Attention fine-grained classification method based on weak supervision position positioning

Abstract

The invention discloses an attention fine-grained classification method based on weak supervision position positioning. The method specifically comprises the following steps: 101) a preprocessing step, 102) an information processing step, 103) an information progress step, 104) a feature data step and 105) a classification confirmation step. The invention provides an attention fine-grained classification method based on weak supervision position positioning, which can be used for positioning areas with relatively small inter-class differences and carrying out positioning and amplification.

Description

technical field [0001] The invention relates to the technical field of image classification, and more particularly, to a fine-grained classification method of attention based on weakly supervised position location. Background technique [0002] Fine-grained classification has attracted a lot of attention from the multimedia and computer vision communities, and the classification goal aims to distinguish visually and semantically very similar categories in general categories, such as various birds, dogs, and different car categories. The fine-grained object classification task is particularly beneficial for multimedia information retrieval and content analysis. An important observational task for fine-grained classification is that objects often play an important role in distinguishing some local part of a sub-category. For example, a dog's head is the key to distinguishing multiple types of dogs. Now studies distinguish dog species by observing dog heads, most of the fine-...

AI Technical Summary

Problems solved by technology

[0004] 1. Manually defined parts may not be the best choice for the final classification task

[0005] Second, labeling parts is much more difficult than collecting image labels. Manually cropping objects and labeling their parts is time-consuming and laborious, which is not conducive to practical applications

[0006] 3. The proposal method for unsupervised object regions will generate a large number of proposals (some as many as thousands), and the cost of computing, processing and classifying candidate regions is very high.

Images