Research On Tactile Recognition Method Of Image For The Blind Based On Convolutional Neural Network

According to the WHO 2014 report,the number of people suffering from blindness is about39 million in the world,which has become a group that can't be ignored.With the rapid development of technology and new machine learning theory are put forward,which brings new hope for the bland man.In this paper,in order to solve those problems that which is difficult to extract the feature of the object,the weak generalization ability,the low recognition rate of the object by using traditional image processing method in blind tactile vision substitution system,proposing the application of convolution neural network algorithm to the blind image recognition and building the standard Template Library that suitable for blind tactile stimulation.The main work of this paper is as follows:1.Improving the classic handwriting recognition network model LeNet-5;Researching about those factors that impact of the performance of the convolutional neural network on MNIST database;Testing my own handwriting samples by the model that has improved.2.Experimenting with different kinds of ResNet networks on the CIFAR10 data set,and studying and analyzing its shortcut network structure;making an training data set,training and testing the set by AlexNet.We utilize the improved network model of AlexNet to identify the objects in the everyday life.3.This paper design an standard template library which can help blind identify target objects efficiency.Which open a new way for the blind to capture the information of outside efficiency.