| 题名 | 类星体识别方法研究 |
| 作者 | 吴永东 |
| 学位类别 | 工学博士 |
| 答辩日期 | 1997-06-01 |
| 授予单位 | 中国科学院自动化研究所 |
| 授予地点 | 中国科学院自动化研究所 |
| 导师 | 马颂德 |
| 学位专业 | 模式识别与智能系统 |
| 中文摘要 | 本文分析了类星体(QSO)光谱特征，建立了一个识别系统模型。在这个模型 中，渗透着类星体光谱知识，综合应用了多尺度形态滤波、空间选择性滤波、弹 性匹配技术以及证据理论知识，并且提出了以下见解： 1．针对平移不变且单调增滤波器，Maragos通过构造它的滤波基，将它表示为一 系列膨胀算子的交(或腐蚀算子的并)，本文指出基底维数可能是无限的，甚至 不可列。 2．提出了一种形态滤波器，它是一种膨胀与腐蚀的复合算子，但是二者尺度不一 样、并且将它应用于特征提取。膨胀(腐蚀)后的曲线可能存在奇异点一导数不 存在的点，为了克服这种缺陷，分析了滤波后曲线的形成只与原始曲线和滤波 核的接触点有关，设计了另外一种自适应膨胀(腐蚀)滤波器，这种滤波器具有 与形态开（闭）运算类似的结果，但运算比它简单得多。本文设计的两种滤波器 都是平移不变而且单调增的。 3．对于一维函数的极值点数，本文指出多尺度膨胀(腐蚀)具有单调性，并且分析 了极值点的产生条件，但是对于二维以上函数，单调性结论不成立。另外，比 较了多尺度高斯滤波器与多尺度膨胀(腐蚀)两者的指纹图。不仅如此，还指出 了极值点数随尺度增加而减少的速率规律——（平均）单调率呈减少趋势，最后为 0。单凋率特性叮用于指导滤波尺度序列的选择。同时，将单调率概念推文到 高斯滤波器的尺度选择。 4．空间选择性滤波方法能综合不同尺度下的滤波结果，本文指出滤波器应满足的 必要条件，改进和推广了Xu方法应用范围。 5．针对证据理论一个很大的不足，即只能用于独立证据之间的组合，我们引入了 解耦算子，将两个相关证据变换成三个独立证据，并且用于推理。 |
| 英文摘要 | Based on QSO spectrum features, a Quasar Stellar Object Recognition model is built. In this system, Guided by the knowledge on QSO, multiscale morphology filtering, spatially selective filtering, elastic matching and Evidence Theory play a critical role. In these fields, we put forward several contributions as follows: l. Maragos proposed any operator with translation invariant and increasing can be represented by intersection of dilation or union of erosion, and gives an approach to select its basis. Here we draw a conclusion that the basis may be infinite, even nondenumerable. 2. Offering two morphological filters. One is the combination operator of dilation and erosion, but their scales are different. This operator can be used on feature extracting. Based on the newly irregular points located on dilated(eroded) curves, we designed another morphology filter by studying point of contact(POC). The filtering result is close to Opening(Closing), but this novel property is much more simply. Both of these two filters are translation invariant and increasing. 3. As a developing approach, multiscale analysis demands monotonicity. In this paper, for ID curves with only limited non-differentiable points, multiscale dilation(erosion) has monotonicity -- the number of extreme decreases with scale increasing. Unfortunately, it does not hold for higher dimensions functions. The fingerprints contrast between dilation and Guassian is put forward. Furthermore, we propose that the decreasing rate almost slows down with scale increasing. Based on the decreasing rate property, the maximal scale can be selected. This concept can be extended to Guassian filter. 4.To combine the results from different scales filtering, the spatially selective filtering approach is a good way, which enhances features and remove noises. We expand Xu's approach and propose the necessary conditions for this filter. 5. As we know, there is an apparent weak on Evidence Theory, which is correct only when evidences is independent. Apparently this demand is seldom satisfied. But the novel de-coupling operator can make all evidences independent, thus Dempster Evidence Combination Formula can be used. |
| 语种 | 中文 |
| 其他标识符 | 413 |
| 内容类型 | 学位论文 |
| 源URL | [http://ir.ia.ac.cn/handle/173211/5673]  |
| 专题 | 毕业生_博士学位论文 |
| 推荐引用方式 GB/T 7714 | 吴永东. 类星体识别方法研究[D]. 中国科学院自动化研究所. 中国科学院自动化研究所. 1997. |
| 个性服务 |
| 查看访问统计 |
| 相关权益政策 |
| 暂无数据 |
| 收藏/分享 |
除非特别说明，本系统中所有内容都受版权保护，并保留所有权利。
修改评论