Recognition of Robot Based on Attention Mechanism and Convolutional Neural Network

In order to accurately recognize the working targets in complex environments, an industrial robot vision model based on attention mechanism and deep learning is proposed. The model uses an improved attention mechanism to achieve fast focusing of the target, and employ a 10-layer-convolutional neural network (CNN) which combines local connection with full connection to accomplish target recognition. The local connection of CNN consists of three convolution layers and three sub-sampling layers, the convolution layers are used for feature extraction and the sub-sampling layers are used to reduce network nodes. The full connection layer of CNN is composed of input layer, hidden layer and output layer as a classifier for target recognition. More than 1000 target images are sampled for CNN network training. The effects of different CNN network structure parameters on the model are analyzed in order to satisfy the rapidity and reliability of robot vision. The test results show that the combination of improved attention mechanism and CNN model can achieve the fast focusing and accurate recognition of working targets in the robot system, the proposed model is can be applied to the visual navigation of industrial robots.