Abstract:Surface Electromyography (sEMG) signal processing has a disruptive technology potential to enable a natural human interface with artificial limbs and assistive devices. However, this biosignal real-time control interface still presents several restrictions such as control limitations due to a lack of reliable signal prediction and standards for signal processing among research groups. Our paper aims to present and validate our sEMG database through the signal classification performed by the reliable forms of o… Show more
“…On these data, Palermo et al [11] reached an inter-session accuracy of 25.4% by feeding Wave Length to a Random Forest. Cene et al [16] successfully employed Extreme Learning Machines (ELMs) to raise this intersession accuracy to 41.8%. It is worth to notice that the reason why the accuracy reached on the NinaPro DB6 is much lower than the one reached on other datasets with a similar number of classes and sensors, is that the hand movements of NinaPro DB6 are all grasps, thus much less diverse and discernable than the gestures in ordinary datasets.…”
Section: B Related Workmentioning
confidence: 99%
“…On the Unibo-INAIL dataset, Milosevic et al [10] showed that multi-posture and multi-day training improve intersession generalization. A Radial Basis Function kernel SVM (RBF-SVM) applied on 4-channel single samples of the RMS signal yielded an intra-session recognition accuracy higher than 90%, with an inter-session accuracy drop up to 20% (a value similar to [11], [16]). The aforementioned approaches showed the major limitation of classical ML: it strongly relies on domain-specific knowledge and hand-crafted features, limiting the capability to generalize over time.…”
Section: B Related Workmentioning
confidence: 99%
“…To evaluate the inference performance, we run TEMPONet on the Ninapro DB6 comparing the results of the multisession testing described in [16], that represents, at the best of our knowledge, the previous state-of-the-art inter-session accuracy for the NinaPro DB6. We obtained an average 49.6% accuracy against 41.8% reported in [16]. In this test, the accuracy is evaluated on the entire gestures (i.e.…”
Section: A Experimental Setupmentioning
confidence: 99%
“…The system is based on the combination of a commercial Analog Front End for biopotential acquisition with GAP8, a multicore low-power IoT processor; • a 20 session dataset, collected with our custom platform, which allows us to validate the algorithm and to profile a quantized version of the TCN, suitable for the deployment on a resource-constrained platform. We tested the performance of TEMPONet on the NinaPro DB6 dataset [11], achieving 65.2% inter-session accuracy on steady signals and 49.6% inter-session on the full dataset -7.8% better than the current state-of-the-art [16]. Moreover, after the system design, we tested the same TEMPONet topology on a new dataset we introduce in this work, comprising 20 sessions on three subjects.…”
Hand movement classification via surface electromyographic (sEMG) signal is a well-established approach for advanced Human-Computer Interaction. However, sEMG movement recognition has to deal with the long-term reliability of sEMG-based control, limited by the variability affecting the sEMG signal. Embedded solutions are affected by a recognition accuracy drop over time that makes them unsuitable for reliable gesture controller design. In this paper, we present a complete wearable-class embedded system for robust sEMGbased gesture recognition, based on Temporal Convolutional Networks (TCNs). Firstly, we developed a novel TCN topology (TEMPONet), and we tested our solution on a benchmark dataset (Ninapro), achieving 49.6% average accuracy, 7.8%, better than current State-Of-the-Art (SoA). Moreover, we designed an energy-efficient embedded platform based on GAP8, a novel 8-core IoT processor. Using our embedded platform, we collected a second 20-sessions dataset to validate the system on a setup which is representative of the final deployment. We obtain 93.7% average accuracy with the TCN, comparable with a SoA SVM approach (91.1%). Finally, we profiled the performance of the network implemented on GAP8 by using an 8-bit quantization strategy to fit the memory constraint of the processor. We reach a 4× lower memory footprint (460 kB) with a performance degradation of only 3% accuracy. We detailed the execution on the GAP8 platform, showing that the quantized network executes a single classification in 12.84 ms with a power envelope of 0.9 mJ, making it suitable for a longlifetime wearable deployment.
