Research On Data Set Expansion Technology Based On Generative Adversarial Network

The widespread application of artificial intelligence is of great significance to production and life,and the sudden emergence of deep learning technology has also promoted the continuous progress of various excellent algorithms in this field,but these are inseparable from the promotion of massive data.In the context of the era of big data,a large amount of valuable data is an important guarantee for the continuous improvement of the performance of artificial intelligence algorithms.Insufficient amount of data,poor quality,and imperfect data labeling are problems that plague every artificial intelligence algorithm researcher.Especially in the field of computer vision,high-quality image data is also an important prerequisite for the continuous advancement of vision algorithms.However,due to various technical reasons or cost reasons,it is difficult to obtain a large amount of image data,which has to require researchers to find more suitable methods to enhance the existing image data.The existing image enhancement methods are mainly divided into traditional image enhancement algorithms and deep learning generation methods.The former,such as image geometric transformation,image color change,and subject pixel transformation,can be enhanced on the basis of existing images,effectively improving the robustness of the algorithm and enhancing the generalization ability of the algorithm.However,such enhancement algorithms also have problems such as poor enhancement of small-scale images,or inability to generate new effective data.The latter is based on the data augmentation algorithm proposed by the generative adversarial network,which has a huge advantage in the field of image generation in recent years.This type of algorithm can not only generate rich content-enhanced images from existing image data,but can also expand a small amount of data to a certain extent,so as to solve the problem of poor image enhancement effects and difficult expansion.After analyzing the advantages and disadvantages of traditional generative adversarial networks and several improved generative countermeasure networks,this paper uses residual structure technology and codec technology to design a new generative adversarial network for data enhancement and expansion.The network is superior to several representative generative adversarial networks in terms of image quality and network training stability.To a certain extent,it solves the problems of image checkerboard effect generated by the generative adversarial network and the collapse of the network training model.And after combining the cycle consistent training process and the new loss function,the style transfer network is designed,which can effectively enhance and expand the face image data,and can improve the accuracy of traditional face detection algorithms for detecting occluded faces by more than 3%.This Verifies that the generated image improves the robustness of the algorithm.At the same time,computer-rendered virtual images are used to transfer the real style,and the target image data of weapons and armors are generated to solve the problem of data image expansion.These experiments verify the effective enhancement and expansion of the image data by the generative adversarial network,and at the same time provide an effective technical idea for the data enhancement method.