstract

The segmentation of mangroves based on remote sensing images is of great significance for the protection of mangroves and the ecological environment. Deep learning methods have been widely used in image segmentation, among them, supervised learning methods based on convolutional neural networks can achieve excellent accuracy in image segmentation most of the time. However supervised learning methods often require the production of complete labels, which is a laborious task. In this paper, we propose an innovative mangrove segmentation model based on domain adaptation method, which integrates the self-attention mechanism to make the model focus on more important image channels, and combines the remote sensing spectral indices to solve the problem that the domain adaptation method may lose edge information. Experimental results on Landsat8 dataset show that our proposed model achieves superior performance, and outperforms several popular semantic segmentation models.