摘要
以地物的混合光谱曲线为主要的研究目标，利用小波变换对混合的光谱曲线进行分解，将混合地物甚至地物内部组成成分的光谱特性曲线分解出来；根据已知的标准光谱特征曲线的自相似性指标，利用分形几何学方法将分解出来的不同光谱特性的曲线进行自相似性指标测量，以实现地物或者地物内部组成成分的识别。将分形几何学和小波变换理论有机地结合起来，为利用成像光谱数据进行地物识别，尤其是解决混合光谱的地物识别提供了一个值得进一步研究的方法。
Mingle spectrum caused by the different landcover's spectrum mixing in one pixel on remote sensing image is the main reason to restrin the improvement of recognition precision.Spectral image is useful data to discriminate landcover, DN distribution curve in wavelength (band) order for each pixel or window on it can be obtained easily. In fact the DN curve represents the corresponding landcover's spectrum in field by the image radiometric correction processing, the DN curve is the mingle spectrum usually. Wavelet transform is a useful method decomposing mingle spectrum into different frequency spectra which correspond to the different landcovers. Fractal dimension is the excellent index to represent these complex curves shape.By analysising the difference of the decomposed curves with the standard landcover spectrum that the landcover is to be reconized we can discriminate and recognize successfully with the spectral image. The reconizing model presented in this paper obviously has improved the recognizing accuracy of landcover, especially minerization feature, the component of landcover and the hydrocarbon microseepage in the soil from the underground oil pools etc.
出处
《遥感技术与应用》
CSCD
1996年第1期1-6,共6页
Remote Sensing Technology and Application
关键词
成像光谱数据
分形几何学
小波理论
地物识别
Spectral image data, Fractal geometry,Wavelet transform, Mingle spectrum intepretation,Object recognition
