一种遥感图像中军用飞机的识别方法

如有对接意向，请电话垂询010-68918340

一种遥感图像中军用飞机的识别方法

本发明涉及一种遥感图像中军用飞机的识别方法，其技术特征是：建立多尺度参考军用飞机模型库；然后提取在目标图像发生模糊的情况下仍保持不变的16个特征，包括六 个 H u 矩 M2 … M 7 、两 个 4 阶 不 变 矩 M 8 和 M 9 和 八 个 复 数 矩 R e( C 3 0 C 1 2 C 0 2 ) ，Re( C 40 C 12 C03 )， Re( C 50 C 02 C 03 )，Re(C50 C 12 C 04 )，Re(C 50 C 05 )；利用粗糙集中的属性频度算法去除冗余属性，降低飞机特征维数，建立新属性集；之后带入多类支持向量机分类器中进行训练，并对待识别飞机目标预测，最终输出飞机型号。有益效果：由于采用多尺度矩特征模型方法，可以适应各种尺度及模糊程度的飞机目标。采用粗糙集，能够减少属性数目的同时不影响分类性能，可提高识别速度；3、支持向量机是专门针对小样本情况，具有较高的泛化能力和较好的推广能力。

Recognizingmethod for military aircraft in remote sensing image

Theinvention relates to a recognizing method for a military aircraft in a remotesensing image. The recognizing method is technically characterized in that amulti-scale reference military aircraft model library is established; then 16features which remain unchanged under the condition that a target image isfuzzy are extracted, the 16 features comprise six Hu moments of M2…M7, twofour-order invariant moments of M8 and M9, and eight complex moments of Re(C<30>C<12>C<02>),Re(C<40>C<12><2>C<02>),Re(C<50>C<12>C<02><2>),Re(C<40>C<12>C<03>),Re(C<50>C<12><2>C<03>),Re(C<50>C<02>C<03>), Re(C<50>C<12>C<04>)and Re(C<50>C<05>); redundant attributes are removed byutilizing an attribute frequency algorithm in a rough set, aircraft featuredimensions are lowered, and a new attribute set is set up; and then the newattribute set is substituted into a classifier of a multi-class support vectormachine to carry out training, to-be-recognized aircraft targets are predicted,and finally aircraft models are output. The recognizing method has thebeneficial effects that since the multi-scale moment feature model method isadopted, the recognizing method can adapt to aircraft targets in various scalesand with various fuzzy degrees; the rough set is adopted, so that classifyingperformance is not affected while the number of attributes is reduced, and therecognizing speed can be increased; and thesupport vector machine is specially used for the condition of small samples,and has high generalization capacity and good popularization capacity.