Improved Fuzzy Adaptive Membership Function Algorithm for High-Density Surface Electromyography Based Hand Gesture Classification

Gesture classification based on high-density surface electromyography (HD-sEMG) is a key research area for assisting prosthetic users in rehabilitation. The existing algorithms have practical applications, although their efficiency and convenience may still require improvement. This study proposes an improved fuzzy adaptive membership function algorithm (IFAMF) to enhance classification accuracy and system robustness. Following data preprocessing, four types of features were extracted from different perspectives. During feature fuzzification, a reinforcement learning algorithm was employed to adaptively select the membership function type, with parameters determined via gradient descent, and these parameters were further refined using particle swarm optimization. Five well-established and reliable classifiers were selected for classification, and their performance was evaluated using a publicly available dataset. The experimental results demonstrated that under single-feature classification, the accuracy improved from 78.89% to 88.09%, with an average increase of 9.20%. Under the optimal feature combination, the final average accuracy reached 92.69%, with performance improvements across the classifiers ranging from 3.08% to 10.36%. These findings validate the superiority of the proposed method and suggest a promising direction for its integration with more advanced classifiers and feature extraction techniques to achieve more precise and intelligent prosthetic control.