摘要
提出了一种基于多分辨率小波高频特征系数的高光谱遥感影像亚像素目标识别方法。首先利用多尺度小波变换将光谱信号分解为不同尺度的高频特征信号,然后借助接收操作特性曲线(ROC)和马氏距离投影寻踪求取一维最佳识别特征,最后通过高斯最大似然决策函数求解亚像素目标的存在概率。通过 38种小波函数的高光谱数据实验证明,该方法对亚像素目标的识别效果较好。
This paper presents a subpixel target detection approach based on multiscale wavelet high-pass coefficients in hyperspectral data.Firstly this algorithm uses multiscale wavelet transforms to obtain multiscale wavelet high-pass features of spectral signals.Secondly it gets the best-discrimination feature with receiver operating characteristics curves(ROC) and projection pursuit based on mahalanobis distance.Lastly it calculates the probability of subpixel target in every pixel with gauss maximum likelihood decision function.Experiment has been made with 38 kind wavelet functions in hyperspectral imagery,and proved that this algorithm is adapted for detecting subpixel targets and has an excellent precision result.
出处
《海洋测绘》
2005年第2期21-25,共5页
Hydrographic Surveying and Charting
基金
国家"863"资助项目(2002AA783050)
关键词
多尺度小波特征
接收操作特性曲线
投影寻踪
高斯最大似然决策函数
multiscale wavelet features
receiver operating characteristics curves
projection pursuit
gauss maximum likelihood decision function
