Research on Objectives Recognition of Robotics based on Convolutional Neural Network

Recently, robotics has widely utilized in tremendous applications including traffic transportation, medical treatments, power system and virtual environment. These applications are almost based on the vision system, which is consisted by detection sensors and cameras. However, existing recognition methods are concentrated on the precision of machine learnings under the ideal setups and ignored the complex environment for robotics. In this paper, we simulate the realistic robotics sensors with non-complete input images and propose a novel convolutional network without redundancy neural layers. Indeed, an attention mechanism is established to enhance the recognition accuracy of proposed framework. Finally, we evaluate our model with existing objectives identification learning methods in the identical testing configurations. From our extensive experiment results and analysis, we can conclude that the proposed model can achieve the objectives recognition in complex robotics scenario with acceptable recognition accuracy and reasonable computation costs.