Deep neural network plug-in based on attention mechanism and image recognition method

Abstract

The invention relates to a deep neural network plug-in based on an attention mechanism and an image recognition method. The plug-in is composed of two layers of LSTMs with the same size and a CNN of amulti-layer structure, wherein one layer of LSTM is used for memorizing context information and generating a mask image with remarkable characteristics, and the other layer of LSTM is used for realizing a glance function and generating classification confidence; the CNN of the multi-layer structure is used for down-sampling, extracting image features and transmitting context information to the LSTM unit. By using the CNN of the multi-layer structure and applying the attention mechanism to guide the CNN of the multi-layer structure to focus the key features of the target object in the image and remove the secondary features and the background pixels, the recognition capability is high, the target object is gradually recognized through multiple propagation, the main features of the object are gradually remembered, and the secondary features of the object are forgotten; and through multiple backward feedback, an error correction function is achieved.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence and image recognition, and in particular relates to an attention mechanism-based deep neural network plug-in and an image recognition method. Background technique [0002] At present, the deep neural network used for image recognition is mainly a convolutional neural network (CNN); however, the applicant found that the existing convolutional neural network (CNN) has the following defects in the image recognition process: [0003] 1. We know that the objects in the image usually only occupy a part of the entire space, or even a small part, and in many cases, there are a large number of background pixels in the image, and many of these pixels in the image have nothing to do with the recognition target, even Interfere with the recognition target; however, for CNN, all pixels in the image are equally weighted; [0004] 2. During the recognition process, CNN can only propagate forward ...

AI Technical Summary

Problems solved by technology

[0003] 1. We know that the objects in the image usually only occupy a part of the entire space, or even a small part, and in many cases, there are a large number of background pixels in the image, and many of these pixels in the image have nothing to do with the recognition target, even Interfere with the recognition target; however, for CNN, all pixels in the image are equally weighted;

[0004] 2. During the recognition process, CNN can only propagate forward once

Therefore, if the perturbation in the adversarial example is effective in this process, it is likely to cause the CNN to misidentify

Images