Convolutional neural networks and genetic algorithm for visual imagery classification
- PDF / 1,846,938 Bytes
- 11 Pages / 595.276 x 790.866 pts Page_size
- 14 Downloads / 220 Views
SCIENTIFIC PAPER
Convolutional neural networks and genetic algorithm for visual imagery classification Fabio R. Llorella1 · Gustavo Patow1 · José M. Azorín2 Received: 12 March 2019 / Accepted: 29 June 2020 © Australasian College of Physical Scientists and Engineers in Medicine 2020
Abstract Brain–Computer Interface (BCI) systems establish a channel for direct communication between the brain and the outside world without having to use the peripheral nervous system. While most BCI systems use evoked potentials and motor imagery, in the present work we present a technique that employs visual imagery. Our technique uses neural networks to classify the signals produced in visual imagery. To this end, we have used densely connected neural and convolutional networks, together with a genetic algorithm to find the best parameters for these networks. The results we obtained are a 60% success rate in the classification of four imagined objects (a tree, a dog, an airplane and a house) plus a state of relaxation, thus outperforming the state of the art in visual imagery classification. Keywords Brain–Computer Interface · Visual imagery · Deep learning · Genetic algorithms · Keras
Introduction A large proportion of the information that human being receive arrives through their vision, which is why the scientific community is interested in understanding the neurological processes that occur in the brain and the areas that are involved. On the other hand, it is logical for BCI technologies to try to exploit whatever possibilities are available to improve the communication between a brain and the computer it is controlling, and because visual imagination is a cognitive tool available to the human being, that is why we proceed to study the possibilities this cognitive capacity may have for use in BCI systems. Many BCI studies have successfully focused on evoked potentials such as P300, SSVEP [1, 2], or the use of motor imagery [3–5], with results that have enabled the development of many practical applications [6,
* Fabio R. Llorella [email protected] Gustavo Patow [email protected] José M. Azorín [email protected] 1
ViRVIG-UdG. Universitat de Girona, Girona, Spain
Brain-Machine Interface Systems Lab, Avda. de la Universidad s/n. Ed. Innova, 03202 Elche, Alicante, Spain
2
7]. Other possibilities that have been studied are the detection and classification of words (also called silent speech) using EEG signals [8], and rehabilitation applications using an exoskeleton controlled by EEG signals [9]. All these examples show a high degree of evolution of BCI technologies with respect to motor signals. The visual imagery can be defined as “the manipulation of visual information that comes not from perception but from memory” [10]. Detecting and classifying of the visual imagery would help to increase the number of applications that could be made more naturally through a BCI, such as CAD systems[11] or more natural drawing systems, and this could help people with mobility problems who would now be able to perform ar
Data Loading...
