Remote sensing scene classification method based on attention network scale feature fusion

Abstract

The invention discloses a remote sensing scene classification method based on attention network scale feature fusion. The method comprises the following steps: inputting a training data set comprising multiple types of remote sensing scene images; preprocessing the training data set; extracting features of multiple scales of a remote sensing image through a convolutional neural network, obtaining attention areas of the image under different scales by using a multi-choice box attention network model, cutting and scaling the attention areas, and inputting the cut and scaled attention areas into a three-layer network structure; fusing features of different scales of an original image and image features of a region of interest of the original image, and by utilizing LBP global feature expression, completing classification prediction by inputting the features into a network full-connection layer; inputting the images of the training data set into a multi-choice box attention network model MS-APN for learning training; and performing scene classification of the remote sensing image through the trained multi-choice box attention network model MS-APN. The method can extract the multi-scale and multi-angle features of the remote sensing image, and has a good scene classification and recognition effect of the remote sensing image.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing, in particular to a remote sensing scene classification method based on attention network scale feature fusion. Background technique [0002] With the continuous development of satellite sensors and remote sensing technology, a large number of high-resolution (HSR) remote sensing images are available. These high-resolution remote sensing images often contain rich spatial and semantic information, and are widely used in land use planning, smart agriculture, etc. , key target detection and military fields. Scene classification for high-resolution images is an important research topic, which aims to assign reasonable semantic labels to each image. Usually high-resolution scenes contain rich semantic information and complex spatial patterns, so their accurate classification is a challenging task. [0003] Due to the difference in acquisition time and location of remote sensin...

AI Technical Summary

Problems solved by technology

However, this research is based on the depth features of a single scale and angle of the image. At present, remote sensing images are gradually becoming more diverse in terms of target categories, feature categories, and feature scales, resulting in large changes in the scale and angle of remote sensing scenes.

[0005] Remote sensing scene images contain a variety of target scales and angles, resulting in large differences in image shape, texture, and color

The traditional convolutional neural network pays more attention to the processing of the global information of the image, and the previous research is only based on the depth features of a single scale of the image, and also ignores the differences in the extracted features due to the rotation angle of the image.

Images