A comparative study of wavelet families for electromyography signal classification based on discrete wavelet transform

Abstract

Automatic detection of neuromuscular disorders performed using electromyography (EMG) has become an interesting domain for many researchers. In this paper, we present an approach to evaluate and classify the non-stationary EMG signals based on discrete wavelet transform (DWT). Most often researches did not consider the effect of DWT factors on the performance of EMG signals classification. This problem is still an interesting unsolved challenge. However, the selection of appropriate mother wavelet and related level decomposition is an essential issue that should be addressed in DWT-based EMG signals classification. The proposed method consists of decomposing a raw EMG signal into different sub-bands. Several statistical features were extracted from each sub-band and six wavelet families were investigated. The feature vector was used as inputs to support vector machine (SVM) classifier for the diagnosis of neuromuscular disorders. The obtained results achieve satisfactory performances with optimal DWT factors using 10-fold cross-validation. From the classification performances, it was found that sym14 is the most suitable mother wavelet at the 8th optimal wavelet level of decomposition. These simulation results demonstrated that the proposed method is very reliable for reducing cost computational time of automated neuromuscular disorders system and removing the redundancy information.