A novel morphology domain description method for visual one-class classification

For many large sample size one-class classification problems, most existing methods fail due to the requirement lengthy execution time and large memory space. To solve these problems, a novel method referred to as Morphology domain description (MDD) is proposed by employing the concepts of Mathematical Morphology. First, the sample space is divided into blocks. Then, training samples are put into these blocks in terms of the values of their features. The block which contains at least one sample is defined as the object block, while the block without any sample is defined as the background block. Next, morphological closing and opening operations are applied to these blocks. Finally, the object blocks corresponding to the morphological operation result are considered as the domain description of the target class. A series of experiments are conducted using artificial datasets and real-world datasets to evaluate the performance of MDD. Besides, a practical example regarding aeroengine gas path condition monitoring is also conducted to demonstrate the efficiency of proposed method. The results show that the MDD is an excellent method with good classification accuracy, especially less execution time.