A Single Platform for Classification and Prediction using a Hybrid Bioinspired and Deep Neural Network (PSO-LSTM)

• Published in: MĀPAN : journal of Metrology Society of India Vol. 37; no. 1; pp. 47 - 58
• Main Authors: Sohane, Anurag, Agarwal, Ravinder
• Format: Journal Article
• Language: English
• Published: New Delhi Springer India 01.03.2022; Springer Nature B.V
• Subjects: Artificial neural networks; Classification; Electromyography; Exoskeletons; Knee; Mathematical and Computational Physics; Mathematical Methods in Physics; Measurement Science and Instrumentation; Muscles; Numerical and Computational Physics; Original Paper; Parameter identification; Particle swarm optimization; Performance prediction; Physics; Physics and Astronomy; Rehabilitation; Short term; Simulation; Theoretical; Walking; Particle swarm optimization algorithm (PSO); Long short-term memory (LSTM); Surface electromyography (SEMG); Prediction
• ISSN: 0970-3950; 0974-9853
• DOI: 10.1007/s12647-021-00478-6

Continuous knee joint angle and surface electromyography (SEMG) signal prediction could improve exoskeleton performance. Prediction of SEMG and knee angle is helpful for the physiotherapist for the improvement in remote rehabilitation. Particle Swarm Optimization-Long Short Term Memory (PSO-LSTM) has been used to classify three movements (Flexion, Extension, Ramp Walking) and predict to improve exoskeleton performance. Five healthy subjects participated in testing the effectiveness of the model. Four knee muscles SEMG signals, namely biceps femoris, vastus medialis, rectus femoris and semitendinosus, and knee angle, were used as model inputs. RMSE, r , and R 2 were taken as evaluation parameters to identify the model's robustness for predicting SEMS signal and knee angle. The proposed model was used to classify three movements (Flexion, Extension, Ramp Walking) with an accuracy of 98.58%. The LSTM-PSO results were compared with random LSTM for predicting and classifying the three movements, and the performance of the proposed model was found to be better. This model could be beneficial in rehabilitating stroke patients in remote areas and designing assistive devices.