Evolved Fuzzy Min-Max Neural Network for Unknown Labeled Data and its Application on Defect Recognition in Depth
- PDF / 1,156,270 Bytes
- 21 Pages / 439.37 x 666.142 pts Page_size
- 13 Downloads / 168 Views
Evolved Fuzzy Min-Max Neural Network for Unknown Labeled Data and its Application on Defect Recognition in Depth Yanjuan Ma1
· Jinhai Liu2 · Yan Zhao1
Accepted: 10 October 2020 © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Pattern classification is one of the most important issue in the data-driven application domains. Unlike the traditional unlabeled data, unknown labeled data refers to the testing data that cannot be classified into the existed category in this paper. How to learn the unknown labeled data is a crucial issue in the data classification. In this paper, an evolved fuzzy min-max neural network for unknown labeled data classification (FMM-ULD) is proposed. In FMM-ULD, the unknown labeled data handling process is designed. Moreover, in the unknown labeled data handling process, a decision function and a threshold function are designed. In addition, FMM-ULD can realize further correction for the unsatisfactory data classification of the known category. The experimental results using UCI benchmark data set show that FMMULD get good performance for handling the unknown labeled data as a general method. In addition, the application result on the pipeline defect recognition in depth shows that FMM-ULD is effective in handling the real-application unknown labeled data problem. Keywords Pattern classification · Fuzzy min-max · Neural network · Unknown labeled data
1 Introduction In recent years, pattern classification have been witnessed an increase in the use of various research domains, such as image, video analysis, computer vision [1,2], detect recognition and so on [3]. Pattern recognition methods can be divided into three groups: unsupervised learning, semi-supervised learning and supervised learning [4]. Unsupervised learning aims to learn useful representations from unlabeled data [5]. Semisupervised learning uses both of the labeled and unlabeled data in the training process [6,7]. Although all of the three methods have got great success in the pattern recognition, the
B
Yanjuan Ma [email protected]
1
School of Renewable Energy, Shenyang Institute of Engineering, Shenyang 110136, Liaoning, China
2
State Key Laboratory of Synthetical Automation for Process Industries, College of Information Science and Engineering, Northeastern University, Shenyang 110819, China
123
Y. Ma et al.
problem in the supervised learning is addressed in this paper. As the most widely used method, supervised learning aims to learn certain classification functions based on known training samples and their labels for pattern recognition [8,9]. In the traditional classification methods, the labeled data is used to train the classifier, and the testing data can be classified into the existed trained category. However, in the real world application, not all the testing data can be classified into the existed category, they may be the unknown labels. (In order to avoid the confusing with other unlabeled testing data, this kind of data is called unknown labeled data.) Unlike the traditional
Data Loading...
