基于TensorFlow2.0的图像分类算法研究

Research on Image Classification Algorithms Based on TensorFlow2.0

面对海量数据的增长,图像分类技术迎来机遇与挑战。深度学习能够很好的解决传统的图像分类方法精度低、耗时长等问题,成为目前主流的图像分类算法。但是深度学习模型结构及其优化算法的实现复杂且效率低,而Tensor⁃Flow2.0能够简化和快速实现深度学习模型设计,并在海量数据集上训练模型,备受研究者青睐。首先对图像分类算法、深度学习框架以及TensorFlow2.0进行介绍;然后使用TensorFlow2.0的Keras API构建一个卷积神经网络图像分类模型,结果表明,在Cifar-10数据集上,模型的准确率达76.12%;最后分析TensorFlow2.0和图像分类算法存在的不足,并对未来研究的方向进行探讨。

The massive increased data has brought great opportunities and challenges to image classification technology.Traditional image classifica⁃tion method had many problems such as low accuracy and slow speed.Deep learning was the most mainstream image classification algo⁃rithm,which was used to solve these problems.However,the implementation of deep learning model structure and its optimization algo⁃rithm was complex and inefficient.TensorFlow2.0,which is the most popular deep learning framework for researchers,can quickly imple⁃ment deep learning model design and train models on massive datasets.First,image classification algorithm,deep learning framework,and TensorFlow2.0 are introduced.Second,this paper used Keras API of TensorFlow2.0 to build an image classification model based on convo⁃lutional neural networks.The results show that the accuracy of the model on cifar-10 dataset was 76.12%.Finally,the shortcomings of Ten⁃sorFlow2.0 and image classification algorithm are analyzed,then the future research direction is discussed.