Contact classification for human–robot interaction with densely connected convolutional neural network and convolutional block attention module

Human–robot interaction (HRI) is a challenging topic to perform various tasks in many repetitive and dangerous tasks. However, humans not only share a workspace with robots, they also use them as helpful assistants. In this study, a solution is sought for the problem of recognizing human hand touch, which is equipped with large-area tactile sensors on the surface of a robot, which comes into contact with these sensors. The main purpose of the study is to distinguish between the contacts of human–robot interaction. Thanks to this separation, it is possible to trigger robot movements, activate them, or make the physical interaction of living-inanimate beings safe. Dense connected Convolutional Block Attention Module based network is proposed for classification of human hand touch. A number of experimental studies are carried out to verify the classification efficiency of the proposed Convolutional Neural Network (CNN) model. The obtained results are extensively compared with the commonly used pre-trained CNN models. From the results of this study, it is seen that the proposed CNN model provides more satisfactory results than other methods with an accuracy rate of 98.20%.