基于NSST混合理论的反恐图像目标识别方法研究

项目名称： 基于NSST混合理论的反恐图像目标识别方法研究

项目编号： No.61309008

项目类型： 青年科学基金项目

立项/批准年度： 2014

项目学科： 自动化技术、计算机技术

项目作者： 孔韦韦

作者单位： 中国人民武装警察部队工程大学

项目金额： 23万元

中文摘要： 非下采样剪切波变换（NSST）是对经典多尺度分析理论的重要发展，具有良好的结合性。NSST理论与相关理论相融合是开辟和发展新的图像处理方法的重要途径之一。针对反恐图像目标识别这一典型的信息处理难题，深入研究基于NSST混合理论的反恐图像目标识别方法。通过研究NSST理论与非负矩阵分解（NMF）理论、与新型神经网络（NN）理论、与感受野（RF）理论的融合机理，探索NSST理论的融合规律；通过研究基于NSST域改进型NMF理论、基于NSST域改进型NN理论、基于NSST域改进型RF理论，寻求发展NSST混合理论求解反恐图像目标识别问题的处理方法；通过突破NSST混合理论在反恐图像目标识别中的应用等关键问题，得到有效的基于NSST域改进型NMF理论、NSST域改进型NN理论、NSST域改进型RF理论的反恐图像目标识别方法。本项目预期研究成果对于提升我国的反恐处突能力具有重要的理论意义和应用价值。

中文关键词： 目标识别；非下采样剪切波变换；非负矩阵分解；神经网络；感受野

英文摘要： Non-Subsampled Shearlet Transform (NSST) is an important extension of the classic multi-scale analysis theories, and has good associativity. The combination between NSST theory and relevant theories is one of the important approaches to the inauguration and evolution of novel methods for image processing. To the typical problem of information processing of counter-terrorism image targets recognition (CTITR), the deep investigation into the techniques for CTITR based on NSST hybrid theories is carried out. By researching on the combination mechanism between the NSST theory and other three ones including nonnegative matrix factorization (NMF), novel neural network (NN), and receptive field (RF), the combination rules of NSST theory can be explored. By conducting an investigation into the theories which are based on NSST domain improved NMF, NSST domain improved NN and NSST domain improved RF, respectively, we can seek for the methods for solving the problems of CTITR based on NSST hybrid theories. With the breakthrough in the key issue of the applications of NSST hybrid theories in CTITR, three kinds of effective techniques for CTITR which are based on NSST domain improved NMF theory, NSST domain improved NN theory and NSST domain improved RF theory, respectively, can be proposed. The expected results of this rese

英文关键词： Target Recognition；Non-Subsampled Shearlet Transform；Non-negative Matrix Factorization；Neural Network；Receptive Field