Nan Su

In recent years, deep learning methods have made progress in Synthetic Aperture Radar (SAR) ship object detection. However, due to the small size and limited learnable features, current SAR ship detectors struggle to focus on and locate small targets effectively. Additionally, information transmission is prone to loss after feature extraction. Therefore, this paper puts forward a novel one-stage detector called MSF-SAR. Firstly, we design a Convolutional Context Attention (CCA) module, which leverages the attention mechanism to enable the model to better prioritize small target regions and enhance feature extraction capabilities. Secondly, we introduce the Multi-Scale Fusion network (MSF) structure to integrate high-level semantic features and low-level spatial features, facilitating comprehensive information interaction. Thirdly, we create the MIoU for small object detection, which not only accelerates convergence but also more accurately reflects the positional information between the true and predicted bounding boxes, thereby improving the localization and enhancing the accuracy for small objects. We carried out experiments on two datasets (LS-SSDD-v1.0, HRSID) and achieved accuracy rates of 81.0% and 93.6%. The results validate the success of the introduced approach for small SAR ship detection.

High resolution remote sensing image change captioning (RSICC) aims to understand the change content in bi-temporal high resolution remote sensing images and generate corresponding descriptive captions. By presenting change information in the form of natural language, it makes the information more intuitive and easier to communicate, which has garnered widespread attention. However, there are still two challenges in RSICC: First, most existing methods adopt a unidirectional interaction from images to text, resulting in insufficient semantic alignment between images and text, which limits method's performance. Second, in remote sensing images, there are interfering factors such as illumination and climate, leading to overall differences between bi-temporal images, which affect the recognition of change information. To address the aforementioned challenges, this paper proposes a bidirectional image-text interaction method for high resolution RSICC (BITCC). BITCC first introduces the image-to-text interaction component based on reconstruction. This approach along with the caption generation component, forms a bidirectional interaction to enhance the semantic correlation between the local change information of the images and the textual information. To address the issue of global discrepancies between bi-temporal images, a noise-based change extractor is designed, which reduces the model's focus on irrelevant factors by adding noise. Finally, the images-and-text interaction component constrains the global representations of both modalities through contrastive alignment, enhancing the global semantic consistency between the image and text in the high-level representation. Experiments on two public datasets show that our method outperforms the current state-of-the-art methods.

Recently, synthetic aperture radar (SAR) target detection methods have attracted extensive attention in the field of remote sensing and achieved impressive progress. Some existing methods primarily focus on the intuitive visual signatures in SAR images, such as contour and texture. However, due to the unique imaging mechanism of SAR, certain false alarms may exhibit visual similarities to real targets, making them difficult to distinguish in complex scenarios. Fortunately, features that appear indistinguishable in the spatial modality may exhibit significant spectrum differences. In this paper, we propose a multi-modality fusion method based on joint representation of visual and scattering features (JRVSNet) for ship detection in complex-valued SAR images. By learning the backscattering characteristics of high-resolution targets vary across sub-bands in the range and azimuth directions, incremental frequency information beyond the spatial modality can be obtained. Capturing and unifying the complementary information between the two modalities enables a more comprehensive feature representation. First, we introduce a scattering-aware dual-branch extractor (SDE) that encodes spatial modality visual features while simultaneously perceiving frequency modality backscattering variations from the complex-valued SAR data. Next, we create an optimal transport-based feature transformation module (OTFT), which transforms frequency modality features to align with spatial modality features, thus reducing discrepancies between the two heterogeneous modalities. Finally, we devise a bidirectional interactive fusion module (BIFM) that integrates incremental information from the frequency modality and fundamental visual features from the spatial modality in a bidirectional interactive manner. It fully explores the complementary clues between different modalities to assist the detector in obtaining more beneficial information. Experimental results show that, compared to other methods, our JRVSNet achieves state-of-the-art detection performance.

