Catalogue Search | MBRL
Are you sure you want to remove the book from the shelf?
{{itemTitle}}
29
result(s) for
"Ben Abdallah, Abderazek"
Sort by:
Study of Deep Learning-based Hand Gesture Recognition Toward the Design of a Low-cost Prosthetic Hand
Background: The surface EMG (sEMG) signal is inherently noisy and, therefore, not a robust input source for prosthetic systems, especially for fatigue, electrode displacement, and sweat conditions. We propose to address these issues by designing a multi-modal approach that combines vision and EMG empowered with appropriate dataset collection. Methods: In Frame-based, the machine learning model used for recognition was a 2D-CNN. The data is image data that is input to the model by preparing videos showing 10 patterns of hand gestures along with multiple backgrounds, and dividing these videos into frames. These image data are then pre-processed and input to the machine learning model. The model is then evaluated in terms of the accuracy of hand gesture identification using the test data and the loss value, which represents the error between the expected data and the correct data output. In EMG, the Myo armband is placed on the forearm and the sEMG of 200 (Hz) is measured. There are six patterns of hand gestures in this process. Similar to the images, these sEMG data are preprocessed and input to a machine learning model for classification. The model is evaluated the model by the accuracy of hand gesture identification using the test data and the loss value, precision, recall , F1-score. Results: The value of the loss function in case of frame-based was 0.0770 and the accuracy was 0.9739 at 1000 epochs of the training data. And the value of the loss function values in the test data were 0.1011 for the loss value and 0.9657 for the accuracy. In the case of EMG, the loss value was 0.931 when the time to maintain the gesture was the longest, and the loss value was 0.171. However, Precision, Recall, and F1-score were not the highest at the longest time for some gestures. Conclusion: In this paper, we created a hand gesture identification software using Frame-based and sEMG, and measured its accuracy and loss value. For sEMG, we used Precision, Recall, and F1-score to check the metrics of each gesture identification. The frame-based results showed good results in both precision and loss values. sEMG showed an improvement in precision and loss values as the time length increased, but there was a tendency to decrease in some indices. In the future, it is necessary to explore the local relationship between finger and forearm to optimize out learning model.
Journal Article
Analytical performance assessment and high-throughput low-latency spike routing algorithm for spiking neural network systems
Large-scale artificial neural networks (ANNs) have been used to mimic the information processing function of the brain. Spiking neural networks (SNNs) are a kind of ANN, which mimic real biological neural networks, conveying information through the communication of short pulses between neurons. Since each neuron in these networks is connected to thousands of others, high bandwidth is required. Moreover, since the spike times are used to encode information in SNN, very low communication latency is also required. The 2D-NoC was used as a solution to provide a scalable interconnection fabric in large-scale parallel SNN systems. The 3D-ICs have also attracted a lot of attention as a potential solution to resolve the interconnect bottleneck. The combination of these two emerging technologies provides a new horizon for IC designs to satisfy the high requirements of low-power and small footprint in emerging AI applications. This paper first presents an analytical model to analyze the performance of different neural network topologies and compare it with a system-level simulation. Second, we present an architecture and a low-latency routing algorithm for spike traffic routing in 3D-NoC of spiking neurons (3DNoC-SNN). The 3DNoC-SNN is validated based on an RTL-level implementation, while area/power analysis is performed using 45-nm CMOS technology.
Journal Article
A low-overhead soft–hard fault-tolerant architecture, design and management scheme for reliable high-performance many-core 3D-NoC systems
The Network-on-Chip (NoC) paradigm has been proposed as a favorable solution to handle the strict communication requirements between the increasingly large number of cores on a single chip. However, NoC systems are exposed to the aggressive scaling down of transistors, low operating voltages, and high integration and power densities, making them vulnerable to permanent (hard) faults and transient (soft) errors. A hard fault in a NoC can lead to external blocking, causing congestion across the whole network. A soft error is more challenging because of its silent data corruption, which leads to a large area of erroneous data due to error propagation, packet re-transmission, and deadlock. In this paper, we present the architecture and design of a comprehensive soft error and hard fault-tolerant 3D-NoC system, named 3D-Hard-Fault-Soft-Error-Tolerant-OASIS-NoC (3D-FETO). With the aid of efficient mechanisms and algorithms, 3D-FETO is capable of detecting and recovering from soft errors which occur in the routing pipeline stages and leverages reconfigurable components to handle permanent faults in links, input buffers, and crossbars. In-depth evaluation results show that the 3D-FETO system is able to work around different kinds of hard faults and soft errors, ensuring graceful performance degradation, while minimizing additional hardware complexity and remaining power efficient.
