基于线性光谱模型的混合像元分解方法与比较

Pixel Unmixing Based on Linear Spectral Mixture Model:Methods and Comparison

线性光谱模型是目前解决城市中等空间分辨率遥感(如Landsat)中存在的混合像元问题的简单、有效的策略。本实验以广州区域为研究区,利用ENVI/IDL影像处理和开发平台对4种混合像元线性光谱分解方法进行了对比,即无约束条件法、带部分约束条件法、普通带全约束条件法和带全约束条件的可变端元法。结果表明,普通带全约束条件法和带全约束条件的可变端元法的分解结果比无约束条件法和带部分约束条件法的分解结果合理,均方根误差明显要小;同时,带全约束条件的可变端元法要优于普通带全约束条件法。光谱归一化处理则对不同分解方法带来不同的影响,应依据实际需要采取合适的光谱处理方式。

At present,Linear Spectral Mixture Model（LSMM） has been considered as a simple but effective way to extract useful information from the mixed pixel which is the problem confronted in the application of spatial medium-resolution remote sensing images（e.g.Landsat） to study the urban environment.Taking Guangzhou as a study region,four unmixture approaches based on the concept of LSMM were developed and compared using ENVI/IDL,which is a popular platform for image processing and code development.Unmixture approaches in this study are termed unconstrained,partial-constrained,generally fully constrained（GFC）,and selective endmember with fully constrained（SEFC） respectively.The results indicate that GFC and SEFC have the advantage over the unconstrained and partial-constrained approaches in the reasonability of unmixing result and with smaller root mean square error（RMSE）,however,SEFC is a better alternative to GFC.It is also shown spectra normalization might bring different effects to each method mentioned in this study.So,in conclusion,the appropriate spectral processing should be taken according to the actual needs.