Locomo-Net: A Low -Complex Deep Learning Framework for sEMG-Based Hand Movement Recognition for Prosthetic Control

Abstract

Background: The enhancement in the performance of the myoelectric pattern recognition techniques based on deep learning algorithm possess computationally expensive and exhibit extensive memory behavior. Therefore, in this paper we report a deep learning framework named 'Low-Complex Movement recognition-Net' (LoCoMo-Net) built with convolution neural network (CNN) for recognition of wrist and finger flexion movements; grasping and functional movements; and force pattern from single channel surface electromyography (sEMG) recording. The network consists of a two-stage pipeline: 1) input data compression; 2) data-driven weight sharing. Methods: The proposed framework was validated on two different datasets- our own dataset (DS1) and publicly available NinaPro dataset (DS2) for 16 movements and 50 movements respectively. Further, we have prototyped the proposed LoCoMo-Net on Virtex-7 Xilinx field-programmable gate array (FPGA) platform and validated for 15 movements from DS1 to demonstrate its feasibility for real-time execution. Results: The effectiveness of the proposed LoCoMo-Net was verified by a comparative analysis against the benchmarked models using the same datasets wherein our proposed model outperformed Twin- Support Vector Machine (SVM) and existing CNN based model by an average classification accuracy of 8.5 % and 16.0 % respectively. In addition, hardware complexity analysis is done to reveal the advantages of the two-stage pipeline where approximately 27 %, 49 %, 50 %, 23 %, and 43 % savings achieved in lookup tables (LUT's), registers, memory, power consumption and computational time respectively. Conclusion: The clinical significance of such sEMG based accurate and low-complex movement recognition system can be favorable for the potential improvement in quality of life of an amputated persons.

Keywords: CNN; data compression; movement classification; sEMG; signal processing; weights compression.

FIGURE 1.

Shows the block diagram of the feed-forward path of a myoelectric controlled prosthetic limb which is bio-inspired and attempts to replicate the physiological motor control system . There are a number of distinct tasks such as supervised data collection, segmentation, feature selection and extraction, training of the network followed by classification.

FIGURE 2.

Proposed LoCoMo-Net with two-stage pipeline compression technique.

FIGURE 3.

Experimental protocol for recording sEMG data. Where data was recorded in 4 trails for each activity discriobed in Table 2.

FIGURE 4.

(a) Workflow of the proposed LoCoMo-Net where T_set1 and V_set1 represent the train-test and validating set of DS1 respectively. The Nested 10 fold CV is used for the selection of best hyper-parameters. (b) The network topology of the proposed LoCoMo-Net model for classifying the task.

FIGURE 5.

(a) Shows block level architecture of the proposed Min/max technique; (b) Mean entropy comparison of sEMG with and without input data compression for 3 amputated and 11 healthy subjects; (c) Comparison of the raw sEMG signal with compressed sEMG from the proposed input data compression module implemented on software (MATLAB) and hardware (FPGA), the analysis shows both the hardware and software results retains same morphology like raw sEMG with less number of samples.

FIGURE 6.

Shows influence of input compression on the computational cost of the model where for the calculation of total number of multiplication and addition units in a particular layer is done by considering kernels of size is getting convolved with input feature map of size with stride rate of , which results in output feature maps of size . Thus, total number of multiplication () and addition () operations at layer are and respectively.

FIGURE 7.

Shows an example of the data-driven weight sharing module. Where weights of the fully connected layer before compression and after compression (yellow color show the weights replaced with their corresponding local maxima or minima depicted with green color in the zoomed plot) are represented with blue and red color respectively.

FIGURE 8.

Block level architecture of the proposed LoCoMo-Net model.

FIGURE 9.

ROC curve of LoCoMo-Net model.

FIGURE 10.

FPGA prototyping of the LoCoMo-Net model shown for Task 15 (DS1), where red color circle indicates output with illuminating LED and yellow color circle indicates input knob.

FIGURE 11.

Scalability analysis for different number of channels.