Change detection (CD) aims to reveal the dynamic evolution of the Earth's surface by analyzing multi-temporal remote sensing imagery, exhibiting significant application value in critical domains such as emergency management. With the continuous improvement in the spatial and temporal resolution of remote sensing imagery, traditional CD methods based on handcrafted features are facing increasing limitations. These approaches have gradually become insufficient to meet the demands of increasingly complex and diverse application scenarios. In recent years, fueled by the powerful feature representation capabilities of various network architectures, deep learning-based methods have achieved remarkable breakthroughs in the field of CD. This progress has also given rise to a diverse range of task types, including binary change detection (BCD), multi-class change detection (MCD), semantic change detection (SCD), change captioning (CC), and change detection data synthesis (CDDS). However, most existing reviews primarily focus on model architectures or supervision strategies, lacking a systematic overview from the perspective of task taxonomy. This paper presents a task-oriented survey of deep learning-based CD methods for high-resolution remote sensing imagery. It comprehensively summarizes the progress, representative approaches, and application scenarios across different task categories, while also organizing widely used datasets and evaluation metrics. Furthermore, potential future directions are discussed to provide valuable insights for subsequent research.

Image-based ship target analysis is an important task for ship monitoring and maritime security. Previous studies achieved remarkable results in ship detection and recognition tasks but mainly relied on single-modality datasets. So far, no ship re-identification dataset has been collected, restricting crossmodality re-identification research. To solve this problem, we construct the first cross-modality ship re-identification dataset, CMShipReID, which contains visible light, near-infrared, and thermal infrared modalities collected by UAV. It consists of 10 categories, about 138 identifications, and 8337 images. This dataset provides rich and high-quality data support for crossmodal ship re-identification research. We use mainstream ReID algorithms as benchmarks for dataset-related research. The dataset will be publicly available when the paper is accepted.

In recent years, reconstruction-based methods have achieved excellent detection results in the field of hyperspectral anomaly detection (HAD). These methods predominantly operate on two aspects regarding their working principles: 1) reconstructing background pixels, and 2) suppressing anomalous pixels. However, most methods only tackle the HAD task from the spatial and spectral domains, making it challenging to effectively suppress anomalies. To eliminate these issues, this paper proposes a Fourier conditional mask based mixed attention (FCMMA) method. Firstly, we propose the Fourier conditional mask based mixed attention (FCMMA) method for hyperspectral anomaly detection. FCMMA generates a conditional mask that suppresses anomalous high-frequency information and preserves background low-frequency information in the frequency domain, optimizing the anomaly detection process. In addition, to achieve fine-grained hyperspectral anomaly detection, we propose the Fourier anomaly suppression filter (FASF). FASF uses Fourier techniques to manage background and anomalies, improving detection via precise frequency decoupling. Finally, a conditional mask (CMASK) network is designed to effectively suppress anomalies. The CMASK network integrated the FASF module and the spatial-spectral multilayer perceptual machine (SSMLP) module together to enhance the transformation and representation capabilities of the generated masks, which can also help suppress anomalies. The results on five different datasets show that the proposed method is more effective and superior when compared to nine state-of-the-art methods.

Despite the successful applications of the remote sensing image in agriculture, meteorology, and geography, its relatively low spatial resolution is hindering the further applications. Super-resolution technology is introduced to conquer such a dilemma. It is a challenging task due to the variations in object size and textures in remote sensing images. To address that problem, we present SymSwin, a super-resolution model based on the Swin transformer aimed to capture a multi-scale context. The symmetric multi-scale window (SyMW) mechanism is proposed and integrated in the backbone, which is capable of perceiving features with various sizes. First, the SyMW mechanism is proposed to capture discriminative contextual features from multi-scale presentations using corresponding attentive window size. Subsequently, a cross-receptive field-adaptive attention (CRAA) module is introduced to model the relations among multi-scale contexts and to realize adaptive fusion. Furthermore, RS data exhibit poor spatial resolution, leading to insufficient visual information when merely spatial supervision is applied. Therefore, a U-shape wavelet transform (UWT) loss is proposed to facilitate the training process from the frequency domain. Extensive experiments demonstrate that our method achieves superior performance in both quantitative metrics and visual quality compared with existing algorithms.

The task of fine-grained object detection aims to accurately detect small objects in images, which has become increasingly valuable in smart agriculture applications. Existing detection methods tend to have high false detection and missed detection when dealing with small objects such as crops and weeds, resulting in low accuracy. To address this issue, we proposed FGOD-YOLOv8, which consists of a cross-scale feature fusion module, a tiny object detection head, and a multi-scale dilated attention module. The cross-scale feature fusion module ensures effective transmission of high-level semantic information and low-level detail information. The tiny object detection head captures finer-grained features. The multi-scale dilated attention module focuses on object features at different scales to further improve accuracy. According to experimental results, our proposed method demonstrates better accuracy on cropandweed datasets compared to other approaches.