如何准确、有效地提取影像中的特征信息是影像目标识别的核心内容，尤其是对中分辨率影像和高分辨率影像的特征信息的提取是影像目标识别的热点和难点。本论文以高分辨率影像为重点研究对象，探讨如何深层次地挖掘其特征信息。目前，多种提取特征的方法（主成分分析、几何特征提取、面向对象特征提取，纹理特征提取，空间关系特征提取等）在高空间分辨率影像目标识别中取得了一定的效果。然而，这些方法或仅考虑了数据的低阶统计特征，对数据源的限制较多；或仅考虑了全局特征，不能很好地捕捉图像中对象的局部特征，并不能满足高空间分辨率影像的复杂性，因此，识别效果不是很理想。因此，本文在高空间分辨率影像的目标识别中引入了一种在信号处理领域中实现盲源分离的方法-独立成分分析（Independent Component Analysis，ICA）。本文第一部分介绍了独立成分分析的起源发展及其数学模型。独立成分分析是将一系列的随机变量表示成若干个统计独立的变量的线性组合的方法。重点阐述了如何使用独立成分分析方法对高空间分辨率影像进行特征提取以及对提取得到的特征进行优化。第二部分详细论述了基于独立成分分析的高空间分辨率影像目标识别系统的设计与实例分析。针对基于传统独立成分分析方法提取的特征空间不能最优区分不同类别样本的缺点，本文首次提出一种改进的基于独立成分分析的目标识别方法（Multi-ICA）。以北京地区的高分辨率卫星Quickbird影像为例， 进行Multi-ICA和传统ICA方法的目标识别对比实验，实验结果表明，本文提出的Multi-ICA算法的识别率有明显的提高，并且在一定程度上，缓解了由于样本数量增加导致样本特征向量维数增加的问题。本文中的基于独立成分分析的高空间分辨率影像目标识别系统包括了若干种常用的模式识别算法，不仅可以对普通的特征数据进行识别，而且提供了特征提取模块，对高空间分辨率影像进行特征提取。针对高空间分辨率影像数据的复杂性，通过标准化、归一化和白化等方法对原始数据进行提取和加工，这样不仅简化了高空间分辨率影像数据的复杂性，而且提高了识别的准确率。

How to exact the feature information of images is the core of recognition of images. Specially how to exact the the feature information of high spatial resolution image is the research's hot spot and difficulty. This paper takes high spatial resolution image as research's target to discuss how to deeply dig the feature information of high spatial resolution images.Until right now, many feature extraction methods (principal component analysis, geometric feature extraction, object-oriented feature extraction, texture feature extraction, etc) in the field of high spatial resolution image's recognition have achieved certain results. However, some of these methods take into account the low-level statistical features of data resource and add many restrictions to the data resource. The other only considers the global feature which can not properly reflect the local feature of image objects and can not meet the complex nature of high spatial resolution images. Therefore, recognition effect is not very satisfactory. In order to improve the recognition effect, independent component analysis namely ICA which is a kind of blind source separation methods in signal processing is introduced into the recognition of high spatial resolution image. The first part of this paper describes the ICA's origin, development and mathematical model. ICA is an approach which can make a series of random variables expressed by the linear combinations of statistically independent variables. This part focuses on how to use ICA to extract the feature of high spatial resolution image and how to optimize the feature space. The second part discusses the design of high spatial resolution image recognition system based on ICA and a case analysis. The feature space, extracted by traditional method based on ICA, cannot optimally distinguish between different types of samples. To solve the problem, this article comes up with the idea of an improved algorithm based on independent component analysis, namely Multi-ICA. The experiment on Beijing's high-resolution Quickbird satellite image showed that the recognition rate of our Multi-ICA algorithm compared with those of traditional ICA was obviously increased, and the recognition rate kept stable when recognition types increased, and it alleviates the problem which is that the dimension of sample feature vector increases as the samples increases. High spatial resolution image recognition system based on ICA in this paper includes commonly-used pattern recognition algorithms which not only can identify common feature data but also provides a feature extraction module to extract the feature of high spatial resolution images. For the complexity of high spatial resolution image data, the system uses the standardized, normalized methods to extract and process raw data, so that not only simplifies the complexity of high spatial resolution image but also improved the recognition accuracy.