Abstract | ||
---|---|---|
More and more social media images are produced on the Internet. Social image classification is an important task in many fields, such as public opinion analysis and sensitive image detection. Although convolutional neural networks (CNN) have achieved amazing achievements in the field of image classification. However, social images are special multi-label images. social image classification still has a lot of space for improvement. In one aspect, we find that many methods using CNN networks for classification tasks mainly use the top-level features, which contain limited information. In the real world, social images usually contain multiple semantic objects with different scales. Using the top-level features to classify different objects may not be the most suitable. Therefore, we propose a multi-branch prediction network. Since the shallow structure of the CNN network can capture some low-level features of the image, such as boundary, color, and so on, while the deeper features can capture some high-level semantics of the image, so different branches can get different representations for multi-label image classification. Subsequently, we also use the feature fusion module for each branch feature, using other branch features as a slight supplement. An efficient attention mechanism is introduced after the fusion module of each branch, which embeds the position information of the feature map into the channel attention so that the branch prediction can locate and identify the object area more accurately. In particular, a multi-branch independent prediction strategy is adopted in this work. The branch result with the highest prediction probability is regarded as the prediction of the whole network. Experiments show that the accuracy of the classification method proposed in this paper is better than other methods for multi-label classification problems. |
Year | DOI | Venue |
---|---|---|
2021 | 10.1109/CBD54617.2021.00042 | 2021 Ninth International Conference on Advanced Cloud and Big Data (CBD) |
Keywords | DocType | ISBN |
multibranch prediction network,multilabel social image classification,social media images,convolutional neural networks,CNN,top-level features,low-level features,deeper features,feature fusion module,branch feature,multibranch independent prediction strategy,attention mechanism,Internet | Conference | 978-1-6654-0746-5 |
Citations | PageRank | References |
0 | 0.34 | 9 |
Authors | ||
4 |
Name | Order | Citations | PageRank |
---|---|---|---|
Junting Lei | 1 | 0 | 0.34 |
Wanqing Zhao | 2 | 15 | 7.07 |
Shaobo Zhang | 3 | 0 | 0.34 |
Jinye Peng | 4 | 284 | 40.93 |