为了分析多类支持向量机(Multi-category support vector machines,M-SVMs)的推广性能,对常用的M-SVMs算法加以概述,推导、总结了理论推广误差公式。对于给定的样本集,可以设计合理的编码来提高ECOCSVMs的推广性能,通过构造合理的层次结构来提高H-SVMs推广性能,其余M-SVMs算法的推广性能均取决于样本空间。研究结果为有效使用M-SVMs提供了依据,为改进M-SVMs指明了方向。
Conventional seismic exploration,mostly based on reflection theory,hardly has accurate imaging results for disaster geologic bodies which have small scale,steep dip,or complex structure.In this paper,we design two typical geologic models for analyzing the characteristics of scattered waves in mines for forward modeling by finite difference and apply the equivalent offset migration（EOM）and EOM-based interference stack migration methods to mine prospecting.We focus on the analysis of scatted imaging’s technological superiority to reflection imaging.Research shows：1）scattered imaging can improve fold and make the best of weak scattered information,so it shows better results than post-stack migration imaging and 2）it can utilize the diffraction stack migration method-based ray path theory for mine seismic advanced prediction,so it provides an new efficient imaging method for improving resolution of mine seismic exploration.