“…On these data, Palermo et al [11] reached an inter-session accuracy of 25.4% by feeding Wave Length to a Random Forest. Cene et al [16] successfully employed Extreme Learning Machines (ELMs) to raise this intersession accuracy to 41.8%. It is worth to notice that the reason why the accuracy reached on the NinaPro DB6 is much lower than the one reached on other datasets with a similar number of classes and sensors, is that the hand movements of NinaPro DB6 are all grasps, thus much less diverse and discernable than the gestures in ordinary datasets.…”
Section: B Related Workmentioning
confidence: 99%
“…On the Unibo-INAIL dataset, Milosevic et al [10] showed that multi-posture and multi-day training improve intersession generalization. A Radial Basis Function kernel SVM (RBF-SVM) applied on 4-channel single samples of the RMS signal yielded an intra-session recognition accuracy higher than 90%, with an inter-session accuracy drop up to 20% (a value similar to [11], [16]). The aforementioned approaches showed the major limitation of classical ML: it strongly relies on domain-specific knowledge and hand-crafted features, limiting the capability to generalize over time.…”
Section: B Related Workmentioning
confidence: 99%
“…To evaluate the inference performance, we run TEMPONet on the Ninapro DB6 comparing the results of the multisession testing described in [16], that represents, at the best of our knowledge, the previous state-of-the-art inter-session accuracy for the NinaPro DB6. We obtained an average 49.6% accuracy against 41.8% reported in [16]. In this test, the accuracy is evaluated on the entire gestures (i.e.…”
Section: A Experimental Setupmentioning
confidence: 99%
“…The system is based on the combination of a commercial Analog Front End for biopotential acquisition with GAP8, a multicore low-power IoT processor; • a 20 session dataset, collected with our custom platform, which allows us to validate the algorithm and to profile a quantized version of the TCN, suitable for the deployment on a resource-constrained platform. We tested the performance of TEMPONet on the NinaPro DB6 dataset [11], achieving 65.2% inter-session accuracy on steady signals and 49.6% inter-session on the full dataset -7.8% better than the current state-of-the-art [16]. Moreover, after the system design, we tested the same TEMPONet topology on a new dataset we introduce in this work, comprising 20 sessions on three subjects.…”
Hand movement classification via surface electromyographic (sEMG) signal is a well-established approach for advanced Human-Computer Interaction. However, sEMG movement recognition has to deal with the long-term reliability of sEMG-based control, limited by the variability affecting the sEMG signal. Embedded solutions are affected by a recognition accuracy drop over time that makes them unsuitable for reliable gesture controller design. In this paper, we present a complete wearable-class embedded system for robust sEMGbased gesture recognition, based on Temporal Convolutional Networks (TCNs). Firstly, we developed a novel TCN topology (TEMPONet), and we tested our solution on a benchmark dataset (Ninapro), achieving 49.6% average accuracy, 7.8%, better than current State-Of-the-Art (SoA). Moreover, we designed an energy-efficient embedded platform based on GAP8, a novel 8-core IoT processor. Using our embedded platform, we collected a second 20-sessions dataset to validate the system on a setup which is representative of the final deployment. We obtain 93.7% average accuracy with the TCN, comparable with a SoA SVM approach (91.1%). Finally, we profiled the performance of the network implemented on GAP8 by using an 8-bit quantization strategy to fit the memory constraint of the processor. We reach a 4× lower memory footprint (460 kB) with a performance degradation of only 3% accuracy. We detailed the execution on the GAP8 platform, showing that the quantized network executes a single classification in 12.84 ms with a power envelope of 0.9 mJ, making it suitable for a longlifetime wearable deployment.
“…ELM is a feed forward neural network raised by professor Guang-Bin Huang [45], with a single input layer, a hidden layer, and an output layer. In most cases, the input weights and hidden nodes are randomly assigned, and the output weights are directly calculated by the least square method in just a single step.…”
Aroma and taste are the most important attributes of alcoholic beverages. In the study, the self-developed electronic tongue (e-tongue) and electronic nose (e-nose) were used for evaluating the marked ages of rice wines. Six types of feature data sets (e-tongue data set, e-nose data set, direct-fusion data set, weighted-fusion data set, optimized direct-fusion data set, and optimized weighted-fusion data set) were used for identifying rice wines with different wine ages. Pearson coefficient analysis and variance inflation factor (VIF) analysis were used to optimize the fusion matrixes by removing the multicollinear information. Two types of discrimination methods (principal component analysis (PCA) and locality preserving projections (LPP)) were used for classifying rice wines, and LPP performed better than PCA in the discrimination work. The best result was obtained by LPP based on the weighted-fusion data set, and all the samples could be classified clearly in the LPP plot. Therefore, the weighted-fusion data were used as independent variables of partial least squares regression, extreme learning machine, and support vector machines (LIBSVM) for evaluating wine ages, respectively. All the methods performed well with good prediction results, and LIBSVM presented the best correlation coefficient (R2 ≥ 0.9998).
For the problem of surface electromyography (sEMG) gesture recognition, considering the fact that the traditional machine learning model is susceptible to the sEMG feature extraction method, it is difficult to distinguish the subtle differences between similar gestures. The NinaPro DB1 dataset is used as the research object, and the sEMG feature image and the Convolutional Neural Network (CNN) are combined to recognize 52 gesture movements. The CNN model effectively solves the limitations of traditional machine learning in sEMG gesture recognition, and combines 1-dim convolution kernel to extract deep abstract features to improve the recognition effect. Finally, the simulation experiment shows that compared with the accuracy of the raw-sEMG images based on the CNN and the sEMG-feature-images based on the CNN and sEMG based on the traditional machine learning, the multi-sEMG-features image based on the CNN is the highest, which coming up to 82.54%.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.