基于特征增量的新类识别方法研究

在分类应用的过程中,经常会出现新的类别,导致数据分布发生显著变化,使得原分类模型不再适用。如何识别新的类别使分类模型能适应其出现已经成为一个亟需解决的问题。本文提出基于特征增量的SVDD(支持向量数据描述)新类识别方法。该方法在SVDD算法的基础上,通过增加新特征,扩大特征空间维度从而提高模型对于新类的识别能力。在多个数据集上的实验结果表明,该方法能有效识别新类,使更新后的模型具有更高的准确度。

New Class Recognition Method Based on Feature incremental

In classification tasks, new classes sometimes emerges, which makes the distribution change significantly and current classification models invalid. How to identify new classes has become an urgent problems. In this paper, a method based on feature incremental is proposed to recognize new classes. This method, which is based on SVDD algorithm with adding new features, expands the dimension of feature space so as to improve the model recognition for new class. The results gathered from multiple data have convincingly demonstrated that effective recognition and more accuracy of the new model.