Fine-grained Image Classification Based on Fusion of Global Contextual Features
A fine-grained image classification method that integrates global contextual features is proposed to address the problem of existing attention mechanism based fine-grained image classification models that overly focus on one or some local features of the image target when extracting image features,while ignoring the correlation between different local features and between local and global features.In this method,a region awareness module is designed to realize the feature representation of one or more local regions of the target in the image by obtaining regional feature perception coding of different regions from the image.Based on the additive attention mechanism,the global attention perception module and the context attention perception module are designed.By constructing the correlation between local features and local features as well as between the local features and the whole features,the key parts of the obscured object can be represented more effectively.Through validation and evaluation on fine-grained image classification datasets such as Stanford Cars,FGVC Aircraft,CUB-200-2011,and the self built FGVC-LAV dataset,it was verified that the proposed method can effectively mine the correlation between local features and local features,as well as between local features and global features,while extracting local features from images,thereby improving the accuracy of fi-ne-grained image classification.
fine grainedimage classificationattentionlocal featureassociation feature
万方数据
维普
