Deep Learning Techniques for Thought-Controlled IoT Devices

The human brain is made up of millions of neurons that are responsible for directing the activity of the human body in response to internal–external stimuli generated by motor and sensory inputs. Quotidian tasks rely on this basic functioning of our body, which could be simplified by harbouring the advancements in technology offered by Internet of Things (IoT), something that could only be fathomed to be fictional until recent years. This allows one to control systems based solely on their thoughts. The goal of IoT in this scenario is to replicate the activity conducted by the neurons to transmit the brain signals to the body using a system that directly analyses and translates these signals into motor actions that the user wishes to perform. The human brain consists of around 100 billion neurons, wherein each of these neurons form connections to other neurons, adding up to roughly one quadrillion connections. This immense network of neurons is capable of incessantly processing brain signals at blazing fast speeds. To come up with a system that almost matches the capabilities of the human brain, one must leverage the state-of-the-art machine learning and deep learning approaches to yield feasible techniques of using the electroencephalogram (EEG) data for said thought-controlled systems. This chapter delves into the various approaches that can extract features from the EEG signals, covering FBSCP, ConvNet, EEGNet, Hybrid Neural Network, Siamese Neural Network, RBN and DBN and discusses the real-world applications like brain-controlled prosthetic arm and brain-controlled wheelchair that use EEG signal in IoT.