混合像元问题是定量遥感的主要障碍之一。将混合像元问题归结为类内与类间像元混合两类，并对类内混合像元分解问题加以研究。混合像元分解的关键在于确定组分光谱，确定组分光谱的方法很多，但大多数方法基于以下假定，即从图像本身可以找到纯组分光谱，然而这一假定对于类内混合像元分解问题来说很难成立。提出采用高光谱与多角度相结合的方法，利用几何光学模型和线性光谱混合模型进行类内混合像元分解。即首先利用多角度数据反演几何光学交互遮蔽(GOMS)模型获得组分光谱，再对高光谱数据进行组分光谱分解。由于该方法直接从混合光谱产生的机理出发，因而更容易获得真正的亚像元信息。为减小反演误差，反演过程中采用改进的多阶段的反演策略，并充分利用多角度图像本身提供的先验信息。用BORE—AS试验获取的高光谱与多角度数据所作的研究表明，该方法可以获得比较理想的分解结果。

The problems of pixel mixture are the main obstacles of quantitative remote sensing. In this paper，we classify the problems of pixel mixture into two groups：mixture in class and mixture between classes. and focus our study on the former. The key problem of subpixel unmixing is to determine the component spectra. Many methods have been developed to determine the component spectra. However，most of these methods base on the assumption that pure component signatures can be found from the image itself. Obviously，this assumption is not correct for pixel mixture in class. In this paper，a new method to combine hyperspectral and multiangular data together to retrieve subpixel information is introduced. This method begins from the definition of the component signatures and tries to get a priori information from the data itself，so it's more likely to acquire the“pure”endmember spectra and the subpixel information extracted by this method is more reliable than some other ways. To demonstrate the result of this method，we use BOREAS（the Boreal Ecosystem-Atmosphere Study）data to extract component signatures and the corresponding areal proportions in SSA （Southern Study Area）OJP（Old Jack Pine）site. The result is encouraging.