Journal Article
Spike-Event X-ray Image Classification for 3D-NoC-Based Neuromorphic Pneumonia Detection
The success of deep learning in extending the frontiers of artificial intelligence has accelerated the application of AI-enabled systems in addressing various challenges in different fields. In healthcare, deep learning is deployed on edge computing platforms to address security and latency challenges, even though these platforms are often resource-constrained. Deep learning systems are based on conventional artificial neural networks, which are computationally complex, require high power, and have low energy efficiency, making them unsuitable for edge computing platforms. Since these systems are also used in critical applications such as bio-medicine, it is expedient that their reliability is considered when designing them. For biomedical applications, the spatio-temporal nature of information processing of spiking neural networks could be merged with a fault-tolerant 3-dimensional network on chip (3D-NoC) hardware to obtain an excellent multi-objective performance accuracy while maintaining low latency and low power consumption. In this work, we propose a reconfigurable 3D-NoC-based neuromorphic system for biomedical applications based on a fault-tolerant spike routing scheme. The performance evaluation results over X-ray images for pneumonia (i.e., COVID-19) detection show that the proposed system achieves 88.43% detection accuracy over the collected test data and could be accelerated to achieve 4.6% better inference latency than the ANN-based system while consuming 32% less power. Furthermore, the proposed system maintains high accuracy for up to 30% inter-neuron communication faults with increased latency.
Journal Article
Parallelization and Hardware Mapping of Deep Neural Network on Reconfigurable Platform for AI-Enabled Biomedical System
COVID-19 is still disrupting many parts of the world. A rapid and accurate diagnosis solution is needed to combat the pandemic. As a part of the AIRBiS(AI-Enabled Real-time Pneumonia Detection Bio-medical System), this work conduct hardware acceleration to speed up the diagnosis. We found that more than 90% of the current diagnosis time is spent on the convolution function and have conducted three methods to speed up the convolution operations. Firstly, by applying the Winograd algorithm on convolution layers, the multiplication operations of the matrices can be decreased, which speeds up the calculation. The next step is to improve the data exchange speed between the FPGA and CPU by replacing the normal buffer with LineBuffer. We also tried to improve the calculation speed by quantization, reducing the number of bits used for the filter and the input image. The FPGA board we used for this research is ZCU102. The application used for high-level synthesis is Xilinx SDSoC 2019.1. Using the mentioned approaches, we improved the inference speed from 106ms to 22.2ms per image.
Journal Article
Study of a Multi-modal Neurorobotic Prosthetic Arm Control System based on Recurrent Spiking Neural Network
The use of robotic arms in various fields of human endeavor has increased over the years, and with recent advancements in artificial intelligence enabled by deep learning, they are increasingly being employed in medical applications like assistive robots for paralyzed patients with neurological disorders, welfare robots for the elderly, and prosthesis for amputees. However, robot arms tailored towards such applications are resource-constrained. As a result, deep learning with conventional artificial neural network (ANN) which is often run on GPU with high computational complexity and high power consumption cannot be handled by them. Neuromorphic processors, on the other hand, leverage spiking neural network (SNN) which has been shown to be less computationally complex and consume less power, making them suitable for such applications. Also, most robot arms unlike living agents that combine different sensory data to accurately perform a complex task, use uni-modal data which affects their accuracy. Conversely, multi-modal sensory data has been demonstrated to reach high accuracy and can be employed to achieve high accuracy in such robot arms. This paper presents the study of a multi-modal neurorobotic prosthetic arm control system based on recurrent spiking neural network. The robot arm control system uses multi-modal sensory data from visual (camera) and electromyography sensors, together with spike-based data processing on our previously proposed R-NASH neuromorphic processor to achieve robust accurate control of a robot arm with low power. The evaluation result using both uni-modal and multi-modal input data show that the multi-modal input achieves a more robust performance at 87%, compared to the uni-modal.
