Combining Transfer Learning and Genetic Algorithms for Real-Time Gesture Recognition using EMG Signals

This paper proposes a hybrid method using Genetic Algorithm and Transfer Learning in Deep Neural Networks for gesture classification using surface EMG (sEMG) signals. The goal is to develop efficient Convolutional Neural Networks (CNN) that can be implemented in real-time situations to recognize gestures with a small number of EMG sensors. By "efficient" we mean good classification accuracy, short data acquisition time, and short CNN training time. The dataset to verify the performance of the proposed algorithm is created in our laboratory. We compare the performance of CNNs trained (a) from scratch, (b) using transfer learning, and (c) using transfer learning with GA-optimized parameters. The performance of the proposed hybrid method shows a good performance.