Bibliographic citations
Barazorda, E., (2023). Diseño de un sistema de control mioeléctrico con clasificación en tiempo real de movimientos manuales basado en Deep Learning para mejorar la destreza de las prótesis de mano de agarre simple [Tesis, Universidad Nacional de Ingeniería]. http://hdl.handle.net/20.500.14076/27086
Barazorda, E., Diseño de un sistema de control mioeléctrico con clasificación en tiempo real de movimientos manuales basado en Deep Learning para mejorar la destreza de las prótesis de mano de agarre simple [Tesis]. PE: Universidad Nacional de Ingeniería; 2023. http://hdl.handle.net/20.500.14076/27086
@misc{renati/712889,
title = "Diseño de un sistema de control mioeléctrico con clasificación en tiempo real de movimientos manuales basado en Deep Learning para mejorar la destreza de las prótesis de mano de agarre simple",
author = "Barazorda Rodríguez, Edgard Jesús",
publisher = "Universidad Nacional de Ingeniería",
year = "2023"
}
In Peru, highly dexterous hand prostheses are primarily acquired through imports, incurring significant costs and lengthy waiting times. Additionally, domestic developments in this field have predominantly focused on simple grip control, thereby limiting the potential of mechanical design. To enhance the dexterity of such prostheses, this research endeavors to design a myoelectric control system with real-time classification of manual movements based on Deep Learning. The proposed control system comprises three subsystems: acquisition, to capture electromyographic (EMG) signals related to five manual movements associated with activities of daily living (ADL); processing, for standardizing, filtering, and converting these signals into images (spectrograms) using Complex Wavelet Transform (CWT); and classification, to train a CNN MobileNet V2 model utilizing spectrograms as input. To enhance user interaction, a Graphical User Interface (GUI) was programmed to acquire, process, and classify signals in real-time. Furthermore, an open-source prosthesis was simulated to provide visual feedback to users. Experimental tests conducted with five volunteers yielded an average real-time classification accuracy of 87.3% and an average control latency of 276.97 ms. These results substantiate a 30% enhancement in dexterity compared to conventional simple grip control systems.
This item is licensed under a Creative Commons License