Fine-grained image classification method and system based on attention mechanism and cut-and-fill

Abstract

The present invention relates to the field of deep learning and image classification, specifically a fine-grained image classification method based on attention mechanism and cutting and filling. The method includes: constructing a convolutional neural network model; inputting the original image into the convolutional neural network model, Combining the attention mechanism to obtain the attention image; cutting the attention image to obtain sub-images, filling the sub-images, and downsampling to obtain the filling image; input the attention image and filling image into the convolutional neural network model, and pass through the linear layer respectively and the softmax classifier to obtain the probability value of the corresponding category; select the maximum probability value, judge the classification result according to the maximum value; mark the classification label on the original image according to the result; the present invention segments the attention image of the original image, and then fills it, It destroys the correlation between various parts, makes the network pay more attention to local features, ensures that the high-level semantics will not be destroyed, and greatly reduces the use of parameters and training time.

Description

technical field [0001] The present invention relates to the field of deep learning and image classification, in particular to a fine-grained image classification method and system based on attention mechanism and cut-and-fill. Background technique [0002] Fine-grained image classification is an important task in the field of computer vision, and this task is very challenging. Fine-grained image classification is different from general-purpose object recognition, because coarsely observed fine-grained objects (such as birds, car models, and airplanes, etc.) are visually similar, so fine-grained image classification is very dependent on the locality of the object feature. General classification methods are also applicable to fine-grained image classification, so how to better learn local features is the key to fine-grained image classification. [0003] In recent years, the attention mechanism has been widely used in fine-grained image classification networks, such as the e...

AI Technical Summary

Problems solved by technology

[0004] However, this method destroys the spatial structure of the object and causes certain damage to the high-level semantic information. At the same time, the use of the class activation map requires the training of category weights to obtain additional attention parts to suppress other categories, resulting in image classification. There is an error

Images