[1]舒 军,李振亚.基于改进Inception-Resnet v2图像分类算法[J].湖北工业大学学报,2020,(5):62-64.
　SHU Jun,LI Zhenya.Research on Improved Inception-Resnet v2 Image Classification Algorithm[J].,2020,(5):62-64.
点击复制

基于改进Inception-Resnet v2图像分类算法()
分享到：

《湖北工业大学学报》[ISSN:1003-4684/CN:42-1752/Z]

卷:
期数:
2020年第5期
页码:
62-64
栏目:
湖北工业大学学报
出版日期:
2020-10-28

文章信息/Info

Title:
Research on Improved Inception-Resnet v2 Image Classification Algorithm
文章编号:
1003-4684(2020)05-0062-03
作者:
舒  军;  李振亚
湖北工业大学电气与电子工程学院， 湖北 武汉 430068
Author(s):
SHU Jun; LI Zhenya
School of Electrical and Electronic Engin., Hubei Univ. of Tech., Wuhan 430070, China
关键词:
工业图片;  Inception-ResNet v2;  深度可分离卷积
Keywords:
industrial pictures;  Inception-ResNet v2;  deeply separable convolution
分类号:
TP751.1
文献标志码:
A
摘要:
根据工业生产要求，需要对不同等级的产品进行分类，Inception-ResNet v2算法模型对其分类准确率相对较低，且检测速度慢，不能够满足工业现场实时性检测分类的需要。针对这些问题提出基于Inception-ResNet v2改进的算法。该方法在网络的Inception-ResNet v2结构中采用密集串联的方式，加强原始图像的特征传递，提取更多表面特征；又对网络的卷积方式进行改进，采用深度可分离卷积，同时降低特征图数量，用以降低网络的运算量从而提升整体网络的训练速度。实验结果表明，改进后的GoogLeNet算法准确率提高7%，运算量降低19%。
Abstract:
According to the requirements of industrial production, different levels of products need to be classified. The Inception-ResNet v2 algorithm model has a relatively low classification accuracy and a slow detection speed, which cannot meet the needs of industrial field real-time detection and classification. This study proposed an improved algorithm based on Inception-ResNet v2 for these problems. This method used the dense series method in the network’s Inception-ResNet v2 structure to strengthen the feature transfer of the original image and extract more surface features. It also improved the convolution method of the network, using deep separable convolutions, while reducing features. The number of graphs was used to reduce the computation of the network and thereby improvd the training speed of the overall network. The experimental results show that the accuracy of the improved GoogLeNet algorithm is increased by 7%, and the operation amount is reduced by 19%.

参考文献/References:

[1] 洪伟,李朝锋.基于区域全卷积网络结合残差网络的火焰检测方法\[J\].激光与光电子学进展,2018,55(4):168-173.
\[2\] 何新宇,张晓龙.基于深度神经网络的肺炎图像识别模型\[J\].计算机应用,2009,39(6):1680-1684
\[3\] 黄双萍,孙超,齐龙, 等.基于深度卷积神经网络的水稻穗瘟病检测方法\[J\].农业工程学报,2017,33(20):169-176.
\[4\] 葛梦颖,于重重,周兰, 等.基于协同半监督的深度学习图像分类算法\[J\].计算机仿真,2019,36(2):196-200.
\[5\] 刘翠玲, 林珑, 于重重, 等. 基于深度学习的花生高光谱图像分类方法研究\[J\]. 计算机仿真, 2020, 37(3):189-192,283.
\[6\] 韩要昌,王洁,史通,等.基于改进GoogLeNet的遥感图像分类方法\[J\].弹箭与制导学报,2019,39(5):139-142，153.
\[7\] Balagourouchetty, Lakshmipriya，K Jayanthi, Pragatheeswaran， Pottakkat, Biju，et al. (2019). GoogLeNet based ensemble FCNet classifier for focal liver lesion diagnosis\[EBOL\]. \[2019-01-04\]www.researchgate.netpublication335956445_GoogLeNet_based_Ensemble_FCNet_Classifier_for_Focal_Liver_Lesion_Diagnosis
\[8\] Kim J H , Seo S Y , Song C G , et al. Assessment of electrocardiogram rhythms by GoogLeNet deep neural network architecture\[J\]. Journal of Healthcare Engineering, 2019, 2019:1-10.
\[9\] Smirnov E A , Timoshenko D M , Andrianov S N . Comparison of regularization methods for ImageNet classification with deep convolutional neural networks\[C\] 2013 2nd AASRI Conference on Computational Intelligence and Bioinformatics(CIB 2013). 0.
\[10\] Szegedy C , Liu W , Jia Y , et al. Going deeper with convolutions\[C\] 2015 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). IEEE, 2015. 
\[11\] He K , Zhang X , Ren S , et al. Deep residual learning for image recognition\[C\] IEEE Conference on Computer Vision & Pattern Recognition. IEEE Computer Society, 2016.
\[12\] Huang G , Liu Z , Maaten L V D , et al. Densely connected convolutional networks\[C\] CVPR. IEEE Computer Society, 2017.

备注/Memo

备注/Memo:
［收稿日期］ 2020-01-09
［第一作者］ 舒 军(1973-)， 男， 湖北荆门人，工学博士，湖北工业大学副教授，研究方向为机器视觉，深度学习
更新日期/Last Update: 2020-10-23