Multi-type Digital Recognition Based on TensorFlow

For the problem of inaccurate identification of numbers, this paper is based on LeNet-5 network structure and optimizes it. The objective function and optimizer are added after the network output, and the sample library is updated to make it more accurate to identify multiple types of numbers. The optimized architecture is applied to identify multiple types of numbers, trained and tested, and the optimization parameters are selected by comparison. The experimental results show that the optimization parameters of certain values have a higher recognition rate for identifying numbers. The study has reference value for multi-type digital recognition.

[1]  Peng Liu,et al.  Multi-feature deep learning for face gender recognition , 2014, 2014 IEEE 7th Joint International Information Technology and Artificial Intelligence Conference.

[2]  Xiaolong Wang,et al.  Convolutional Deep Networks for Visual Data Classification , 2012, Neural Processing Letters.

[3]  Shih-Wei Sun,et al.  Digit Recognition in Natural Scene Texts , 2017 .

[4]  Adil M. Bagirov,et al.  A convolutional recursive modified Self Organizing Map for handwritten digits recognition , 2014, Neural Networks.

[5]  Claus Nebauer,et al.  Evaluation of convolutional neural networks for visual recognition , 1998, IEEE Trans. Neural Networks.

[6]  Ioannis A. Kakadiaris,et al.  Facial landmark detection in uncontrolled conditions , 2011, 2011 International Joint Conference on Biometrics (IJCB).

[7]  Jimmy Ba,et al.  Adam: A Method for Stochastic Optimization , 2014, ICLR.