Myoelectric signals (MES) are electrical impulses generated during muscle contractions. Advanced signal processing enables their use for precise prosthesis control in rehabilitation, gaming, interactive simulations, and robotics. The inventions introduce an AI-based method to optimize detection and classification of MES segments for specific gestures, inspired by fuzzy audio search and Convolutional Neural Networks (CNNs), like how music recognition services recognize a song from a short, imperfect snippet!
Physical Sciences
Reference
y24000
IP right year
IP status
Three patent applications filed (2025, 2025, 2026)
Applicant

Challenge and innovation

Conventional AI-based MES classification depends heavily on the number of features extracted from recorded signals. Commercial sensors often provide low‑resolution data due to hum and noise reduction. Classifiers are usually chosen empirically and require time‑consuming training, while online learning during operation is rarely feasible in such frameworks.

The inventions fuse data from multiple sensors for richer, more reliable MES, treating signals as music and modeling features via Mel Frequency Cepstral Coefficients (MFCC). Fuzzy audio search algorithms classify MES, which can be stored long term and compressed (e.g., MP3) to reduce data. Cloud-based CNN processes multi-user EMG to detect gestures, learn temporal patterns, and exploit cross-user correlations, so each prosthesis benefits from shared transfer learning. With Wi‑Fi, powerful cloud CNNs classify; otherwise, an on‑device CNN on the prosthesis is used.