A strategy for improving the precision of material removal model based on deep neural networks was proposed. A deep learning algorithm with ability of feature selecting was proposed. A series of simulation samples composed of a material removal rate and corresponding polishing parameters were generated based on the model of material removal rate for robot polishing. The deep learning algorithm learned both the simulation samples and practical samples and then a deep learning model was established. The error between material removal depth of the test samples and material removal depth estimated by polishing parameters by using proposed deep learning model was calculated and compared. The results show that the improved model can achieve higher accuracy than the traditional models.