Amputee walking mode recognition based on mel frequency cepstral coefficients using surface electromyography sensor

Walking mode recognition through surface electromyography (sEMG) sensors is an active field of smart prostheses technologies. This work presents the mel frequency cepstral coefficients of the nonlinear sEMG signals as an effective feature for the recognition of different walking modes. Principal component analysis and mutual information were used for feature reduction and optimum sEMG channel selection respectively. The proposed recognition system identifies five walking modes such as normal walking, slow walking, fast walking, ramp ascending and ramp descending. The proposed method was evaluated using 11 channels of lower limb sEMG signals recorded from six subjects including four able-bodied, one unilateral transtibial and one unilateral transfemoral amputee. Several classifiers were trained with a pool of collected data. The experimental result exhibits that the proposed system achieved the highest accuracy of 97.50% using support vector machine. The promising results of this work could promote the future developments of neural-controlled lower limb prosthetics.