Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning

Haorun Li

doi:10.47297/taposatWSP2633-456918.20210206

PDF
导出
分享
收藏
专辑

Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning

2卷第6期

浏览量: 0 Downloads: 9

Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning
Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning
Horace Mann School
2021年2卷第6期页码：116-124
纸质出版日期： 2021 ，
DOI：10.47297/taposatWSP2633-456918.20210206
Accepted：

Scan QR Code

Haorun Li. (2021). Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning. 科技理论与实践（英文版）, 2(6), 116-124.

HAORUN LI. (2021). Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning. Theory and Practice of Science and Technology, 2(6), 116-124..
Haorun Li. (2021). Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning. 科技理论与实践（英文版）, 2(6), 116-124. DOI： 10.47297/taposatWSP2633-456918.20210206.

HAORUN LI. (2021). Increasing the Efficiency of Training a Deep Neural Network for Food Photo Recognition Via Transfer Learning. Theory and Practice of Science and Technology, 2(6), 116-124.. DOI： 10.47297/taposatWSP2633-456918.20210206.

摘要

Abstract

In this research project

we created a food image classifier using machine learning. However

training an image classifier from scratch has many complications. It is difficult to acquire a vast amount of data to train an accurate model

and often the training process is very slow. We were able to overcome these difficulties by utilizing transfer learning on the RESNET-18 deep neural network as a feature extractor. The accuracy of our model increased from 6% to 60% after we employed transfer learning

while the training time decreased by half. The benefits of transfer learning were evident throughout this research

as it made our model more efficient in training while improving accuracy.