At present, the research of image classification mainly focuses on the detection and classification of objects. However, most of the current classification methods still rely on a large number of pixel level manual annotation samples, which often has a greater impact on practical problems. This paper will focus on the task of image classification in the weakly supervised mode. By combining attention mechanism with convolution neural network, the model can extract salient regions in the image circularly, and then use the multi branch VGG deep neural network model to adjust the classification of the extracted features, so that the model can focus on the features of the salient region, so as to improve the ability of this model. The experimental results show that the improved weak supervised learning model can achieve 87.2% and 85.7% classification accuracy under the premise of reducing the use of artificial features, which means that it can achieve better image classification effect when compared with other advanced image classification methods.
Weakly Supervised Image Classification Based on Attention Mechanism
14.10.2020
320966 byte
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
Dual-attention Focused Module for Weakly Supervised Object Localization
BASE | 2019
|Video Classification via Weakly Supervised Sequence Modeling
British Library Online Contents | 2016
|Video Classification via Weakly Supervised Sequence Modeling
British Library Online Contents | 2016
|Classification and weakly supervised pain localization using multiple segment representation
British Library Online Contents | 2014
|