Image-enhanced Adaptive Learning Rate Handwritten Vision Processing Algorithm Based on CNN
Keqin Chen, Kun Zhu, Meng Meng · 2019
Handwritten images are an important source of information in the human world, and machine learning has always been an important method for dealing with handwritten visual problems. Most of today's machine learning uses shallow structures such as SVM (Support Vector Machine) and kernel regression. These algorithms cannot be complex. The image information is more accurately predicted, and the recognition method of machine learning has quite high requirements on the quality of the image. This paper takes the MNIST data set of the National Institute of Standards and Technology as an example to enhance the data of the handwritten image data. The Convolutional Neural Network (CNN) algorithm builds the model and uses the adaptive learning rate algorithm (Adadelta) to enhance the model accuracy. The experimental results on the MNIST dataset show that the proposed method can effectively improve the recognition of handwritten images and quickly and correctly identify handwritten images. Its accuracy of 99.6% is significantly better than neural network and logistic algorithm.