Use of Three Distinct Types of Cuneate Neurons in the Classification of Naturalistic Textures

The tactile sensation is encoded by mechanoreceptors in the skin that characterize information such as pressure and vibration. This information is processed by the somatosensory system and the first major processing step is in the cuneate nucleus. Based on the functionality of the cuneate nucleus, a model of the intercellular dynamics of this processing stage has recently been proposed. The validation of this model considered that there was only one type of cuneate neuron that received synaptic input from slow and fast adapting mechanoreceptors; however, physiologically there are several types of cuneate neurons with specific synaptic inputs. Considering this specificity, this paper hypothesizes that the use of different types of artificial cuneate neurons would allow better segregation and classification of tactile artificial data. To prove this hypothesis, we implemented a neuromorphic model with three types of bioinspired artificial cuneate neurons and evaluated its ability to classify eight naturalistic textures. Two indexes were extracted from the output of these tree cuneate neuron models and used as a feature vector for five different support vector machines (SVM). The results show that combining the output of the three neural models it was obtained an average accuracy in the texture's classification of 90.47%.