Human-robot interaction systems based on recognizing hand gestures from images with the help of graphical user interface are used in smart systems today. For this purpose, image processing algorithms are used to recognize hand gestures. One of the most effective methods in this field is the use of deep neural networks to detect hand gestures. In this work, a hand state classification system is presented with the help of a combined convolutional neural network. In this thesis, two pre-trained networks ResNet and GoogleNet have been used to extract features from images. The combination of these two features increases the accuracy of recognizing hand gestures. Then the optimal features are selected with the help of bat optimization algorithm. The results of the simulations show the superiority of the proposed method in terms of accuracy compared to other works.
