Browsing by Author "Moofarry, Jhon Freddy"
Now showing 1 - 2 of 2
Results Per Page
Sort Options
Item Proposal for the Implementation of MLP Neural Networks on Arduino Platform(Springer, 2020-03-03) Suaza Cano, Kevin Andrés; Moofarry, Jhon Freddy; Castillo García, Javier FerneyThis paper presents implementation MLP artificial neural networks on embedded low-cost microcontrollers that can be dynamically configured on the run. The methodology starts with the training process, goes through the codification of the neural network into the microcontroller format, and finishes with the execution process of the embedded NNs. It is presented how to compute deterministically the memory space require for a certain topology, as well as the required fields to execute the neural network. The training and verification was done with Matlab and programming with a IDE Arduino compiler. The results show statistical and graphical analysis for several topologies, average execution times for various transfer function, and accuracy.Item Selection of Mental Tasks for Brain-Computer Interfaces Using NASA-TLX Index(Springer, 2020-03-03) Moofarry, Jhon Freddy; Suaza Cano, Kevin Andrés; Saavedra Lozano, Diego Fernando; Castillo García, Javier FerneyThe brain-computer interfaces - BCIs allow people with disabilities to interact with the outside world using different communication channels than conventional ones. This article deals with the selection of tasks in the protocols for the development of BCI based on the paradigm of mental tasks. It is proposed to use the NASA-TLX index to evaluate the effect of the mental load of each of the tasks and contrast the performance of the interface task by task. In the implementation of BCI, the OPENBCI hardware was used for signal acquisition and the MATLAB software for processing. Five mental tasks were defined that activated different regions of the cerebral cortex. The acquisition protocol consisted of defining the rest time, execution and recovery for the tasks. The extraction methods used temporal, frequency and time-frequency combination characteristics. The classifiers used were neural networks, nearby neighbors and support vector machines. The evaluation of the TLX index seeks to quantify the appreciation of the effort, frustration and complexity of the task, therefore after the acquisition of signals for each task, the participant proceeded to evaluate the mental overload using the NASA-TLX index. The results obtained show that those tasks that require greater complexity to be performed presented a greater repeatability and higher success rate.