Segment Anything Model (SAM) is an advanced foundational model for image segmentation, which is gradually being applied to remote sensing images (RSIs). Due to the domain gap between RSIs and natural images, traditional methods typically use SAM as a source pre-trained model and fine-tune it with fully supervised masks. Unlike these methods, our work focuses on fine-tuning SAM using more convenient and challenging point annotations. Leveraging SAM’s zero-shot capability, we adopt a self-training framework that iteratively generates pseudo-labels. However, noisy labels in pseudo-labels can cause error accumulation. To address this, we introduce Prototype-based Regularization, where target prototypes are extracted from the dataset and matched to predicted prototypes using the Hungarian algorithm to guide learning in the correct direction. Additionally, RSIs have complex backgrounds and densely packed objects, making it possible for point prompts to mistakenly group multiple objects as one. To resolve this, we propose a Negative Prompt Calibration method, based on the non-overlapping nature of instance masks, where overlapping masks are used as negative signals to refine segmentation. Combining these techniques, we present a novel pointly-supervised segment anything model, PointSAM. We conduct experiments on three RSI datasets, including WHU, HRSID, and NWPU VHR-10, showing that our method significantly outperforms direct testing with SAM, SAM2, and other comparison methods. Additionally, PointSAM can act as a point-to-box converter for oriented object detection, achieving promising results and indicating its potential for other point-supervised tasks. The code is available at https://github.com/Lans1ng/PointSAM.

Unlike optical sensors, synthetic aperture radar (SAR) sensors acquire images of the Earth’s surface with all-weather and all-time capabilities, which is vital in a situation such as a disaster assessment. However, SAR sensors do not offer as rich visual information as optical sensors. SAR-to-optical image-to-image translation generates optical images from SAR images to benefit from what both imaging modalities have to offer. It also enables multisensor image analysis of the same scene for applications such as heterogeneous change detection. Various architectures of generative adversarial networks (GANs) have achieved remarkable image-to-image translation results in different domains. Still, their performances in SAR-to-optical image translation have not been analyzed in the remote-sensing domain. This letter compares and analyzes the state-of-the-art GAN-based translation methods with open-source implementations for SAR-to-optical image translation. The results show that GAN-based SAR-to-optical image translation methods achieve satisfactory results. However, their performances depend on the structural complexity of the observed scene and the spatial resolution of the data. We also introduce a new dataset with a higher resolution than the existing SAR-to-optical image datasets and release implementations of GAN-based methods considered in this letter to support the reproducible research in remote sensing.

Object detection in remote sensing images relies on a large amount of labeled data for training. However, the increasing number of new categories and class imbalance make exhaustive annotation impractical. Few-shot object detection (FSOD) addresses this issue by leveraging meta-learning on seen base classes and fine-tuning on novel classes with limited labeled samples. Nonetheless, the substantial scale and orientation variations of objects in remote sensing images pose significant challenges to existing few-shot object detection methods. To overcome these challenges, we propose integrating a feature pyramid network and utilizing prototype features to enhance query features, thereby improving existing FSOD methods. We refer to this modified FSOD approach as a Strong Baseline, which has demonstrated significant performance improvements compared to the original baselines. Furthermore, we tackle the issue of spatial misalignment caused by orientation variations between the query and support images by introducing a Transformation-Invariant Network (TINet). TINet ensures geometric invariance and explicitly aligns the features of the query and support branches, resulting in additional performance gains while maintaining the same inference speed as the Strong Baseline. Extensive experiments on three widely used remote sensing object detection datasets, i.e., NWPU VHR-10.v2, DIOR, and HRRSD demonstrated the effectiveness of the proposed method.

Object detection in remote sensing imagery usually suffers from inaccurate target localization and bounding box regression uncertainty, mainly due to the varying sizes of objects and the complexity of the background. Most detectors address these challenges by adding various feature extraction modules, which increases the size and computational burden of the network. In this article, we propose a more accurate and faster detector named AFDet, which is composed of two parts: a backbone pretrained on ImageNet and a head that includes a center prediction branch (CPB), semantic supervision branch (SSB), and boundary estimation branch (BEB). CPB produces a keypoint heatmap using an elliptical Gaussian kernel to adapt to the ground truth with a large aspect ratio. SSB, which is used only during training, extracts extra keypoint features from boundary and interior points rather than only from the center point, thereby improving the quality of object localization. BEB predicts the distributions of the bounding box in four directions, which is further supervised by the focus loss, and the gather loss raises the box prediction accuracy. To verify the effectiveness and robustness of AFDet, we conduct extensive experiments on three widely used optical remote sensing object detection datasets, i.e., NWPU VHR-10, DIOR, and HRRSD, for which AFDet achieves state-of-the-art results.

In conventional remote sensing change detection (RSCD) procedures, extensive manual labeling for bi-temporal images is first required to maintain the performance of subsequent fully supervised training. However, pixel-level labeling for CD tasks is very complex and time-consuming. In this paper, we explore a novel self-supervised contrastive framework applicable to the RSCD task, which promotes the model to accurately capture spatial, structural, and semantic information through domain adapter and hierarchical contrastive head. The proposed SSLChange framework accomplishes self-learning only by taking a single-temporal sample and can be flexibly transferred to main-stream CD baselines. With self-supervised contrastive learning, feature representation pre-training can be performed directly based on the original data even without labeling. After a certain amount of labels are subsequently obtained, the pre-trained features will be aligned with the labels for fully supervised fine-tuning. Without introducing any additional data or labels, the performance of downstream baselines will experience a significant enhancement. Experimental results on 2 entire datasets and 6 diluted datasets show that our proposed SSLChange improves the performance and stability of CD baseline in data-limited situations. The code of SSLChange is available at https://github.com/MarsZhaoYT/SSLChange.