A fine-grained classification method of ship targets in remote sensing images based on spatial fusion attention

Abstract

The fine-grained classification method of remote sensing image ship targets based on spatial fusion attention, the convolution operation is performed on the target after image acquisition, and the convolution feature map is obtained; for the input convolution feature map, self-attention modeling is performed to extract The one-dimensional mean correlation coefficient between channels is obtained, and the convolution feature map under the mean correlation coefficient is constructed, and then spatial attention is used to extract the two-dimensional spatial attention weight in the convolution feature map under the mean correlation coefficient. , the cascaded combination of self-attention modeling module and spatial attention modeling module can be embedded into the classification network as a spatial fusion attention module, and the classification network can be used for final fine-grained classification. The invention improves the learning ability of the fine-grained features of the ship target through the cascading combination of the two modules and the classification network, thereby improving the accuracy of the classification result. A classification network.

Description

technical field [0001] The invention belongs to the field of image processing, and more particularly relates to a fine-grained classification method of remote sensing image ship targets based on spatial fusion attention. Background technique [0002] Accurate identification of ship targets at sea from remote sensing images helps to obtain more effective class information and improve the accuracy of identification. Significance. In recent years, with the rapid development of deep learning technology, research related to fine-grained classification has received more and more attention. However, because the current research on ship targets in remote sensing images mainly focuses on the detection of ship targets and the subsequent binary classification of background and targets. A small number of literatures specifically carry out the task of multi-classification of ship targets, but the number of classification categories is also very limited. In general, there is little rese...

AI Technical Summary

Problems solved by technology

[0005] Therefore, in order to solve the problem of low accuracy of fine-grained classification of ship targets in remote sensing images, a fine-grained classification technology of ship targets in remote sensing images based on spatial fusion attention is proposed.

Images