Journal Article
An Affordable 3D-printed Open-Loop Prosthetic Hand Prototype with Neural Network Learning EMG-Based Manipulation for Amputees
Despite the advancement of prosthetic hands, many of the conventional products are difficult to control and have limited capabilities. Even though these limitations are being pushed by many state-of-the-art commercial prosthetic hand products, they are often expensive due to the high cost of production. Therefore, in the Adaptive Neuroprosthesis Arm (NeuroSys) project, we aim to develop a low-cost prosthetic hand with high functionalities that let users perform various gestures and accurate grasp. This paper mainly focuses on the sEMG signal recognition and control for a prototype 3D printed prosthetic hand model. In this work, we have considered the prosthetic hand to operate from a non-intrusive sensor, surface Electromyographic signal (sEMG). The signal used to control the prosthetic hand is received from a low-cost, 8-channel sEMG sensor, Myo armband. The sensor is placed around a person’s upper forearm under the elbow, and the signal is sent wirelessly to a computer. After the signal is received, a neural network is used to recognize and classify the intention of the signals. The network model is designed for specific individuals to increase the controllability of the prosthetic hand. Also, to mimic the real-world usage, evaluation on two different sessions is conducted. With the use of Recurrent Neural Networks (RNNs) family, sEMG data recognition can reach around 85% of accuracy. While Gated Recurrent Units (GRUs) and Long Short Term Memory (LSTM) have similar results, simple RNN unit produces very low accuracy. Also, the more session the sample data is taken, the more robust the recognition system can be. Using the Myo armband sensor, sEMG signal data during a steady state with force or no force can affect the accuracy performance of the decoding hand gestures. In terms of real-world usage, however the constant force must be applied, otherwise, the system fails to classify the gestures. Also, the variation of sensor placement can affect the deep learning model. Although, there is a trade-off between accuracy and delay, optimal window size can be explored. Using the mentioned method, a prototype of an affordable 3D printed prosthetic hand controlled using sEMG is realized, although it is still far from real-world usage.
Journal Article
A Low-cost Raspberry Pi-based Control System for Upper Limb Prosthesis
recent years, robots have been introduced in most factories. However, manual work still continues to be done in some places where giant robots cannot be installed. In particular, traditional Japanese crafts are done by hand, and people that engage in such crafts are called craftsmen. Generally, such artisans need years of training and cannot become experts right away. One of the problems these artisans face is the lack of successors. To address this challenge, this paper proposes a raspberry pi hardware based control method for a prosthetic hand using hand gestures from camera sensor, which will allow a prosthetic hand to learn the hand movements of the craftsmen and perform the crafts. The advantage of this is that there is no need for training, which usually takes years. To control the prosthetic hand, hand gestures are captured from a camera sensor, converted to HSV and binarized, and then classified into one of five gestures using a CNN implemented on the raspberry pi hardware. The recognized gesture is then relayed to the prosthetic hand to mimic the classified gesture. A dataset containing 2000 captured images of each gesture was created to evaluate the performance, and these gestures clearly define the closing and opening of the fingers. Using a 32×32 hand gesture image dataset captured from camera, we validated the trained CNN first in software for hand recognition without using Raspberry Pi, and achieved an accuracy of 99.63%, and then implemented on the raspberry pi, and performed real-time evaluation by recognizing five hand gestures captured from the camera sensor in real-time. Out of the hand gestures, four were correctly recognized. We presented the design of a low-cost prosthetic hand based on raspberry pi hardware, and evaluated its real-time hand gesture recognition. The evaluation result show that the proposed system is able to correctly recognize four hand gestures.
Journal Article
Real-time Hand-Gesture Recognition based on Deep Neural Network
Hand gestures are a kind of nonverbal communication in which visible bodily actions are used to communicate important messages. Recently, hand gesture recognition has received significant attention from the research community for various applications, including advanced driver assistance systems, prosthetic, and robotic control. Therefore, accurate and fast classification of hand gesture is required. In this research, we created a deep neural network as the first step to develop a real-time camera-only hand gesture recognition system without electroencephalogram (EEG) signals. We present the system software architecture in a fair amount of details. The proposed system was able to recognize hand signs with an accuracy of 97.31%.
Journal Article