Conventional remote sensing image retrieval (RSIR) systems perform single-label retrieval with a single label to represent the most dominant semantic content for an image. Improved spatial resolution dramatically boosts the remote sensing image scene complexity, as a remote sensing image always contains multiple categories of surface features. In this case, a single label cannot comprehensively describe the semantic content of a complex remote sensing image scene and therefore results in poor retrieval performance in practical applications. As a result, researchers have begun to pay attention to multi-label image retrieval. However, in the era of massive remote sensing data, how to increase retrieval efficiency and reduce feature storage while preserving semantic information remains unsolved. Considering the powerful capability of hashing learning in overcoming the curse of dimensionality caused by high-dimensional image representation in Approximate Nearest Neighbor (ANN) search problems, we propose a new semantic-preserving deep hashing model for multi-label remote sensing image retrieval. Our model consists of three main components: (1) a convolutional neural network to extract image features; (2) a hash layer to generate binary codes; (3) a new loss function to better maintain the multi-label semantic information of hash learning contained in context remote sensing image scene. As far as we know, this is the first attempt to apply deep hashing into the multi-label remote sensing image retrieval. Experimental results indicate the effectiveness and promising of the introduction of hashing methods in the multi-label remote sensing image retrieval.

As the degradation factors of remote sensing images become increasingly complex, it becomes challenging to infer the high-frequency details of remote sensing images compared to ordinary digital photographs. For super-resolution (SR) tasks, existing deep learning-based single remote sensing image SR methods tend to rely on texture information, leading to various limitations. To fill this gap, we propose a remote sensing image SR algorithm based on a multi-scale texture transfer network (MTTN). The proposed MTTN enhances the texture feature information of reconstructed images by adaptively transferring texture information according to the texture similarity of the reference image. The proposed method adopts a multi-scale texture-matching strategy, which promotes the transmission of multi-scale texture information of remote sensing images and obtains finer-texture information from more relevant semantic modules. Experimental results show that the proposed method outperforms state-of-the-art SR techniques on the Kaggle open-source remote sensing dataset from both quantitative and qualitative perspectives.

Infrared and visible image fusion (IVIF) aims to integrate the information from source images into a single image, achieving a comprehensive representation of the scene. Existing methods typically focus on either the correlation or complementarity (uncorrelated) between different modalities; however, both aspects are equally important for image fusion. Hence, in this paper, we propose a novel IVIF method based on feature decomposition to enhance both the correlation and complementarity of information, which is termed as FDFuse. Firstly, FDFuse employs dual-branch encoders with distinct structures to decompose the features of source images into modality-shared and modality-specific components. We then introduce differently designed cross-attention mechanisms to enhance the correlation between modality-shared features and complementarity between modality-specific features. We further propose different fusion strategies for these two types of features to preserve important information in the fused results. Finally, the fused features are concatenated and fed into an image decoder to generate the final fused image. Extensive experiments conducted on public datasets demonstrate that FDFuse outperforms state-of-the-art IVIF methods, exhibiting superior performance across multiple metrics. We also show that FDFuse can improve the performance in downstream object detection tasks under the same experiment settings. The code is available at https://github.com/cheng411523/ FDFuse.

In this paper, a novel adaptive bandwidth mean shift algorithm toward 2D object detection (ABMSOD) is proposed. It can not only identify whether an object of certain classes exists or not, but also get the scale and orientation besides position very fast. The feature histogram weighted by a kernel with adaptive bandwidth is used for representing the target object model and the candidate object model. Features such as color, texture, gradient and so on can be used. A single piece of image is enough to build a model by calculating the weighted feature histogram of the object in the image. There is no exhaustive training. The similarity of the target model and the candidate model is measured by the Bhattacharyya coefficient. After gathering the models of targets, the algorithm can be used for object detection. In the first step, the algorithm searches the whole image to find the rough positions of possible candidate objects. If the similarities are all below a certain threshold, it reports no object existence. If the similarities are above the threshold, the second step or the adaptive bandwidth mean shift search step is executed to find the best position, orientation and scale of these objects. Experiments show that it successfully detects the position, scale and orientation of objects.

To enhance the accuracy of remote sensing data analysis, cloud detection from the complex ground environment is crucial. We refer to clouds that are easily confused with similar background as weak targets clouds, including thin clouds, tiny clouds, cloud boundaries, clouds with snow’s existence or highlighted background’s existence. This paper proposes a coarse-to-fine cloud detection network for weak target recognition. The network consists of two subnetworks: the Scalable Weak Target Feature Extraction Subnetwork (SWTFES) and the Cascade Weak Target Refinement Subnetwork (CWTRS). SWTFES incorporates a Multi-scale Feature Extraction Module (MFEM) with different scale receptive field branches and an Attention-based Cross-layer Fusion Module (ACFM) to characterize cloud at various scales. The improved reverse attention operation and the Cascade Group Reverse Attention Module (CGRAM) serve as the guiding principles in CWTRS, driving the network to progressively add and refine the weak target’s details to distinguish it from the complex background surface. We evaluate our methodology on four cloud datasets with various resolutions, varying from 0.5m to 16m, and different satellites (including Gaofen-1 WFV, Sentinel-2, Gaofen-2, WorldView-2). The experimental results demonstrate that WodNet has achieved excellent results in cloud detection in a variety of complex scenarios, compared to other models, performing SOTA in four challenging datasets.