| With the continuous development of Internet technology and mobile terminal equipment,the image resources are increasing day by day.The huge image resources put forward urgent requirements for the current image recognition and classification and other computer vision technologies.For a long time,the image recognition were mainly carried out in a text-based way.However,due to the complexity of the information contained in the image and the existence of the "semantic gap",the text-based approach alone cannot meet the requirements of high accuracy.In order to solve these problems,image semantic automatic annotation technology came into being.and how to effectively improve the efficiency of image recognition and accuracy become the key of this technology.To this end,this paper mainly does the following work:(1)In this paper,the deep convolutional neural network technology and GPU parallel computing technology are used to realize the classification and recognition of images.Moreover,through experiments on data sets of different sizes,the recognition efficiency and recognition accuracy of the current classical deep convolutional neural network model are analyzed and compared in detail.(2)After comparing and analyzing the current network model,this paper proposes an improved structure of deep convolutional neural network model to solve the problems such as low learning efficiency and poor recognition accuracy.Experiments on ImageNet,cifar-10 and Mnist data sets have proved that compared with the traditional CNN network model,the network model proposed in this paper can greatly improve the classification accuracy with less network depth and less training time.(3)Based on the improved deep convolutional neural network model,the improved model is further optimized and adjusted to adapt to the multi-label classification problem.Based on the recognition and classification results of the network model,the automatic semantic annotation of the target image is realized. |
PDF Full Text Request