Small-scale Picture Classification Based On Convolutional Neural Network VGG

The image is one of the basic tools for human understanding of the world.It has a simpler and more direct message transmission than word.Image classification recognition is an application of computer vision,and it is also an important research content in the field of artificial intelligence.Research on image classification and recognition is the way the only way which must be passed to open the door of visual applications.In academia,image classification using deep learning has reached an accuracy similar to that of the human eye,and the style transfer of art paintings using deep learning and transfer learning is an important topic in the field of image research.In the industrial field,the restoration of bank bills,face-lift withdrawals,fingerprint identification,robot AlphaGo competitions,etc.all show the application of image classification and recognition.Based on the VGG model of the convolutional neural network,this paper makes an empirical analysis of 10 categories of pictures in Food-101 and 17 Category Flower Datase.The concrete work is to briefly summarize the development of neural networks at home and abroad.After a brief description of the deep convolutional neural network,the structure of the VGG16 convolutional neural network is elaborated.Then a convolutional neural network model VGG16 with 13-layer convolution and migration learning for classifying small-scale images was constructed.In addition,a four-layer convolutional VGG model VGG6 suitable for small sample data is also constructed.Food-101 distribution is not uniform,using its 10000 data for experiments.After analysis and comparison of fine-tuned,VGG6 has a higher classification accuracy.17 The data in the Category Flower Datase is evenly distributed.After analysis of 1360 images' experiments,it is concluded that VGG16 based on transfer learning has higher accuracy.The computer environment used in this paper is Windows 7 with 8G memory and one CPU.The programming language environment is the deep learning frameworks Tensorflow and Keras.The research topic of this article has certain practical value in the context of small sample data sets and simple devices.