基于优化分类的数据增广方法

Data augmented method based on optimized classification

为提高卷积神经网络对图像分类的正确率,对数据增广提高网络正确率进行研究,提出优化分类的数据增广方法。通过对测试集所有类别进行分析,找到分类效果不好的单类进行数据扩增,改善网络模型因训练样本少、结构复杂引起分类效果差的现象。基于Caffe深度学习框架,采用CaffeNet网络模型对Caltech-101和Corel1K数据集进行训练分析,提取图像特征信息,对测试集进行验证,将优化分类前后的测试集正确率进行对比,优化后的正确率有较大的提升。

To improve the correctness of the image classification through convolution neural network,the data augmentation was studied,and the data augmentation method based on optimizing classification was proposed.Through the analysis of all categories of the test set,the single class with poor classification was found to carry on the data amplification,and the poor classification of the network model caused by the small training samples and the complexity of the structure was alleviated.Based on the Caffe depth learning framework,the CaffeNet network model was used to train and analyze the Caltech-101 and Corel1K datasets,and the image feature information was extracted.The test set is validated,compared with the test set before optimization,the correctness of the optimized classification is greatly improved.