Based on convolution neural network model,this paper proposes an improved automatic image data enhancement strategy.Focusing on the discretization of the original strategy search space and the instability of super parameter optimization in model training process,the original strategy is improved by reducing the complexity of search space strategy,optimizing the training process of sub-network model,and selecting more efficient enhanced random search algorithm to realize super parameter optimiza-tion.And the different types of datasets are used to verify.The results show that the proposed enhancement strategy improves the ac-curacy of image classification on CIFAR-10,CIFAR-100 and ImageNet datasets,it achieves advanced experimental results.
关键词
卷积神经网络/图像数据增强/超参数优化/增强随机搜索/图像分类
Key words
convolution neural network/image data enhancement/super parameter optimization/enhanced random search/image classification
维普
万方数据