基于组合不变矩与神经网络的卫星目标识别

Recognition of Satellite Targets Based on Combined Invariant Moments and Artificial Neural Network

摘要:

首先介绍了Hu提出的不变矩和Jan Flusser提出的仿射不变矩,在不增加特征向量维数的前提下,将不变矩的部分分量与仿射不变矩组合成新的特征向量——组合不变矩。利用组合不变矩提取图像中的不变特征向量,并结合基于人工神经网络的分类器,较好的实现了对卫星目标的分类识别。

Abstract:

Firstly, the theory about Hu’s invariant moments and affine invariant moments was introduced. Then, the first four components of Hu’s invariant moments were combined with affine invariant moments to create a new invariant moments “combined invariant moments” without increasing the dimension of the feature vectors. Combined invariant moments were extracted as the feature vector of the satellite target. Finally, with the help of classifier, which was based on the Artificial Neural Network, the target classification in the satellite image has been implemented successfully.