An Attention-Guided Multilayer Feature Aggregation Network for Remote Sensing Image Scene Classification

Ming Li,Gangyao Kuang,Lin Lei,Yuli Sun,Yuqi Tang

doi:10.3390/rs13163113

Abstract

Remote sensing image scene classification (RSISC) has broad application prospects, but related challenges still exist and urgently need to be addressed. One of the most important challenges is how to learn a strong discriminative scene representation. Recently, convolutional neural networks (CNNs) have shown great potential in RSISC due to their powerful feature learning ability; however, their performance may be restricted by the complexity of remote sensing images, such as spatial layout, varying scales, complex backgrounds, category diversity, etc. In this paper, we propose an attention-guided multilayer feature aggregation network (AGMFA-Net) that attempts to improve the scene classification performance by effectively aggregating features from different layers. Specifically, to reduce the discrepancies between different layers, we employed the channel–spatial attention on multiple high-level convolutional feature maps to capture more accurately semantic regions that correspond to the content of the given scene. Then, we utilized the learned semantic regions as guidance to aggregate the valuable information from multilayer convolutional features, so as to achieve stronger scene features for classification. Experimental results on three remote sensing scene datasets indicated that our approach achieved competitive classification performance in comparison to the baselines and other state-of-the-art methods.

Highlights

With the rapid development of remote sensing imaging technology, a large amount of high-resolution remote sensing images, captured from space or air, can provide rich detail information, e.g., spatial layout, shape, and texture, about the Earth’s surface
The main contributions of this paper are listed as follows: (1) We propose an attention-guided multilayer feature aggregation network, which can capture more powerful scene representation by aggregating valuable information from different convolutional layers, as well as suppressing irrelevant interference between them; (2) Instead of only considering discriminative features from the last convolutional feature map, we employed channel–spatial attention on multiple high-level convolutional feature maps simultaneously to make up for information loss and capture more complete semantic regions that were consistent with the given scene
We presented a novel attention-guided multilayer feature aggregation network in this paper, which consisted of three parts: the multilayer feature extraction module, the multilayer feature aggregation module, and the classification module

Summary

Introduction

With the rapid development of remote sensing imaging technology, a large amount of high-resolution remote sensing images, captured from space or air, can provide rich detail information, e.g., spatial layout, shape, and texture, about the Earth’s surface. This information is a significant data source and has been used to many applications, such as land use classification [1,2], land use change detection and management [3,4], geospatial object detection [5], etc. A remote sensing scene labeled “bridge” consists of five different land cover units including vehicle, trees, ship, river, and bridge. To classify this scene, we only need to pay more attention to the “bridge” regions, i.e., the red-box-covered region; the other regions can be considered

Methods

Results

Conclusion