A Dense Residual Correlation Attention Multi-scale Network for Remote Sensing Scene Classification

Abstract

Remote sensing scene classification is a crucial task in the field of remote sensing technology, as it helps identify and categorize various geographical features on the Earth's surface. Recently, convolutional neural networks (CNNs) have been widely employed for scene classification in remote sensing images, leading to significant performance improvements. However, current CNN methods have certain limitations, primarily manifested in their tendency to overly focus on extracting high-level features, often overlooking essential information at different levels within the images. Furthermore, some approaches attempt to directly merge features from different levels, potentially resulting in issues of information redundancy or mutual exclusion. To address these challenges, this paper proposes an efficient method for remote sensing scene classification. Our approach involves three main contributions. First, we introduce a feature extraction module called multi-stream feature extraction, which effectively utilizes features at various scales and extracts features at different levels and depths. Second, we propose a dense residual connection feature fusion technique that enables comprehensive feature interactions and improves overall accuracy. Additionally, we introduce a correlated attention network to learn powerful feature representations at multiple levels, further improving classification performance. The method outperforms existing algorithms in terms of effectiveness and accuracy, achieving state-of-the-art results on four widely used remote sensing scene classification benchmarks.