篇名 | A Data Augment Method for Fine-grained Recognition |
---|---|
卷期 | 29:3 |
作者 | Yin Zhang 、 Zhifeng Hu 、 Shenjing Tian |
頁次 | 012-018 |
關鍵字 | convolutional neural network 、 fine-grained recognition 、 region proposal 、 EI 、 MEDLINE 、 Scopus |
出刊日期 | 201806 |
DOI | 10.3966/199115992018062903002 |
Fine-grained recognition is a very challenge problem, because of the similarity between different subcategories and scarce training data. Even in the same subcategories, there can be some differences due to the distinct color and pose of objects. We focus our thoughts on the details of specific object parts to settle these limitations. We propose a model for finegrained recognition by taking advantage of deep Convolutional Neural Network (CNN) combined with bottom-up region proposals. Our method evaluates these proposals and utilize the evaluated proposals to determine the subcategory of the object. Our final result shows that our methods can increase the accuracy by 2-3% on average.