Zynq-based acceleration of robust high density myoelectric signal processing

Boschmann, Alexander; Agne, Andreas; Thombansen, Georg; Witschen, Linus; Kraus, Florian; Platzner, Marco

doi:10.1016/j.jpdc.2018.07.004

Cited by 8 publications

(3 citation statements)

References 34 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Recent studies on gesture recognition via sEMG only focuses on improving the performance of algorithms that require high computation and memory hardware environments. Therefore, it is unsuitable for implementation in resource-constrained smart prostheses with daily human-computer interaction devices [12], [13]. To address these issues, the application of sEMG recognition algorithms on embedded platforms has become a new research hotspot [14].…”

Section: Introductionmentioning

confidence: 99%

FPGA-based Lightweight QDS-CNN System for sEMG Gesture and Force Level Recognition

Guo,

Gou,

Yao

et al. 2024

IEEE Trans. Biomed. Circuits Syst.

View full text Add to dashboard Cite

Deep learning (DL) has been used for electromyographic (EMG) signal recognition and achieved high accuracy for multiple classification tasks. However, implementation in resourceconstrained prostheses and human-computer interaction devices remains challenging. To overcome these problems, this paper implemented a low-power system for EMG gesture and force level recognition using Zynq architecture. Firstly, a lightweight network model structure was proposed by Ultra-lightweight depth separable convolution (UL-DSC) and channel attention-global average pooling (CA-GAP) to reduce the computational complexity while maintaining accuracy. A wearable EMG acquisition device for real-time data acquisition was subsequently developed with size of 36 mm×28 mm×4 mm. Finally, a highly parallelized dedicated hardware accelerator architecture was designed for inference computation. 18 gestures were tested, including force levels from 22

show abstract

Section: Introductionmentioning

confidence: 99%

FPGA-based Lightweight QDS-CNN System for sEMG Gesture and Force Level Recognition

Guo,

Gou,

Yao

et al. 2024

IEEE Trans. Biomed. Circuits Syst.

View full text Add to dashboard Cite

show abstract

“…The authors of [58] presented an architecture for embedded HD EMG prosthetics control capable of accelerate up to 4.8 over the software-only solution. In [59] the same authors presented a second architecture for accelerating HD EMG-based control algorithms, where the system achieved a speed-up of 5.5 over the software-only version (Figure 15). This new architecture has been tested with a new pattern recognition method [59][60], based on computer vision features and SVM classification, which results robust to electrodes shift and noisy channels.…”

Section: Exploitation Of Spatial Features Of Hd Emgmentioning

confidence: 99%

“…In imagine categorization tasks the DSIFT features are extracted at multiple scales and combined with histogram-of-visual-words (HOW) [62], forming scale invariant features [63]. In [59] The increasing number of studies using HD EMG, and in particular in the field of classification of movement, demonstrate that they can represent a valuable means to improve the control in upper limb prostheses. They can potentially reduce the drawbacks of the electrodes shift and of the noisy channels, when coupled with more powerful microprocessors.…”

Section: Exploitation Of Spatial Features Of Hd Emgmentioning

confidence: 99%

Increasing the robustness of active upper limb prostheses

Stango¹

View full text Add to dashboard Cite

This thesis is based on my work done at the Institute for Neurorehabilitation Systems at the University Medical Center Goettingen. My work has been partially founded by German Ministry for Education and Research (BMBF) via the Bernstein Focus Neurotechnology (BFNT) Göttingen under grant number 1GQ0810 The local ethics committee approved all studies involving human subjects, and all subjects signed informed consents prior to their participation in the studies. The entire thesis has been originally written by me. Part of the materials used in this thesis have also been published in journals or conferences, where I am the first or corresponding author. All rights for re-use of previously published material were obtained. Reused figures and tables of IEEE publications are marked with © [Year] IEEE.Hereby I declare that I have written this thesis independently and with no other aids and sources than quoted.

show abstract