Research On Instance Specific Sequential Feature Extraction Algorithms And Application

With the fast development of data collection techniques, lots of complicated objects are described and stored. As a consequence, accelerated feature learning approaches should be proposed for reducing the time costs in the intelligent data mining approaches. Existing feature learning, including feature selection and extraction methods mainly aim at reducing the number of features while leaving the properties of single instances unconsidered. In this work, we focus on instance specific extraction techniques which can further cut off the expense of feature extraction. Besides, we also bring forward a fast feature extraction based systematic solution for abnormal fabric detection. There are two major achievements in this paper as follows:1. We proposed an instance specific feature extraction technique (DFE), which assigns different feature extraction sequences for particular instances and can further reduce the expense of feature extraction. Empirical investigations reveal the effectiveness and efficiency of the proposed DFE method;2. We proposed a systematic solution for abnormal fabric detection in which the fast feature extraction techniques and feature selection are implicitly used. We use DFE method for feature selection in this system. Besides, a classifier refine approach is also applied for ensuring the detection performance.