A multi-view model fusion network with double branch structure

摘要

The composition of things is multifaceted. Agents need to capture multiple perspectives of things in order to establish a better understanding of the real world. Machine vision has made great achievements, but it often performs well only in specific perspectives and environments. In this work, we propose a multi perspective network model with dual branch configuration, build our own data set through the built physical experiment platform, and use high-definition cameras to collect visual images from the vertical and side of opaque cans with eight contents, Eight kinds of content have very high similarity. The data from two perspectives are extracted by convolution neural network, and finally the model level fusion is carried out to realize classification. The results show that the accuracy of the model is 95.2%, which can be applied to kitchen robots and other operational tasks to improve the humanoid ability of robots.