Feature Selection And Its Stability For Typical Geoobjects Of The High-Resolution Remote Sensing Image

The High-resolution remote sensing image provides more and more details on Geoobjects. Different Geoobjects usually possess different characteristic. Because of its rich information and large volume, the research on the typical Geoobjects can’t be completed without feature selection. It depends on feature selection with higher stability to better describe typical Geoobjects because these methods can get more stable results. Through analyzing the feature subsets, the progress of characteristics research on typical Geoobjects can be greatly promoted.As the existing stable feature selections mainly focus on the outside of the procedure, ignoring the relationship between the accuracy and stability of feature subset, a new feature selection considering stability is proposed. It puts the stability together with the relevance and redundancy based on mutual information, and gets the feature subset satisfying the criterion of max-relevance, stability and min-redundancy by iterative calculation.The proposed method is applied on the 21 land-use classes image set to demonstrate it can get a more stable result with satisfying classification accuracy, comparing with other feature selections in terms of accuracy and stability. The RADARSAT– 2 experiment in Suzhou area shows that this new method is more suitable for characteristics research on typical Geoobjects. The proposed feature selection method is applied to get the respective feature subsets of the 6 typical Geoobjects and get more stable and higher accuracy than other methods.