摘要
针对生产任务中常用的卫星高光谱数据空间分辨率不高、地物复杂的特点,提出一种实用性和灵活性较强、效率较高、不依赖空间信息的分类方法。对高光谱遥感影像数据进行分析,依据其光谱信息丰富,但在复杂地物中空间特征不足的实际情况,采用离散采样的方法,充分利用质量较好的样本点进行特征提取。对传统卷积神经网络进行改进,通过卷积层与池化层的重组等措施,使其更充分地利用地物的光谱特性。该方法在珠海一号高光谱影像上实现了对地物的有效分类。
Aimed at the characteristics of low spatial resolution and complex ground features of satellite hyperspectral data commonly used in production tasks,a practical,flexible,efficient classification method which is independent of spatial information is proposed in the paper.The analysis of hyperspectral remote sensing image data is based on the fact that the spectral information is rich,but the spatial features are not enough in the complex features.The method of discrete sampling is adopted to make full use of the good quality sample points to extract the features.The traditional convolutional neural network is improved to make full use of the spectral characteristics of ground objects through the recombination of convolutional layer and pooled layer.Effective classification of ground objects on the Zhuhai-1 hyperspectral image has been achieved by this method.
作者
杨晔
龚志辉
刘相云
陈旭东
YANG Ye;GONG Zhihui;LIU Xiangyun;CHEN Xudong(Information Engineering University, Zhengzhou 450001, China)
出处
《测绘科学技术学报》
CSCD
北大核心
2021年第2期160-165,共6页
Journal of Geomatics Science and Technology
关键词
光谱特性
复杂地物
离散采样
卷积神经网络
珠海一号
spectral characteristic
complex ground objects
discrete sampling
convolutional neural network
Zhuhai-1
