ME-Net: A Deep Convolutional Neural Network for Extracting Mangrove Using Sentinel-2A Data

Mangroves play an important role in many aspects of ecosystem services. should be accurately extracted from remote sensing imagery to dynamically map and monitor the mangrove distribution area. However, popular extraction methods, such as object-oriented method, still have some defects for imagery, being low-intelligence, time-consuming, laborious. A pixel classification model inspired by deep learning technology was proposed solve these problems. Three modules were designed improve performance. multiscale context embedding module extract information. Location information restored global attention module, boundary feature optimized fitting unit. Remote ground truth labels obtained through visual interpretation applied build dataset. Then, dataset used train convolutional neural network (CNN) extracting mangrove. Finally, comparative experiments conducted prove potential extraction. We selected Sentinel-2A data acquired on 13 April 2018 Hainan Dongzhaigang National Nature Reserve China conduct a group experiments. After processing, exhibited 2093 × 2214 pixels, generated. The made satellite, which includes five original bands, namely R, G, B, NIR, SWIR-1, six multispectral indices, normalization difference vegetation index (NDVI), modified normalized water (MNDWI), forest discrimination (FDI), wetland (WFI), (MDI), first principal component (PCA1). has total 6400 images. Experimental results based datasets show that overall accuracy trained reaches 97.48%. Our method benefits CNN achieves more accurate intersection union ratio than other machine methods analysis. embedding, unit are helpful