• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文学位 >Fuzzy-connective-based information fusion networks and their application to computer vision.
【24h】

Fuzzy-connective-based information fusion networks and their application to computer vision.

机译：基于模糊连接的信息融合网络及其在计算机视觉中的应用。

获取原文
获取原文并翻译 | 示例
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Computer vision involves a lot of decision making under uncertain circumstances. In this dissertation, we propose a new information fusion method for decision making. The proposed method is capable of aggregating and propagating degrees of satisfaction of criteria in a hierarchical network, where the various nodes in the network represent sub-criteria at different levels of abstraction. The inputs to each node are the degrees of satisfaction of criteria, and the output is the overall degree of satisfaction. The proposed approach is very flexible since the type of connective at each node can be chosen by training algorithms. The training methods treat a set of the fuzzy-set-connective-based aggregation operators as activation functions, and determine the desirable aggregation connectives and the associated values of parameters as in a back-propagation neural network. In addition, the training algorithms can detect redundancy and reduce (or eliminate) the weights connected to the input which provides redundant information. At the end of training, the network can capture a knowledge base in a structure of connectives, and the structure can be used for fusion and propagation of uncertainties generated from various information sources. Through an extensive simulation study ranging from simple logic operations to object recognition, we show the effectiveness of the proposed method for decision making, especially for computer vision. Even though the scheme is based on the fuzzy set theory, we believe that any type of uncertainty including probability and Demster-Shafer belief measure as well as possibility can be adopted in the proposed framework in a practical sense.
机译：在不确定的情况下，计算机视觉涉及很多决策。本文提出了一种新的决策信息融合方法。所提出的方法能够在分层网络中聚合和传播标准的满足程度，其中网络中的各个节点代表处于不同抽象级别的子标准。每个节点的输入是标准的满意度，而输出是整体的满意度。所提出的方法非常灵活，因为可以通过训练算法选择每个节点处的结缔类型。训练方法将一组基于模糊集的基于连接的聚合算子视为激活函数，并确定所需的聚合连接词和参数的关联值，就像在反向传播神经网络中一样。另外，训练算法可以检测冗余并减少（或消除）连接到提供冗余信息的输入的权重。在培训结束时，网络可以捕获连接体结构中的知识库，并且该结构可用于融合和传播从各种信息源生成的不确定性。通过从简单逻辑运算到对象识别的广泛仿真研究，我们证明了所提出的决策方法尤其是计算机视觉决策的有效性。尽管该方案基于模糊集理论，但我们认为，在提议的框架中，从实际意义上讲，可以采用任何类型的不确定性，包括概率和Demster-Shafer信念测度以及可能性。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号