基于非负稀疏表示的多标签学习算法.doc


基于非负稀疏表示的多标签学习算法#
陈思宝1,2,徐丹洋1,2,罗斌1,2*
(1. 安徽大学计算机科学与技术学院,合肥 230601;
5
10
15
20
25
30
35
40
2. 安徽省工业图像与分析重点实验室,合肥 230039)
摘要:为提高多标签数据分类性能,提出基于非负稀疏表示多标签学习算法。首先基于
LASSO 稀疏最小化方法,将测试样本用训练样本集进行非负稀疏线性重构,得到稀疏的非
负重构系数;然后根据非负重构系数计算测试样本的各个标签隶属度;根据隶属度的排序完
成多标签分类。提出一个全局最优的迭代更新求解算法及其相应的分析与证明。该算法保留
了更多的重构后图像的信息。在多标签分类识别的实验结果中显示所提出的方法比经典的
ML-SRC 和 ML-KNN 方法性能更优。
关键词:多标签学习;非负稀疏表示;LASSO 稀疏最小化;非负重构
中图分类号:
A Multi-Label Learning Algorithm Based on Non-negative
Sparse Repre-sentation
CHEN Sibao1,2, XU Danyang1,2, LUO Bin1,2
(1. College puter Science and Technology, Anhui University, HeFei 230601;
2. Key Laborary of Industrial Image Processing and Analysis of Anhui Province, HeFei 230039)
Abstract: To improve the multi-label data classification performance, a new multi-label learning
algorithm based on non-negative sparse representation is proposed. Firstly, based on LASSO
sparse minimization, a testing sample is linearly reconstructed by the training samples with
non-negative sparse coefficients. Then, using the sparse non-negative reconstruction coefficients,
the memberships of all labels on the testing sample are calculated. Finally, the classification is
achieved by ranking these memberships. In addition, an iterative update global optimization
algorithm and its associated theoretic analysis are provided to solve the proposed model, which
reserves more