Remote sensing has significantly advanced water detection by applying
semantic segmentation techniques to satellite imagery. However, semantic
segmentation remains challenging due to the substantial amount of annotated
data required. This is particularly problematic in wetland detection, where
water extent varies over time and space, necessitating multiple annotations for
the same area. In this paper, we present DeepAqua, a self-supervised deep
learning model that leverages knowledge distillation to eliminate the need for
manual annotations during the training phase. DeepAqua utilizes the Normalized
Difference Water Index (NDWI) as a teacher model to train a Convolutional
Neural Network (CNN) for segmenting water from Synthetic Aperture Radar (SAR)
images. To train the student model, we exploit cases where optical- and
radar-based water masks coincide, enabling the detection of both open and
vegetated water surfaces. Our model represents a significant advancement in
computer vision techniques by effectively training semantic segmentation models
without any manually annotated data. This approach offers a practical solution
for monitoring wetland water extent changes without needing ground truth data,
making it highly adaptable and scalable for wetland conservation efforts.

本研究介绍了 DeepAqua，这是一个自我监督的深度学习模型，通过知识蒸馏来消除训练阶段中手动注释的需求，并利用归一化差异水体指数 (NDWI) 作为教师模型，训练卷积神经网络 (CNN) 从合成孔径雷达 (SAR) 图像中分割水域。该方法实现了有效地训练语义分割模型而无需手动注释数据，为监测湿地水域范围变化提供了实用的解决方案，无需地面真实数据，因此具有高度的适应性和可扩展性，可以应用于湿地保护工作。