L1正则化的贝叶斯分类器研究及在陨石坑检测中的应用

发布时间：2018-02-27 16:15

本文关键词： 陨石坑检测监督分类贝叶斯分类器特征选择　出处：《合肥工业大学》2012年硕士论文　论文类型：学位论文

【摘要】：陨石坑是陨石体高速运动撞击地表或其他天体表面之后所形成的坑穴。这种坑穴的重要性源于其丰富的信息，这些信息描述了它们的分布以及其产生的形态。因此，陨石坑在行星探测方面得到了广泛及深入的研究和应用。然而由于在一个陨石坑生命周期存在侵蚀、掩埋、覆盖以及转化问题，检测圆形状的陨石坑比单纯从图像检测完整圆形更困难。陨石坑大小的分布呈现规律，即服从幂律分布：大的很容易被检测出来的陨石坑是稀少的，小的亚公里(sub-kilometer)陨石坑是大量存在的。监督分类算法对于小陨石坑检测比较有效，因此可以充分利用领域知识以标注训练集的形式来指导分类算法。贝叶斯分类模型是一种建立在贝叶斯统计学习和贝叶斯网络基础上的监督分类算法，具有坚实的数学基础和清晰的模型可解释性等优点，同时也是数据挖掘和机器学习中的研究热点。本文针对监督学习下的贝叶斯分类器展开研究，从而提出一种新的有效的贝叶斯分类算法，并将其应用于陨石坑的检测过程中。本文的主要内容包括如下几个方面：（1）对陨石坑检测的研究背景及国内外的研究现状进行了论述，然后对描述了常用的陨石坑检测方法。（2）在研究贝叶斯分类器的过程中，，提出了一种新的贝叶斯分类方法，该算法需要学习一组有效的属性变量拓扑序列。通过实验与数据挖掘中常用的其他分类算法进行比较，该方法取得很好分类效果。（3）由于陨石坑检测数据的高维性，因此一般的贝叶斯分类器不能用于对其进行分类。本文特将提出的贝叶斯分类器进行了改进使其能应用于高维数据中。通过实验表明，本文提出的贝叶斯分类器即L1正则化的贝叶斯分类器应用在自动检测陨石坑框架中是有效可行的。
[Abstract]:The crater is a meteorite hit the ground after the high-speed movement of body or other celestial body surface formed by the hole. The hole's importance stems from its rich information, this information describes their distribution and shape of the crater. Therefore, it has got extensive and in-depth research and application in planetary exploration. However, due to a crater in the life cycle of erosion, burial, coverage and transformation, crater detection circular shape than from complete circular image detection more difficult. The crater size distribution is regular, i.e. obeying power-law distribution: very easily detected crater is rare. Small sub kilometer (sub-kilometer) crater is substantial. Supervised classification algorithm is effective for detecting small craters, so it can make full use of domain knowledge in the form of labeled training sets to guide the classification Algorithm.
Bias classification model is based on a statistical learning Bias and Bias network based on supervised classification algorithm has a solid mathematical foundation and a clear explanation of the advantages of the model, mining and research hotspot in machine learning and data. This paper studied Bias classifier under supervised learning, and put forward a the new Bias classification algorithm, and its application in the detection process of the crater.
The main contents of this paper include the following aspects:
(1) the research background of crater detection and the current research status at home and abroad are discussed, and then the commonly used methods for detecting craters are described.
(2) in the process of Bias classifier, Bias proposed a new classification method, the algorithm needs to learn a set of valid attribute variable topology sequence. Through the comparison of experiment and data mining in other commonly used classification algorithms, this method has good classification effect.
(3) due to the high dimensionality of the crater detection data, and therefore can not be used for general Bias classifier to classify it. This paper will put forward the Bias classifier is improved so that it can be used in high-dimensional data. Experimental results show that the Bias classifier is presented in this paper is the application of L1 regularization Bias classifier is effective feasible in automatic detection of craters in the framework.

【学位授予单位】：合肥工业大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP18;P185.83

【相似文献】