A Novel Feature Extraction for Robust EMG Pattern Recognition

Abstract: Varieties of noises are major problem in recognition of Electromyography (EMG) signal. Hence, methods to remove noise become most significant in EMG signal analysis. White Gaussian noise (WGN) is used to represent interference in this paper. Generally, WGN is difficult to be removed using typical filtering and solutions to remove WGN are limited. In addition, noise removal is an important step before performing feature extraction, which is used in EMG-based recognition. This research is aimed to present a novel feature that tolerate with WGN. As a result, noise removal algorithm is not needed. Two novel mean and median frequencies (MMNF and MMDF) are presented for robust feature extraction. Sixteen existing features and two novelties are evaluated in a noisy environment. WGN with various signal-to-noise ratios (SNRs), i.e. 20-0 dB, was added to the original EMG signal. The results showed that MMNF performed very well especially in weak EMG signal compared with others. The error of MMNF in weak EMG signal with very high noise, 0 dB SNR, is about 5-10 percent and closed by MMDF and Histogram, whereas the error of other features is more than 20 percent. While in strong EMG signal, the error of MMNF is better than those from other features. Moreover, the combination of MMNF, Histrogram of EMG and Willison amplitude is used as feature vector in classification task. The experimental result shows the better recognition result in noisy environment than other success feature candidates. From the above results demonstrate that MMNF can be used for new robust feature extraction.

• The paper introduces novel frequency-based features (MMNF and MMDF) that reduce error margins to 5-10% under high noise levels.
• The experimental framework validates these features across SNRs from 20 dB to 0 dB using varied gestures recorded via Ag-AgCl electrodes.
• Combining MMNF with complementary metrics forms a robust feature vector that significantly enhances classification accuracy in EMG-driven prosthetics.

A Novel Approach to Feature Extraction for EMG Pattern Recognition

The paper presents a significant exploration in the domain of Electromyography (EMG) pattern recognition, focusing on the development of features robust to White Gaussian Noise (WGN). Electromyography is crucial in fields such as assistive technology and rehabilitation, where precise signal interpretation is essential for controlling prosthetics and assistive devices. Noise, specifically WGN, poses a persistent challenge in cleanly extracting features from EMG signals due to its broadband and random frequency nature.

The authors introduce two novel frequency-based features: Modified Mean Frequency (MMNF) and Modified Median Frequency (MMDF). These metrics deviate from traditional calculations by operating on the amplitude spectrum instead of the power spectrum, resulting in enhanced robustness against noise. Unlike earlier models which may rely extensively on preprocessing noise removal techniques, the MMNF and MMDF features are resilient to WGN, potentially eliminating the need for prior noise filtration.

Experimental Framework

The robustness of these features is validated through comprehensive experimentation involving both strong and weak EMG signals subjected to varying levels of WGN, from 20 dB to 0 dB SNR. The authors conduct assessments on a dataset that includes gestures such as wrist flexion and hand opening, recorded using widely adopted Ag-AgCl surface electrodes. Additionally, the dataset encompasses a variety of feature extraction methods, spanning both time and frequency domains.

Key Findings and Implications

The findings suggest that MMNF, in particular, significantly outperforms existing features under high noise conditions. With an error margin as low as 5-10% under 0 dB SNR conditions and an average error of 6% in strong EMG signals, MMNF demonstrates remarkable noise resilience. These error rates starkly contrast with other tested features, which commonly exceed an error of 20%. Moreover, when MMNF is combined with Histogram of EMG and Willison amplitude, it forms a robust feature vector that enhances classification accuracy in noisy environments.

The implications of this research are notable for the field of biomedical signal processing. The ability to extract reliable features from noisy EMG signals without extensive preprocessing broadens the applicability of EMG devices in real-world settings, where signal noise is unavoidable. This advancement augments both the reliability and efficiency of EMG-controlled prosthetics, potentially improving user experience and device performance.

Future Directions

Although MMNF shows promising robustness against WGN, further research might explore its integration with other machine learning classifiers to optimize classification performance. Additionally, exploring MMNF's adaptability across diverse motion types and noise environments would extend its applicability and ensure its utility in varied clinical and engineering applications.

Overall, this study contributes a substantial advancement in developing noise-resilient EMG feature extraction methods, offering practical benefits for enhancing human-machine interface technologies. As advancements continue in this field, the resilience and adaptability of feature extraction methodologies like MMNF will play a pivotal role in advancing the capabilities and precision of EMG-based control systems.