多光谱图像水域变化检测方法研究  

Water Change Detection for Multi-spectral Images

作　　者：张永梅[1,2] 刘海伟[1] 姜明[2] 

机构地区：[1]北方工业大学信息工程学院,北京100144 [2]北京大学数学科学学院,北京100871

出　　处：《计算机测量与控制》2013年第2期305-307,339,共4页Computer Measurement &Control

基　　金：国家科技支撑计划资助项目(2012BAH04F00);北京市高校学术创新团队资助项目(PHR201007121);北京市图像信息处理与智能识别科研平台建设项目(PXM2012_014212_000024);北方工业大学重点研究计划项目资助

摘　　要：针对遥感图像中水域的变化检测问题,提出一种综合特征级和像素级变化检测的多光谱图像变化检测方法,用于检测多时相遥感图像中水域的变化情况;首先利用BP神经网络,选取输入层为5,选定隐含层为10,完成多光谱图像的水域有效分类,再根据多光谱图像的水域特征,通过HSV空间准确定位水域区域,最后利用差值法进行多光谱图像水域变化检测;实验结果表明,该方法提高了变化检测的自动化程度和变化检测效率,具有较高的检测精度。A novel two--step change detection algorithm combining feature--level and pixel--level techniques is proposed to detect chan- ges of water in multi--temporal remote sensing images. Firstly, water areas in multi--spectral images should be accurately classified on the basis of BP neural network, the input layer is selected as 5, the hidden layer is selected as 10. Secondly, water areas should be accurately located by the water features in HSV space. At last, water change detection in multi--spectral images will be finished by the difference method. The experiment results show this method improves change detection efficiency and the degree of automation of change detection. The method has higher detection precision.

关 键 词：变化检测 BP神经网络 多光谱图像 特征提取 

分 类 号：TP75[自动化与计算机技术—检测技术与自动化装置]