基于半监督局部保持投影的磨粒图像特征降维

来源期刊:中南大学学报(自然科学版)2015年第8期

论文作者:张云强 张培林

文章页码:2937 - 2944

关键词:磨粒分析;局部保持投影;特征提取;Parzen窗;成对约束;半监督

Key words:wear particle analysis; locality preserving projection; feature extraction; Parzen windows; pairwise constrains; semi-supervised

摘    要:为有效提取磨粒图像的数字化特征,引入局部保持投影算法。针对局部保持投影在磨粒特征降维中的不足,提出一种基于Parzen窗和成对约束的半监督局部保持投影算法(PSS-LPP)。利用Parzen窗估计高维特征空间中样本的密度,然后根据各样本密度自适应调整邻域参数,并且充分利用样本的标签信息和实例约束生成成对约束集,进而指导投影权矩阵的构造,从而实现特征参数的半监督降维。将PSS-LPP应用于磨粒图像的纹理特征降维,研究结果表明:PSS-LPP对邻域参数初值和热核参数不敏感,降维性能比较稳定,磨粒识别精度明显提高。PSS-LPP可以更有效提取磨粒图像的低维特征。

Abstract: To effectively extract digital features of wear particle images, the locality preserving projection algorithm was employed. For the disadvantages of locality preserving projection for feature dimensionality reduction of wear particles, a semi-supervised locality preserving projection algorithm(PSS-LPP) based on Parzen windows and pairwise constrains was proposed. Parzen windows were utilized to estimate the density of samples in high-dimensional feature space, and the formation of samples of labels and constrains were employed to create pairwise constrains sets which guided the construction of the projection right matrix. Then, with the projection right matrix, semi-supervised dimensionality reduction of feature parameters was implemented. PSS-LPP was applied for texture feature dimensionality reduction of wear particle images. The results indicate that PSS-LPP is not sensitive to the original value of neighborhood parameter and the kernel parameter, thus it has very stable dimensional-reduction performance. The classification accuracy is improved obviously. PSS-LPP can extract low-dimensional features of wear particle images more effectively.

有色金属在线官网  |   会议  |   在线投稿  |   购买纸书  |   科技图书馆

中南大学出版社 技术支持 版权声明   电话:0731-88830515 88830516   传真:0731-88710482   Email:administrator@cnnmol.com

互联网出版许可证:(署)网出证(京)字第342号   京ICP备17050991号-6      京公网安备11010802042557号