中文摘要：

为实现无人作战飞机（UCAV）认知导航的高鲁棒性特征点提取，提出一种基于自适应非极大值抑制（ANMS）的多元量化Hessian—Affine迭代式尺度不变特征变换（SIFT）方法。针对认知导航对特征点均匀分布的需求，提出基于ANMS的初始特征点优选算法。为确保特征点的仿射不变性，利用引入迭代调节因子的Hessian-Affine迭代算法估计仿射不变区域，并在对应归一化圆形区域进行主方向确定以及圆形描述子生成。针对模拟特征序列分布不均匀、正确匹配率不高的缺陷，采用多值量化与比特抽取结合法对模拟特征序列进行多元量化，并且分析验证了该方法的优越性能。仿真结果表明，本文方法具有较高的正确匹配率，具有旋转不变性和尺度不变性，其抗噪性能提高了10dB，并且在大视角变化范围内具有较优的抗仿射性能。

英文摘要：

In order to extract high robust keypoints for the cognitive navigation of unmanned combat aerial vehicle （UCAV）, a method named multi-quantifying Hessian-Affine iterative scale invariant feature transform （SIFT） with adaptive non-maxi- mum suppression （ANMS） is proposed. Considering the demand of well-distributed keypoints, an optimization arithmetic based on ANMS is first presented to choose the keypoints. In order to ensure the chosen keypoints＇ affine invariant property, a Hessian-Affine iterative arithmetic with an iterative adjusting factor is used to estimate the affine invariant regions, and then the main orientation assignment and cycle descriptor are further realized in the corresponding normalized cycle regions. In view of the deficiencies of analog feature vectors in balanced distribution and correct matching score, a method combining the multiple value quantizatien and reshaping operation is presented to quantify the analog feature vectors. The analysis and simulation results verify this quantifying method as having better properties. Simulation results prove that the method proposed in this paper has higher correct matching score. It is invariant to image rotation and scaling. It can improve the anti- noise property by over 10 dB, and it possesses robust affine invariant property within a large visual angle range.