基于学习率自增强的图像识别深度学习算法  

A DEEP LEARNING ALGORITHM FOR IMAGE RECOGNITION WITH SELF-ENHANCEMENT OF LEARNING RATE

作　　者：吕伏 刘铁 LüFu;Liu Tie(Department of Basic Education,Liaoning Technical University,Huludao 125105,Liaoning,China;School of Software,Liaoning Technical University,Huludao 125105,Liaoning,China)

机构地区：[1]辽宁工程技术大学基础教学部,辽宁葫芦岛125105 [2]辽宁工程技术大学软件学院,辽宁葫芦岛125105

出　　处：《计算机应用与软件》2021年第12期268-273,共6页Computer Applications and Software

基　　金：国家自然科学基金青年基金项目(51904144);国家自然科学基金项目(51874166);国家重点研发计划项目(2016YFC0801404);辽宁省自然科学基金项目(20102090)。

摘　　要：在深度学习模型中,为了进一步提高网络的收敛速度和识别精度,提出一种学习率自增强的图像识别算法。当距离极值点比较远时,以大于1的常数进行学习率自增强,加快网络向极值点附近逼近的速度。随着模型接近收敛,根据代价函数的变化情况调整学习率,学习率的变化和代价函数的变化情况成反比。在MNIST数据集和CIFAR-10数据集上进行实验。实验结果表明,结合该算法的深度学习模型在进行图像识别时,能有效地提高识别的准确率和收敛速度,并具有较好的表现能力。In the deep learning model,we propose an image recognition algorithm with self-enhancing learning rate to further improve the convergence speed and image recognition accuracy of the network.When the distance from the extreme point was relatively long,the learning rate self-enhancement was performed with a constant greater than 1,so as to speed up the network s approach to the extreme point.As the model approached convergence,the learning rate was adjusted according to the change of the cost function,and the change of the learning rate was inversely proportional to the change of the cost function.The experiments were performed on the MNIST dataset and the CIFAR-10 dataset.The experimental results show that the deep learning model combined with the algorithm in this paper can effectively improve the recognition accuracy and convergence speed in image recognition,and has better performance.

关 键 词：深度学习 学习率自增强 代价函数 图像识别 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]