The Design And Implementation Of Handwritten Digit Recognition Software Based On Convolution Neural Network

Convolution neural network is the combination of artificial neural network and depth of learning two technologies which is a newer and faster network. The depth of learning technology is hierarchical mode of visual simulation of the visual system. With local receptive fields, sub-sampling and sharing weights, hidden feature extraction and classification of a combination of features in the image recognition, it has been widely applied.This thesis introduces the basic concepts of convolution neural network model, analyzing the advantages and disadvantages and the scopes of application. It has been used to handwritten digit recognition and gained a good result. The main work is as follows:(1) This thesis follows the LeNet-5 model, based on specific parameters, reducing the number of connections and network training, depending on the number of different filters and feature map, designed and implemented a number of network models using the same data set as training samples, used in digital identification.(2) Contrasted to the several network models, the results give the network structure suitable to this program, and it explains the streamlined, lightweight network model feasibility.(3) By comparing the learning characteristics of the network models and the classification results, it analyzes the relationship between the number of filters and feature maps and the recognition rate, outlining the advantages and disadvantages of each model.(4) In this thesis concluded that on the basis of the existing network models provided on hardware resources existing conditions, CNN2 model was the most suitable model, network training 50 times, accuracy was 99.63%.