基于深度学习框架的流媒体图像多标签分类

MULTI-LABEL CLASSIFICATION OF STREAMING MEDIA IMAGES BASED ON DEEP LEARNING FRAMEWORK

考虑到存在未知类别的大规模流媒体图像多标签分类问题,提出一种基于深度学习框架的多标签分类方法。为了检测图像中是否含有新类标签,提出一种递归类检测器,它通过对图像特征和多个标签之间的关系进行有效编码来学习。为提升方法对大规模数据集处理能力,通过假设新的类图像在特征空间中远离已知类,从而有效地实现分类器和检测器交替学习的批处理模式。实验结果验证了该方法对大规模未知类流媒体图像多标签分类有效性。

Considering the problem of multi-label classification of large-scale streaming images with unknown classes,a multi-label classification method based on a deep learning framework is proposed.To detect whether an image contains a new class of labels or not,a recursive class detector was proposed,which learned by efficiently encoding the relationship between image features and multiple labels.To enhance the method's ability to handle large-scale datasets,a batch mode of learning the classifier and detector alternately was effectively implemented by assuming that the new class media images were far away from the known classes in the feature space.The experimental results verify the effectiveness of the method for multi-label classification of large-scale unknown class streaming media images.