Handwritten Digit Recognition Based on the Deep Learning of CNN
Beginning with the structure and function of deep learning of CNN,the recognition properties of the handcraft script digits are chiefly investigated through CNN,and the recognition properties of the handcraft script digits coupled with noise(such as Gaussian,salt&pepper,poisson,speckle etc.).The technology of learning transfer is adopted to improve the properties of the designed CNN,replacing the learning unit ReLu with LeakyReLu,the other layers unchanged.It improves the defect that the sigmod function is easily saturated,and the learning efficiency.
万方数据
维普
