项目名称: 融入时空关系联合判别学习的地基云图序列分类方法研究
项目编号: No.61501327
项目类型: 青年科学基金项目
立项/批准年度: 2016
项目学科: 无线电电子学、电信技术
项目作者: 刘爽
作者单位: 天津师范大学
项目金额: 19万元
中文摘要: 地基云观测技术对气候研究,天气预报,国民经济以及社会服务等工作具有重要的意义。地基云图分类技术是自动云观测的核心,本项目针对地基云图序列的分类问题开展研究。本项目拟在以下三个方面开展研究:首先,研究集合张量的局部特征提取方法,使得到的特征保留对分类有益的时空信息;其次,研究基于联合判别学习的特征表示方法,通过同时学习具有判别性的编码向量和词典,提高特征的表示能力;最后,研究基于概率一致性的特征聚合方法,使其在聚合过程中对环境变化鲁棒。本项目的研究可以提高地基云图序列分类方法的实用性,同时也为图像、视频、动态纹理分类提供新的研究方法。
中文关键词: 图像特征提取;图像分类;图像特征;纹理特征;地基云图序列
英文摘要: Ground-based cloud observation technology is of great significance for climate research, forecasting, national economy, social services etc. and as such cloud classification is an important technique. This project studies the issue of classification based on cloud sequences. We focus on the following three aspects. First, we propose an ensemble tensor method to extract the local features, which can reserve the useful spatio-temporal information for classification. Second, we study the joint discriminative learning method for feature representation. We improve the ability of feature representation by simultaneously learning discriminative coding vectors and dictionary. Finally, we study the probability consistency for pooling stage, which is robust to the environmental change. This project will enhance the accuracy of ground-based cloud sequence classification. It also provides a novel method for image, video and dynamic texture classification.
英文关键词: image feature extraction;image classification;image feature;texture feature;ground-based cloud sequences