Aircraft Recognition Based on Feature Fusion and Feature Selection

Due to the diversity of aircraft 3D models and the complexity of the identification environments, anyone invariant features (for example Hu moments, Affine moments, Wavelet moments, etc.) has its own shortcomings. So combined invariant features are proposed for improving recognition rate. But combined invariant features are usually not optimal and will lower real-time performance sometimes. In this paper, an invariant-feature selection method is proposed. First, Multiple invariant features are listed and aircraft 3D model training image library (including original images, noise-added images, and occlusion images) is build. Second, multiple invariant features for every aircraft image is extracted. Third, both the features stability, which is acquired by the ratio of the mean square error to the mean value and the features distinguish-ability, which is acquired by the Markov distance, are used for sorting multiple invariant features. In the end, those invariant features which are more stable and farther Markov distance are chosen as new optimized invariant features, and sent to support vector machine for aircraft recognition. Simulation results are shown that the new optimized invariant features can improve recognition rate under less invariant features.