A Convolutional Neural Network for Soft Robot Images Classification

Abstract

In this work, a Convolutional Neural Network (CNN) is used to classify the images of soft robotic actuators as bending, triangle, and muscle actuators. The classifier model is built with a total 390 images of soft actuators comprising the soft actuators with 130 images for bending, triangle, and muscle actuators, respectively. 70% of the images were used for training, while 30% were used for validation. The developed CNN model achieved a loss of 7.63% and accuracy of 97.6% for the training data while a loss of 9.64% and accuracy of 85.71% was obtained on the validation data.