Hand Gesture SEMG Signals Recognition Based On ART2 Neutral Network

The surface electromyographic (SEMG) signal caused by voluntary muscle contraction or passive muscle stimulation is recorded non-invasively using surface electrodes directly located on the skin, it can be considered as the compound of the superposition of motor unit action potential (MUAP) trains generated by the activated motor units and various noise, the surface potential distribution is influenced by the volume conductor and the signal acquisition equipments. Surface Electromyography, as an important biological signal, has been used widely in bionics, biofeedback, sports medicine and rehabilitation engineering. Recently, with the development of human computer interaction, hand gestures recognition based on action SEMG signals has become one of key research directions. SEMG-based hand gesture recognition technology has been widely researched and applied in rehabilitation, interactive human-computer interfaces, and virtual input devices as a supplementary or an alternative interaction modality.A method of hand gestures recognition based on the adaptive resonance neutral network 2 (ART2) was proposed to improve the adaptive capacity and stability in this dissertation. In order to explore the classification role of ART2 for hand gesture SEMG pattern, we conducted same-user and multi-user recognition experiments on 8 subjects and 8 kinds of hand gestures. The experimental results demonstrated that the ART2 classifier has the advantage of rapid recognition for hand gesture SEMG signals with high adaptability and stability. This research is meaningful for the development and realization of the real-time gesture-based myoelectric control system.The main work and achievement of the dissertation could be presented as the follows:1. Based on the comprehension and analysis of SEMG signals, we designed the collection experiment of multi-channel SEMG signals, and proposed effective SEMG signals processing algorithms including SEMG signals pre-processing, hand gesture action activities detection, and feature extraction. A 20-400Hz band-pass filter was used for the raw SEMG signal preprocessing to suppress noise and eliminate baseline drift. Moving average method was used for the detection of the activities related to hand gesture action. Normalized MAV, as well as AR model coefficients, were used to extract SEMG signal features.2. Adaptive resonance neutral network 2 (ART2) classifier was investigated and developed for the pattern recognition of hand gesture SEMG signals based on the research and analysis of ART2 neutral network theory and methods of hand gestures recognition. In view of the problems of efficiency and stability existing in hand gestures SEMG signals recognition, a method of hand gestures recognition based on ART2 was proposed. The same-user experimental results shows that, compared with the back propagation (BP) network classifier, ART2 classifier can effectively improve the recognition rate, reduce the signal processing time, and is not sensitive to individual differences. Multi-user experimental results also indicated that ART2 classifier has good adaptive capacity and stability in discriminating hand gestures SEMG signals.