论文部分内容阅读
鉴于迁移成分分析(TCA)忽略了样本间的局部信息差异性,提出一种基于样本局部判别权重的加权迁移成分分析.首先,通过对局部近邻圆内样本分布情况的分析,为共享特征子空间中的每个样本均设计一个局部判别权重;然后,通过将局部判别权重与最大均值差异相结合构造的分布差异矩阵引入进TCA的目标函数中,从而体现样本对维持局部结构的贡献度差异;最后,结合联合分布调整和线性判别分析,使算法不仅能够同时缩小领域间的边缘分布差异和条件分布差异,而且能够提高算法的类间可分特性.36组跨领域图片数据集上的实验结果表明,所提算法能够获得65.67%的平均分类精度.
Since Migration Component Analysis (TCA) ignores the difference of local information between samples, this paper proposes a weighted Migration Component Analysis (MTA) based on local discriminant weights of samples.Firstly, by analyzing the distribution of samples within a circle of local neighbors, Then, the difference matrix of the contribution to the local structure is represented by introducing the matrix of the distribution variance, which is constructed by combining the local discriminant weight and the maximum mean difference, into the objective function of TCA. Finally, combined with the joint distribution adjustment and linear discriminant analysis, the algorithm can not only reduce the edge distribution difference and the conditional distribution difference in the field at the same time, but also improve the inter-class separability of the algorithm.Experimental results on the 36 sets of cross-field picture datasets The results show that the proposed algorithm can obtain an average classification accuracy of 65.67%.