面向弱点目标检测的规则集创建研究

项目名称： 面向弱点目标检测的规则集创建研究

项目编号： No.61303080

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： 张海英

作者单位： 厦门大学

项目金额： 23万元

中文摘要： 作为自动目标识别领域中的难点问题,弱点目标的检测一直受到研究者们的关注。为了提高算法的性能和通用性，研究日渐呈现多技术手段综合的趋势，但是却以增加复杂度和降低实时性为代价。本项目提出构建"规则集"的设想，以图像挖掘为理论基础，通过空间和时间的多层面挖掘工作，建立与目标关联的"组合特征"。以此为基础，形成"规则"，从而将传统的检测过程改变为与规则匹配的过程，减少了中间环节带来的漏检和虚警问题。针对仿真图像库，规则集的构建经历了特征生成、特征挖掘、规则优化以及规则验证等训练过程。该"规则集"的建立，相比于基于估计理论的时频滤波器无需已知目标的运动方程；同时，也克服了基于背景抑制方法的特征单一易于衰减的缺陷，为解决检测算法的通用性和实时性提供了一条新思路。

中文关键词： 红外小目标；聚类分析；图像分割；SURF；回溯

英文摘要： As the hard problem of ATR (Automatic Target Recognition), the detection of dim point targets has aroused many researchers' attentions. In order to improve the performance and the universality of the algorithms, the integrated technique has been a research trend; however, it is at the cost of increasing the complexity of the algorithm and lowering the real-time. In this project the idea of creating "Rule set" is presented and it aims to create "compound features" associated with the target by carrying out multi-level mining based on image mining (IM) theory. Consequently, with the forming of the "rules" the traditional detection process is change to the matching process with the rules and it redeuce the missing detection and false alarm.Based on the simulated image database, the creation of rule set experience the training process including feature creation, feature mining, rule optimization and rule verification. Compared with the time-frequency filtering based on estimation theory the foundation of the "rule set"has the superiority of without knowing of the target dynamic; and at the same time, it also overtakes the performance decline of the background suppression methods caused by single feature. So, it provides a new scheme to resolve the universality and real-time of the algorithm.

英文关键词： infrared small target；clustering analysis；image segementation；SURF；backtracking