An Image Classification Approach based on Deep Learning and Transfer Learning. Issue 7 (March 2020)

• Record Type: Journal Article
• Authors: Han, Kaixu; He, Jinxin; Wang, Yongzhi; Xiong, Yue; Zhang, Chi
• Is Part Of: IOP conference series. Volume 768:Issue 7(2020)
• Publication Date: 2020-03
• Subjects: Materials science -- Periodicals; 620.1105
• DOI: 10.1088/1757-899X/768/7/072055
• Languages: English
• ISSNs: 1757-8981

Abstract

Abstract: Target recognition of high-resolution images is an important direction of today's classification technology, and some classification models have emerged, but there are still many technical problems to be solved. The main source of this paper is the Google Photo Collection, which includes five types of city daisy, rose, tulip, dandelion and sunflower. Using CNN (Convolutional Neural Networks) deep learning model and Google's Inception transfer learning model to train and classify the sample images, the final accuracy of the overall test set can reach 88.3%; However, the accuracy of training without transfer learning is only 60.2%. Thus, it is more efficient to combine deep learning with transfer learning to image classifications.