摘要
由于事物的多样性、现代战争的欺骗性和破坏性,军事目标经常会发生部分外形改变的情况,基于几何散列表技术可以有效解决这个问题,但在特征库中已知目标不满足360度全姿态角时,该方法的识别效果下降。为了提高正确识别率,提出一种基于D-S证据理论融合的识别方法,并详细探讨了在贝叶斯结构和准贝叶斯结构下基本概率赋值的构建。利用MSTAR数据对该方法进行了仿真实验,验证了该方法的有效性和可行性。
Because of the diversity of things and the fraudulence or devastation of modern war, military targets are often partially distorted. Geometric Hashing technology can effectively resolve this problem, but when the known targets in training data set don't satisfy with the condition of 360 azimuths, the effect of recognition degrades. A new fusion method based on D-S evidence reasoning was proposed to improve the recognition probability, and the construction of basic probability assignment function under Bayes frame and quasi-Bayes frame was also discussed. Experimental results with MSTAR dataset show that this fusion method is effective and feasible.
出处
《系统仿真学报》
EI
CAS
CSCD
北大核心
2007年第9期2053-2057,2112,共6页
Journal of System Simulation
基金
国家自然科学基金(60541001)
全国优秀博士学位论文作者专项基金(200443)