目的 针对两期高分辨率遥感影像，提出一种结合邻域相关影像（NCI）和最大相关性最小冗余性特征选择（mRMR）的面向对象变化检测方法。方法 为了验证该方法的有效性，设计了3组对比实验：1）只使用mRMR特征选择与未使用mRMR特征选择的效果比较；2）使用NCI与mRMR特征选择相结合与只使用NCI的效果比较；3）使用NCI与mRMR特征选择相结合与只使用mRMR特征选择的效果比较。结果 实验结果表明，使用NCI与mRMR特征选择相结合的变化检测效果要优于只使用NCI或是只使用mRMR特征选择的效果，更优于两者都不使用的效果。结论 理论上本文方法并不会因为采用了不同的高分辨率遥感数据源而影响其对变化检测的优越性，实际情况是否如此，还需进一步通过实验验证。

Objective The methods of change detection based on high resolution images have important applications in land use/land change, water quality change, forest resource monitoring and military target detection. In this paper, we introduce a new object-oriented change detection method for two high-resolution remote sensing images, which combines neighborhood correlation images (NCI) and the minimum-redundancy-maximum-relevance feature selection (mRMR) together. Content of main experiments:In this paper, study area is located in Shunyi district, Beijing. We choose two rapideye images as experimental data and both of them are without any cloudy covering them. Comparisons with reviewed researches:These two image data both contain abundant kinds of object types while the number of object types for traditional change detection are relative small. Method We design three comparative experiments to verify the effectiveness of this method: 1) compare the results of using mRMR feature selection versus without using mRMR feature selection; 2) compare the results of using NCI and mRMR feature selection together versus only using NCI; 3) compare the results of using NCI and mRMR feature selection together versus only using mRMR feature selection. Result The results show that the effect of change detection combining NCI and the mRMR feature selection is better than only using NCI or only using mRMR feature selection, and is more superior than neither NCI nor mRMR was introduced. Meanings:The most innovative contents about this paper is the method about how to deal with features, involving how to condense features and how to deal with high dimension features. That means we can use much more knowledge of data mining and machine learning to deal with remote sensing change detection. However,we need use more high resolution images to test this method. Conclusion In theory, effectiveness of this method for change detection does not be affected by using different high resolution remote sensing data sources; However, the reality needs to be further verified by experiments.