基于视觉注意机制的迷彩伪装评价模型和检测方法研究

项目名称： 基于视觉注意机制的迷彩伪装评价模型和检测方法研究

项目编号： No.61472115

项目类型： 面上项目

立项/批准年度： 2015

项目学科： 计算机科学学科

项目作者： 薛峰

作者单位： 合肥工业大学

项目金额： 82万元

中文摘要： 迷彩伪装在军事作战中具有重要的应用价值，研究迷彩伪装效果的评价模型和检测方法具有重要的理论意义和应用前景。本课题将人类视觉注意机制引入迷彩的评价和检测研究，旨在建立科学、有效的迷彩评价模型和检测方法。为此，（1）首先探索研究人类视觉注意机制的生物学原理，揭示探索人眼对迷彩图案的显著性感知和迷彩图案识别的内在关系，建立数码迷彩的评价理论模型；（2）然后，深入分析数码迷彩的本质特征，采用基于深度机器学习方法挖掘可以刻画迷彩纹理本质特征的高级抽象特征；（3）基于人类自底向上的视觉注意模型，研究基于多任务学习特征融合算法的迷彩目标评价模型，并创新性地提出从显著特征出发，研究优质迷彩模式挖掘推荐；（4）最后，借鉴人眼从顶向下的目标搜索机制，研究基于隐马尔科夫模型的迷彩目标检测方法。

中文关键词： 数码迷彩；视觉注意机制；评价模型

英文摘要： Camouflage is becoming more and more useful in military application, and it has important theoretical significance and application prospects to establish evaluation model and detection method for camouflage. Based on human vision mechanism, we attempt to build a scientific and effective evaluation model, and propose a fast camouflage detection method. To make this come true, (1)Firstly, we do research on biological principle of focus mechanism of human vision, and explore the essential relation between saliency of camouflage and identification mechanism of camouflages to build the evaluation model of digital camouflage. (2)Secondly, based on the in-depth analysis of the essential characteristics of digital camouflage,we propose a novel Deep Learning algorithm to automatically train these features that have good desciption ability for the camouflages, which are called High Level and Abstract Feature.(3) Thirdly,based on Bottom-Up focus attention theory, we propose a novel camouflage evaluation model based on the multi-task feature learning.Moreover, we can do camouflage pattern recommendation by using the dictionary learning, which can be useful for camouflage designing. (4)Finally, with the help of target searching mechanism of human eyes, we propose an effective camouflaged target detection algorithm via HMM.

英文关键词： digital camouflage;human visual mechanism;evaluation model