Fine-grained image classification method and device based on spherical features

Abstract

The invention discloses a fine-grained image classification method based on spherical features. The method comprises the following steps: extracting general features of an input image by using a trained general convolutional neural network; converting the general features of the input image into image spherical features by using deep hyperspherical inlaying; and classifying the input images through angles between the image spherical features of the input images and different fine-grained category features. The method is efficient in operation and low in time complexity. According to traditional bilinear pooling, Kronecker product operation between vectors is carried out on high-dimensional features obtained after low-dimensional general feature conversion, so that the algorithm operation amount is very huge. When spherical features are used, the time complexity is reduced from traditional bilinear pooling O (N2) to existing O (N). The network is convenient to visualize, and the fine-grained identification principle is effectively understood. The invention further provides a corresponding fine-grained image classification device based on the spherical features.

Description

technical field [0001] The invention belongs to the technical field of image classification, and more specifically relates to a fine-grained image classification method and device based on spherical features. Background technique [0002] Compared with ordinary image classification tasks, fine-grained image classification needs to distinguish many subcategories under the basic category, and more attention needs to be paid to the subtle differences between each subcategory. [0003] The main difficulty of fine-grained image classification lies in how to distinguish images of different subcategories with subtle differences. On the one hand, the image differences of different subcategories are only reflected in certain details of the image. For example, Boeing's 737Max aircraft differs only in engine details from its predecessor, the 737 aircraft. On the other hand, images of the same subcategory vary widely. For example, the 737 aircraft photographed under different lightin...

AI Technical Summary

Problems solved by technology

[0006] In view of the above defects or improvement needs of the prior art, the present invention provides a fine-grained image classification method and device based on spherical features, the purpose of which is to use spherical features (Hypersphere Feature) instead of bilinear pooling to extract image detail information , so as to solve the technical problems in the prior art that the calculation is complex and the calculation time is large, it is difficult to distinguish the categories of similar images, and there is a lack of visual understanding of image differences

Images