Wavelet Moments Algorithm Applied Research In Image Recognition

With the development of science and technology, automatic target recognition (ATR) will play more and more important role in modern weapon systems. Feature extraction is a necessary process of automatic target recognition. There are many methods of Feature extraction. Because invariant moments are invariable to transportation, rotation and scale transformation, they are widely used in automatic target recognition. The main works are as follows:1. The traditional invariant moments are briefly introduced and their characteristics are analyzed and compared in detail.2. Because traditional invariant moments have their limitations in recognition and computation, the wavelet moment invariants are introduced and applied to the image recognition of airplane. Simulation results show that the wavelet moment method is satisfying both in recognition efficiency and computation burden.3. In consideration of the large feature dimensions of the wavelet moment invariants, we first choose a set of better features by combining divergence with Sequential Forward Selection and this can be an off-line computation process. Then we can use the set of features to do the on-line identification of airplane.4. When the regular BP neural network is used, it may have poor performance because of bad convergence. So we choose improved BP neural network as classifier, which combines momentum and adaptive learning rate, and improve the learning and recognition efficiency.