Abstract:
Fine-grained classification is the classification of subclasses. Since subclasses usually only have subtle differences, the local areas become more important in fine-grai...Show MoreMetadata
Abstract:
Fine-grained classification is the classification of subclasses. Since subclasses usually only have subtle differences, the local areas become more important in fine-grained classification. Aiming at the problem that most existing methods cannot obtain reliable local areas, we propose a priority selection algorithm of local areas. Our method first uses the attention model to obtain candidate local areas. Then, according to the classification ability differences of the candidate areas in different subclasses, the geometric relationship between the areas is determined by removing the unnecessary area and the abnormal position area. Finally, the selected areas are classified using a convolutional neural network (CNN). The algorithm we proposed achieves a correct rate of 87.0% in the CUB-Birds bird dataset and 90.2% in the FGVC-Aircraft aircraft dataset, reaching the current state-of-the-art level.
Published in: 2019 12th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI)
Date of Conference: 19-21 October 2019
Date Added to IEEE Xplore: 23 January 2020
ISBN Information: