Research On Multi-Feature Based Target Classfication And Recognition Algorithm For Optical Remote Sensing Image

Target detection and recognition based on optical remote sensing image is one of the most active research topics in the field of image processing and pattern recognition. Currently the classification and recognition based on single target has achieved some successes, but some problems have emerged for real scenarios based on multi-target such as inadequate representation of single feature, poor adaptability of features and lower recognition accuracy. A new algorithm framework based on multi-feature for multi-target classification and recognition of optical remote sensing image is proposed. Specific contents are as follows:Aiming at the problem of poor feature rotation adaptability, a novel hierarchical BoF-SIFT feature is proposed for target representation in remote sensing image. It combines three advantages: the translation, scaling and rotation adaptability of SIFT feature, intuitive representation of BoF feature, and the distribution characteristics of targets of Hierarchical feature. The proposed feature is tested by using recognition algorithm based on SVM and AdaBoost. Finally the results verified the effectiveness of hierarchical BoF-SIFT features.A novel multi-feature decision level fusion based recognition algorithm is proposed due to the problem of poor accuracy based on single feature for multi-target classification and recognition. Firstly, three kinds of features which are hierarchical BoF-SIFT features, improved SC features and Hu moment invariants features are extracted. Secondly, a strategy of multi-feature decision level fusion is designed. A large number of experiments show that the algorithm for multi-target classification of optical remote sensing image have better performance.Seen from a view of practical application, a feasible algorithm framework of multi-target classification and recognition based on optical remote sensing image is proposed and implemented. It can achieve multi-target automatic detection and recognition. The results show that the proposed framework is workable and has a certain practical significance.