ABSTRACT

Remote sensing image segmentation has large uncertainty related to the heterogeneity of similar objects and complex spectrum in satellite images, causing the traditional segmentation methods to be greatly limited. Existing semantic segmentation methods represented by deep learning have made breakthrough progress. However, traditional deep learning methods, such as deep convolution neural network, are a completely deterministic model, which cannot describe the uncertainty of remote sensing image well. To solve this problem, a new deep neural network combined with fuzzy logic units is proposed in this paper, called RSFCNN (Remote Sensing image segmentation with Fuzzy Convolutional Neural Network). The network integrates convolution units and fuzzy logic units. Convolution units are used to extract discriminant features with different proportions, thus providing comprehensive information for pixel-level remote sensing image segmentation. Fuzzy logic units are used to deal with various uncertainties and provide more reliable segmentation results, and each unit handles the feature maps at a particular image scale by Gaussian blur function. Experiments were carried out on two data sets, and the results showed that RSFCNN has higher segmentation accuracy and better performance than state-of-the-art algorithms.

摘要

遥感影像分割由于卫星影像中相似物体的异质性和复杂的光谱具有较大的不确定性，使得传统的分割方法受到很大限制。现有以深度学习为代表的语义分割方法取得了突破性进展。然而，传统的深度学习方法，如深度卷积神经网络，是一个完全确定性的模型，不能很好地描述遥感图像的不确定性。为了解决这个问题，本文提出了一种结合模糊逻辑单元的新型深度神经网络，称为RSFCNN（Remote Sensing image detection with Fuzzy Convolutional Neural Network）。该网络集成了卷积单元和模糊逻辑单元。卷积单元用于提取不同比例的判别特征，从而为像素级遥感图像分割提供全面的信息。模糊逻辑单元用于处理各种不确定性并提供更可靠的分割结果，每个单元通过高斯模糊函数处理特定图像尺度下的特征图。在两个数据集上进行了实验，结果表明 RSFCNN 比最先进的算法具有更高的分割精度和更好的性能。