Remote sensing images have become one of the most important imaging resources recently. Thus, it is important to develop high-performance techniques to process and manipulate these images. On the other hand, image processing techniques are enhanced spatially based on neural networks. Deep learning is one of the most important techniques in use for computer vision tasks and has been deployed successfully to solve many tasks. But when dealing with remote sensing images, the deep learning method faces two main problems: the underfitting problem, because of the small amount of learning data and the unbalanced receptive field problem, because of the structural stereotype of the remote sensing images. In this paper, we propose to use a complex-valued neural network to segment high-resolution remote sensing images. The proposed network can deal with the problems of remote sensing images by using an ensemble of Complex-Valued Auto-Encoder. Based on an adaptive clustering technique, this network can be used to solve the multi-label segmentation problem of remote sensing images. The proposed method achieves state-of-the-art performance when evaluated on the ISPRS 2D dataset.

中文翻译：

---

基于神经网络的高分辨率遥感影像分割新技术

遥感图像已成为最近最重要的成像资源之一。因此，重要的是开发高性能的技术来处理和处理这些图像。另一方面，基于神经网络在空间上增强了图像处理技术。深度学习是用于计算机视觉任务的最重要技术之一，并且已成功部署以解决许多任务。但是在处理遥感图像时，深度学习方法面临两个主要问题：由于学习数据量少而导致的拟合不足问题，以及由于遥感图像的结构定型而导致的接收场不平衡问题。在本文中，我们建议使用复数值神经网络对高分辨率遥感影像进行分割。所提出的网络可以通过使用复数值自动编码器的集合来解决遥感图像的问题。该网络基于自适应聚类技术，可用于解决遥感图像的多标签分割问题。当在ISPRS 2D数据集上进行评估时，所提出的方法可实现最新的性能。