The spatial distribution of remote-sensing scene images is highly complex in character, so how to extract local key semantic information and discriminative features is the key to making it possible to classify accurately. However, most of the existing convolutional neural network (CNN) models tend to have global feature representations and lose the shallow features. In addition, when the network is too deep, gradient disappearance and overfitting tend to occur. To solve these problems, a lightweight, multi-instance CNN model for remote sensing scene classification is proposed in this paper: MILRDA. In the instance extraction and classifier part, more discriminative features are extracted by the constructed residual dense attention block (RDAB) while retaining shallow features. Then, the extracted features are transformed into instance-level vectors and the local information associated with bag-level labels is highlighted by the proposed channel-attention-based multi-instance pooling, while suppressing the weights of useless objects or backgrounds. Finally, the network is constrained by the cross-entropy loss function to output the final prediction results. The experimental results on four public datasets show that our proposed method can achieve comparable results to other state-of-the-art methods. Moreover, the visualization of feature maps shows that MILRDA can find more effective features.
Accurate classification of remote sensing scene images is crucial for diverse applications, from environmental monitoring to urban planning. While convolutional neural networks (CNNs) have dramatically improved classification accuracy, challenges remain due to the complex distribution of small objects, varied spatial configurations, and intra‐class multimodality in remote sensing images. In this work, we make three key contributions to address these challenges. (1) We propose the adaptive channel and coordinate relational attention network (ACDR‐CRAFF), a novel multi‐scale feature fusion framework designed to enhance feature representation across scales. (2) We introduce two innovative modules: the adaptive channel dimensionality reduction (ACDR) module, which dynamically adjusts channel representations to retain essential low‐dimensional features, and the coordinate relational attention multi‐scale feature fusion (CRAFF) module, which effectively captures and transfers spatial information between feature levels. (3) By integrating ACDR and CRAFF, our model achieves a progressive fusion of local to global features, ensuring robust feature expressiveness at multiple scales. Experimental results on four widely used benchmark datasets demonstrate that ACDR‐CRAFF consistently outperforms several state‐of‐the‐art methods, achieving significant improvements in classification accuracy and setting a new benchmark for complex remote sensing scene classification tasks. These results underscore the effectiveness of our approach in addressing the limitations of existing methods and advancing the state of the art in remote sensing image analysis.
We obtain a characterization of pair matrices A and B of order n such that where denotes the j-th largest singular values of It can imply not only some well-known singular value inequalities for sums and direct sums of matrices but also Zhan’s result related to singular values of differences of positive semidefinite matrices. In addition, some related and new inequalities are also obtained.
Remote sensing scene classification is often challenged by domain shifts arising from variations in regions and sensor types, which can significantly degrade performance when models are applied to unlabelled target domains. Multi‐source domain adaptation (MSDA) leverages multiple labelled source domains to improve generalization, but its effectiveness is still limited by discrepancies among source domains and the coarse semantic alignment of existing methods. In this work, we propose a multi‐source domain adaptation framework based on multi‐scale attention aggregation (MSAA‐MSDA). The proposed approach exploits multi‐scale semantic information to achieve more stable cross‐domain feature alignment. It combines a multi‐scale attention mechanism with a domain‐specific adaptation strategy that aligns source‐domain features with the conditional distribution of the target domain and maintains discriminative consistency through domain‐specific classifiers. In addition, two loss functions are introduced to enhance semantic transfer and reduce class confusion in the target domain. Experiments on a newly constructed cross‐domain remote sensing scene dataset show that MSAA‐MSDA consistently outperforms existing MSDA methods and achieves improved robustness and generalization.
6 May 2026 in IEEE Geoscience and Remote Sensing Letters
This paper proposes a cross-domain scene classification method that effectively leverages source-domain knowledge to classify unlabeled data, thereby addressing the challenges of limited labeled samples and weak generalization capability in remote sensing scene classification. Traditional deep domain adaptation approaches typically focus on global distribution alignment while neglecting the discrepancies among related subdomains of the same category across different domains, resulting in the loss of fine-grained semantic information. In addition, existing methods struggle to properly balance the importance of marginal and conditional distribution alignment. To overcome these limitations, we propose a Multi-scale Dual-Perception Subdomain Dynamic Adaptation (MDP-SDA) method. The proposed approach dynamically adjusts the relative weighting between marginal and conditional distributions according to the cross-domain influence of samples, enabling more flexible and adaptive distribution alignment. Furthermore, the dual-perception attention module enhances the representation capability of critical features, making the extracted features more discriminative. Experimental results demonstrate that the proposed MDP-SDA method outperforms existing deep domain adaptation approaches in overall performance.