Research On Image Classification And Its Applications Based On Deep Learning

Internet technology,computer hardware and software technology has made great progress.Digital storage equipment costs continue to decline,resulting in the fact that the size of the image data shows an exponential growth,and further expands the trend.Image data show the new features of massiveness,diversity,and complexity.The image classification is a significant part of computer vision field,which is an important means to obtain effective information in the image and has a wide application in the real society.With the development of machine learning,data mining,bigdata and artificial intelligence technology in computer field,it is an urgent problem to extract the main features of massive visual images with high performance computers and classify them.Starting at 2012,deep learning becomes a research hotspot in the field of machine learning,resulting in becoming a widespread concern in academy and the industry.The method of image classification based on deep learning is more versatile than traditional image classification method.It does not need artificial feature extraction steps and many prior knowledge.It is the mainstream trend of image classification development.Based on researching a large number of references,this paper studies the image classification method based on deep learning with the concrete application scene.The main work is as follows:(1)Convolution neural network.In this paper,the composition of the convolutional neural network,the common model design,training and optimization algorithm and the selection of the classifier are studied.This paper also introduces commonly used deep learning framework Caffe.(2)Application of Convolution Neural Network in Typical Image Detection and Recognition.This paper chooses the classical vehicle image as the research object explores the application performance of the convolutional neural network in the detection and recognition task of the vehicle,focuses on the feature extraction process of the convolutional neural network.The method proposed in this paper has achieved 93.8% classification accuracy,and meet the need of practical application.(3)Application of convolutional neural network in remote sensing image scene classification.In this paper,we design different convolution neural network models to study the influence of different models for the classification of remote sensing images.In this paper,we study the remote sensing images with the characteristics of multitarget and multi-object coverage.In this paper,the method combines dual channel VGG16 models by using bilinear convolution feature fusion,and achieves 93.30% overall classification accuracy,which is 3.55% higher than that of the single VGG16 model with good classification accuracy.The deep learning can establish the complex representation model of the original data with key step lying in the design of the deep learning model.This paper establishes different convolutional neural network models for vehicle image and remote sensing image.By using the model,this paper obtains higher classification accuracy in vehicle image classification and remote sensing scene classification.It is of great significance for the future study of image classification using convolutional neural network and other deep learning methods.