基于局部Haar和PHOG特征的月球撞击坑综合检测方法

发布时间：2018-05-11 19:23

本文选题：撞击坑检测 + 塔式梯度方向直方图特征　；参考：《中国科学:物理学力学天文学》2013年11期

【摘要】：基于月貌图像的撞击坑的检测需要采用合理的特征选择和机器学习策略,我们提出了一种基于区域局部灰度和梯度分布特征与机器学习方法相结合的撞击坑检测方法.这种方法将Haar特征与AdaBoost结合,使候选撞击坑区域的定位更加快捷,采用局部区域的塔式梯度方向直方图(PHOG)与高效的支持向量机学习算法相结合的方法用来精确地对撞击坑候选区域进行分类.考虑到Haar特征数的繁多而采用AdaBoost作为特征提取和分类方法,并由于PHOG特征的每一项都对分类起作用,将撞击坑区域统一预处理为不含阴阳面的各向梯度向量基本一致的圆形模糊边界,使圆形撞击坑的正样本特征具备更多的稳定性.文中还讨论了几种特征和分类方法的机理和集成,以及参数调整对撞击坑检测的效率分析.
[Abstract]:The detection of impact craters based on lunar images requires reasonable feature selection and machine learning strategy. We propose a new method of impact crater detection based on the combination of regional local gray and gradient distribution features and machine learning methods. This method combines Haar features with AdaBoost, which makes the location of candidate crater area faster. The method of combining the tower gradient direction histogram (PHOG) of the local area with the efficient support vector machine learning algorithm is used to classify the candidate area of impact crater accurately. Considering the large number of Haar features, AdaBoost is used as the feature extraction and classification method, and because every item of PHOG features has an effect on the classification, The region of impact crater is pretreated as a circular fuzzy boundary with almost identical gradient vectors without yin and yang, so that the positive sample features of circular impact crater have more stability. The mechanism and integration of several features and classification methods are also discussed, as well as the efficiency analysis of impact crater detection by adjusting parameters.
【作者单位】：华东交通大学基础科学学院;
【基金】：国家自然科学基金(批准号:61262031) 中国科学研究院地理研究院科研项目(编号:YQZX-HT-KY-QT-20120119-1)资助
【分类号】：TP181;P184.5

【参考文献】