遥感图像包含丰富的光谱和空间信息,现已广泛用于农业、林业、灾害评估等众多领域。对遥感图像进行分类是各领域进行后续研究的基础,但因遥感图像数据高维、小样本等特点给分类精度带来了挑战。近年来,在计算机视觉领域取得显著成功的...遥感图像包含丰富的光谱和空间信息,现已广泛用于农业、林业、灾害评估等众多领域。对遥感图像进行分类是各领域进行后续研究的基础,但因遥感图像数据高维、小样本等特点给分类精度带来了挑战。近年来,在计算机视觉领域取得显著成功的深度学习已被广泛应用于遥感图像分类中。本文首先介绍了遥感图像分类的背景及目前存在的问题,然后对遥感图像分类领域中应用较为广泛的深度学习经典模型堆叠自编码器(Stacked Autoencoder, SAE)、深度置信网络(Deep Belief Networks, DBN)、卷积神经网络(Convolutional Neural Networks, CNN)、生成对抗网络(Generative Adversarial Networks, GAN)做了简要概述,接下来介绍了SAE、DBN、CNN和GAN在遥感图像分类领域中的应用发展状况,并对这些经典深度模型作了对比分析,最后对遥感图像分类的未来研究方向进行了展望。Remote sensing images contain rich spectral and spatial information, and have been widely used in many fields such as agriculture, forestry, and disaster assessment. The classification of remote sensing images is the foundation for subsequent research in various fields, but the high dimensionality and limited sample size of remote sensing image data present challenges to classification accuracy. In recent years, deep learning, which has achieved remarkable success in the field of computer vision, has been widely applied in remote sensing image classification. This paper first introduces the background and current problems of remote sensing image classification, and then provides a brief overview of the widely used deep learning classic models Stacked Autoencoder (SAE), Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), and Generative Adversarial Networks (GAN) in the field of remote sensing image classification. Next, it introduces the development status of SAE, DBN, CNN, and GAN in the field of remote sensing image classification, and provides a comparative analysis of these classic deep learning models. Finally, it looks forward to the future research directions of remote sensing image classification.展开更多
密度峰值聚类(Density peaks clustering简称DPC)算法是2014年在美国Science期刊上发表的一种非常简洁优美的聚类算法,它不需要像经典K-means算法那样迭代,也不需要很多参数。DPC算法的核心思想在于对聚类中心的刻画,它通过计算数据集...密度峰值聚类(Density peaks clustering简称DPC)算法是2014年在美国Science期刊上发表的一种非常简洁优美的聚类算法,它不需要像经典K-means算法那样迭代,也不需要很多参数。DPC算法的核心思想在于对聚类中心的刻画,它通过计算数据集中每个数据点的局部密度和该点到具有更高局部密度的点的最小距离,当数据点的■的值较大时,该点为聚类中心。然而通过分析,发现这样选取聚类中心得聚类效果不具有稳健性,依赖于和的量纲。本文提出一种改进的密度峰值聚类算法,将和归一化后的和记为每个点的权重,构造函数■作为选取聚类中心的判决函数,结合模拟计算,验证本文的方法更鲁棒,选取聚类中心效果更好,且复杂度降低。展开更多
文摘遥感图像包含丰富的光谱和空间信息,现已广泛用于农业、林业、灾害评估等众多领域。对遥感图像进行分类是各领域进行后续研究的基础,但因遥感图像数据高维、小样本等特点给分类精度带来了挑战。近年来,在计算机视觉领域取得显著成功的深度学习已被广泛应用于遥感图像分类中。本文首先介绍了遥感图像分类的背景及目前存在的问题,然后对遥感图像分类领域中应用较为广泛的深度学习经典模型堆叠自编码器(Stacked Autoencoder, SAE)、深度置信网络(Deep Belief Networks, DBN)、卷积神经网络(Convolutional Neural Networks, CNN)、生成对抗网络(Generative Adversarial Networks, GAN)做了简要概述,接下来介绍了SAE、DBN、CNN和GAN在遥感图像分类领域中的应用发展状况,并对这些经典深度模型作了对比分析,最后对遥感图像分类的未来研究方向进行了展望。Remote sensing images contain rich spectral and spatial information, and have been widely used in many fields such as agriculture, forestry, and disaster assessment. The classification of remote sensing images is the foundation for subsequent research in various fields, but the high dimensionality and limited sample size of remote sensing image data present challenges to classification accuracy. In recent years, deep learning, which has achieved remarkable success in the field of computer vision, has been widely applied in remote sensing image classification. This paper first introduces the background and current problems of remote sensing image classification, and then provides a brief overview of the widely used deep learning classic models Stacked Autoencoder (SAE), Deep Belief Networks (DBN), Convolutional Neural Networks (CNN), and Generative Adversarial Networks (GAN) in the field of remote sensing image classification. Next, it introduces the development status of SAE, DBN, CNN, and GAN in the field of remote sensing image classification, and provides a comparative analysis of these classic deep learning models. Finally, it looks forward to the future research directions of remote sensing image classification.
文摘密度峰值聚类(Density peaks clustering简称DPC)算法是2014年在美国Science期刊上发表的一种非常简洁优美的聚类算法,它不需要像经典K-means算法那样迭代,也不需要很多参数。DPC算法的核心思想在于对聚类中心的刻画,它通过计算数据集中每个数据点的局部密度和该点到具有更高局部密度的点的最小距离,当数据点的■的值较大时,该点为聚类中心。然而通过分析,发现这样选取聚类中心得聚类效果不具有稳健性,依赖于和的量纲。本文提出一种改进的密度峰值聚类算法,将和归一化后的和记为每个点的权重,构造函数■作为选取聚类中心的判决函数,结合模拟计算,验证本文的方法更鲁棒,选取聚类中心效果更好,且复杂度降低。
