An Approach for Signal Discrimination Based on Category-Preserving BoW

This study proposes a novel category-preserving bag-of-words model for 1D signals discrimination (CPBoW-1D). In the proposed CPBoW-1D, the Taguchi’s quality engineering method was used for developing a category-preserving distance (CPD) metric for executing a clustering process to generate category-preserving codewords (CPWs). This facilitated preserving the categorical information in the codebook and consequently increased the effectiveness of the discrimination process. The proposed mathod was verified through an experiment that involved discriminating musical instruments. The results indicated that the proposed framework demonstrated an average discrimination accuracy of 87.39% for 18 musical instruments, which is superior to other benchmark methods.