Space attention calculation method based on channel attention

Abstract

The invention discloses a space attention calculation method based on channel attention, relates to the technical field of deep convolutional neural networks, and designs a channel attention module to forward derive a channel attention map and a channel refining feature map. According to a channel attention map, a double-branch space attention module groups channels of a channel refinement feature map into important channels and secondary important channels. This organically links space attention and channel attention. And performing a series of nonlinear operations on the divided channel groups to obtain a two-dimensional space attention map. According to the method, space attention and channel attention are closely and organically linked, the method can be conveniently embedded into any current mainstream deep convolutional neural network, and the performance of a network model is improved.

Description

technical field [0001] The invention relates to the technical fields of deep convolutional neural network, channel attention and spatial attention, in particular to a method for calculating spatial attention based on channel attention. Background technique [0002] In the past few years, many scholars have been keen to explore deeper neural networks to improve the performance of the model on vision tasks. However, as the number of neural network layers deepens, its accuracy in visual tasks such as image classification and object detection does not improve significantly, that is, the improvement in network performance is not directly proportional to the deepening of the number of network layers. [0003] Researchers in the biomedical community have found that in the human visual system, the attention mechanism plays an important role in obtaining important information. Inspired by this phenomenon, many scholars try to improve the performance of the model by combining the att...

AI Technical Summary

Problems solved by technology

However, the spatial attention submodule in the CBAM module is not organically combined with the channel attention submodule, that is, the spatial attention still puts important channels and secondary important channels together for pooling operations

This will cause the original key features in the feature map to be weakened with the average pooling operation, and the performance of the network model will not be fully utilized.

Images