A concept ontology triplet network for learning discriminative representations of fine-grained classes
- PDF / 2,009,971 Bytes
- 26 Pages / 439.642 x 666.49 pts Page_size
- 101 Downloads / 145 Views
A concept ontology triplet network for learning discriminative representations of fine-grained classes Guiqing He1
· Qiqi Zhang1 · Haixi Zhang1 · Yuelei Xu2 · Jianping Fan3
Received: 20 May 2019 / Revised: 20 March 2020 / Accepted: 22 May 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Triplet network is an efficient method of metric learning, but with the increase of the number of fine-grained images and sample categories, the training of Triplet network is more and more challengeable. In order to solve this problem, this paper proposes an algorithm that effectively combine Concept Ontology Structure with the Triplet network trained of Two-layer Ontology Loss. It not only utilizes semantic knowledge to guide the Concept Ontology Structure of the network, but also makes use of the relationship between the layers to make the network more effective to see the triplets, which enhances the separability of the learned features. At the same time, we also use the bilinear function jointly trained with the Triplet network to enhance the image details, further improving the performance of the network. Finally, the effectiveness of the proposed algorithm is also proved by the results of classification experiments on the fine-grained image databases - Orchid and Fashion60. Keywords Metric learning · Two-layer ontology loss · Concept ontology structure · Bilinear
Guiqing He
guiqing [email protected] Qiqi Zhang [email protected] Haixi Zhang [email protected] Yuelei Xu [email protected] Jianping Fan [email protected] 1
School of Electronics and Information, Northwestern Polytechnical University, Xi’an, China
2
Unmanned System Research Institute, Northwestern Polytechnical University, Xi’an, China
3
University of North Carolina at Charlotte, Charlotte, NC, USA
Multimedia Tools and Applications
1 Introduction As an important algorithm in machine learning, convolutional neural networks have become more and more mature [13, 17, 18, 21, 22, 30, 35, 47, 50]. Deep networks can extract the deep features of images , which can be effectively applied to other related fields such as image recognition [25, 43, 48, 49, 52]. However, as the depth of the network and the amount of computation increases, the loss function of the traditional deep learning algorithm has already shown its shortcomings, so people have begun trying to study more efficient algorithms from other aspects. Metric learning is such an efficient method of calculating network error and updating network parameters by calculating the similarity between two images. The goal of metric learning is to calculate the similarity between pictures, so that the similarity of the pictures in different categories is small, the similarity of pictures in same category is large, and finally the purpose of learning the characteristics of strong separability is achieved. From the original Siamese network [2] to the current Triplet network [16] and many variants, the method of metric learning achieves unique superiority. A Triplet network is [
Data Loading...
