项目名称: 基于视觉信息处理机制的启发式聚类算法研究
项目编号: No.61305070
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: 李春忠
作者单位: 安徽财经大学
项目金额: 23万元
中文摘要: 作为数据挖掘与机器学习的一个重要研究领域,数据聚类在自动化程度和数据集适应能力方面有很大的提升空间。为避免大量的先验指导及算法人为干预,本课题从聚类问题的本源入手,把聚类看作认知问题,通过模拟视觉认知系统精妙的工作机制构建出一种启发式聚类算法。课题从已构建的模型框架的完善和图像处理应用两个方面进行研究,主要包括:1)正向反馈初始滤波器的构建;2)反向反馈过程中的信息收集与利用;3)在图像分割领域的应用。算法融入了视觉信息反馈机制,并构建基于认知特征的、逐点不同的、智能化的感受野滤波器,通过感受野滤波器之间的相互关系来匹配任意类型的数据集聚类特征,从而达到最优的聚类效果。其最大优势是其智能性,在尽可能少的人为干预情况下让算法自动的去适应数据集。新算法中的反馈机制可充分利用图像数据中的结构化信息进行反馈,进而应用于基于语义的图像分割问题,有望为该领域提供一条新的解决途径。
中文关键词: 聚类;信息反馈;多尺度信息融合;岩石物理;方向一致性
英文摘要: As an important part of data mining and machine learning, clustering has much room to be improved in automation and adaptability to various data sets. To avoid much prior guidance and artificial involvement, the present study analyzes the root of clustering, and constructs a kind of hierarchical clustering algorithm through simulating the delicate working mechanism of visual system and regarding clustering as cognitive problem. The study proceeds the research in two aspects—the improvement of the established frame and the application in image processing, which mainly includes: 1) the construction of forward feedback filter; 2) the collection and employment of the information in backward feedback process; 3) the application in image segmentation. The algorithm builds intelligent receptive field filter which is based on cognitive features, and is different point by point. Then, the algorithm matches arbitrary types of clustering features of data sets through the interrelationship of receptive field filters, so as to achieve the optimized clustering results. The remarkable superiority of the model is its intelligence, which enables the algorithms, with as little artificial interference as possible, to automatically adapt to data sets with various structures in feature space. The feedback mechanism in the new algori
英文关键词: Clustering;Information feedback;Multi-scale information fusion;Petrophysical;Direction consistence