Implementation Of Recognition System Based On Tensorflow Deep Learning And Research On Optimization Of Mobile Terminal Recognition

As the basis of recognition technology,image recognition is an enduring topic in the field of artificial intelligence.It is a technique for image processing,analysis and understanding to recognize pictures with a computer.So far,image recognition technology has been applied to many areas,such as medical image analysis,remote sensing image recognition,fingerprint identification.The traditional image recognition method is divided into five parts,including picture input,preprocessing,feature extraction,classification and matching,and the basic idea is template comparison.On the other hand,with the development of deep learning,especially the convolutional neural network in the field of image applications become more mature,image recognition technology has been an unprecedented development.Nowadays,people can use the personal computer to easily complete the image recognition task,with the help of neural networks.At the same time,with the popularity of smart phones and hardware innovation,deep learning and mobile devices continue to show the trend of combining.In September 2016,TensorFlow,the depth learning platform,began providing deep technical support for mobile devices.This makes it possible to use deep learning results on mobile devices.However,because the technology has just started,there are still many places can be improved that use the phone to identify the picture with deep learning.The paper has compiled the development and research results of artificial neural network both at home and abroad,introduced the basic flow and the key algorithm using neural network for image recognition,and use the relevant methods and interfaces provided by TensorFlow to implement a recognition system that recognizes five kinds of flowers.The paper analyzes the factors that affect the accuracy of system identification,including the training data set model,training method,data set size,different data enhancement methods,the process of moving the model into the mobile device process,and the way of getting the picture.Analyze the key factors in them and use them to build recognition programs that can identify 12 plants,thereby improving the reliability of identification.In the paper,we prove that to retrain a pre-training model can obtain a new reliable classification model if the pictures waiting training are similar to the pictures used to construct the pre-training model.By optimizing the training process and the key steps involved in the identifying process,the accuracy of mobile device identification can be improved.