RSPrompter: Learning to Prompt for Remote Sensing Instance Segmentation Based on Visual Foundation Model

IEEE Transactions on Geoscience and Remote Sensing, Год журнала: 2024, Номер 62, С. 1 - 17

Опубликована: Янв. 1, 2024

Leveraging the extensive training data from SA-1B, Segment Anything Model (SAM) demonstrates remarkable generalization and zero-shot capabilities. However, as a category-agnostic instance segmentation method, SAM heavily relies on prior manual guidance, including points, boxes, coarse-grained masks. Furthermore, its performance in remote sensing image tasks remains largely unexplored unproven. In this paper, we aim to develop an automated approach for images, based foundational model incorporating semantic category information. Drawing inspiration prompt learning, propose method learn generation of appropriate prompts SAM. This enables produce semantically discernible results concept have termed RSPrompter. We also several ongoing derivatives tasks, drawing recent advancements within community, compare their with Extensive experimental results, derived WHU building, NWPU VHR-10, SSDD datasets, validate effectiveness our proposed method. The code is publicly available at https://kychen.me/RSPrompter.

Язык: Английский

---

Segment Anything Is Not Always Perfect: An Investigation of SAM on Different Real-world Applications

Deleted Journal, Год журнала: 2024, Номер 21(4), С. 617 - 630

Опубликована: Апрель 12, 2024

Abstract Recently, Meta AI Research approaches a general, promptable segment anything model (SAM) pre-trained on an unprecedentedly large segmentation dataset (SA-1B). Without doubt, the emergence of SAM will yield significant benefits for wide array practical image applications. In this study, we conduct series intriguing investigations into performance across various applications, particularly in fields natural images, agriculture, manufacturing, remote sensing and healthcare. We analyze discuss limitations SAM, while also presenting outlook its future development tasks. By doing so, aim to give comprehensive understanding SAM’s This work is expected provide insights that facilitate research activities toward generic segmentation. Source code publicly available at https://github.com/LiuTingWed/SAM-Not-Perfect .

Язык: Английский

---

Adapting Segment Anything Model for Change Detection in VHR Remote Sensing Images

IEEE Transactions on Geoscience and Remote Sensing, Год журнала: 2024, Номер 62, С. 1 - 11

Опубликована: Янв. 1, 2024

Vision Foundation Models (VFMs) such as the Segment Anything Model (SAM) allow zero-shot or interactive segmentation of visual contents, thus they are quickly applied in a variety scenes. However, their direct use many Remote Sensing (RS) applications is often unsatisfactory due to special imaging properties RS images. In this work, we aim utilize strong recognition capabilities VFMs improve change detection (CD) very high-resolution (VHR) remote sensing images (RSIs). We employ encoder FastSAM, variant SAM, extract representations To adapt FastSAM focus on some specific ground objects scenes, propose convolutional adaptor aggregate task-oriented information. Moreover, semantic that inherent SAM features, introduce task-agnostic learning branch model latent bi-temporal RSIs. The resulting method, SAM-CD, obtains superior accuracy compared SOTA fully-supervised CD methods and exhibits sample-efficient ability comparable semi-supervised methods. best our knowledge, first work adapts VHR

Язык: Английский

---

SAM-Assisted Remote Sensing Imagery Semantic Segmentation with Object and Boundary Constraints

IEEE Transactions on Geoscience and Remote Sensing, Год журнала: 2024, Номер 62, С. 1 - 16

Опубликована: Янв. 1, 2024

Язык: Английский

---

Advances and Challenges in Deep Learning-Based Change Detection for Remote Sensing Images: A Review through Various Learning Paradigms

Remote Sensing, Год журнала: 2024, Номер 16(5), С. 804 - 804

Опубликована: Фев. 25, 2024

Change detection (CD) in remote sensing (RS) imagery is a pivotal method for detecting changes the Earth’s surface, finding wide applications urban planning, disaster management, and national security. Recently, deep learning (DL) has experienced explosive growth and, with its superior capabilities feature pattern recognition, it introduced innovative approaches to CD. This review explores latest techniques, applications, challenges DL-based CD, examining them through lens of various paradigms, including fully supervised, semi-supervised, weakly unsupervised. Initially, introduces basic network architectures CD methods using DL. Then, provides comprehensive analysis under different summarizing commonly used frameworks. Additionally, an overview publicly available datasets offered. Finally, addresses opportunities field, including: (a) incomplete supervised encompassing semi-supervised methods, which still infancy requires further in-depth investigation; (b) potential self-supervised learning, offering significant Few-shot One-shot Learning CD; (c) development Foundation Models, their multi-task adaptability, providing new perspectives tools (d) expansion data sources, presenting both multimodal These areas suggest promising directions future research In conclusion, this aims assist researchers gaining understanding field.

Язык: Английский

---

A trained Mask R-CNN model over PlanetScope imagery for very-high resolution surface water mapping in boreal forest-tundra

Remote Sensing of Environment, Год журнала: 2024, Номер 304, С. 114047 - 114047

Опубликована: Фев. 13, 2024

Small water bodies (< 0.01 km2) showing diverse limnological properties occur in great abundance across the boreal forest and tundra landscapes of Arctic Subarctic. However, their classification, geographical distribution collective importance for water, heat, nutrient, contaminant carbon cycles are still poorly constrained. One important step better understanding role evolution small fast-changing northern is to develop image analysis protocols that allow automatic remote sensing detection, delineation inventory. In this study, we set an protocol (High Latitude Water – HLWATER V1.0) based on a trained supervised Mask R-CNN deep learning model over PlanetScope imagery detection lakes ponds were absent existing datasets. Most our training dataset comprised smaller than km2 (97%) spanned wide range environmental hydrological settings, from sporadic continuous permafrost zones Canada. The was tested as fully autonomous approach eastern Hudson Bay, Nunavik (Subarctic Canada), region poses challenges given variety bodies. These mainly thaw glacial basin forest-tundra challenging optical settings influenced by vegetation or topography shadowing, revealing peat logging, fen bog pond conditions. A multi-scale validation developed using body delineations ultra-high resolution orthomosaics Unoccupied Aerial Systems. This procedure allowed sub-pixel assessment identified limitations strengths detecting large results varied according different landscape units, with mean Intersection Union (IoU) 0.5 F1 Scores 0.53 0.71 0.62 0.95. Considering 166 m2 minimum size threshold, IoU 0.7 0.91 0.76 0.83, evaluated comparing manual delineations. show high potential extension other regions Subarctic, allowing detailed inventories optically morphologically areas circumpolar North.

Язык: Английский

---

Estimating aboveground biomass of tropical urban forests with UAV-borne hyperspectral and LiDAR data

Urban forestry & urban greening, Год журнала: 2024, Номер 96, С. 128362 - 128362

Опубликована: Май 11, 2024

Язык: Английский

---

PointSAM: Pointly-Supervised Segment Anything Model for Remote Sensing Images

IEEE Transactions on Geoscience and Remote Sensing, Год журнала: 2025, Номер 63, С. 1 - 15

Опубликована: Янв. 1, 2025

Язык: Английский

---

Mapping Vegetation Changes in Mongolian Grasslands (1990–2024) Using Landsat Data and Advanced Machine Learning Algorithm

Remote Sensing, Год журнала: 2025, Номер 17(3), С. 400 - 400

Опубликована: Янв. 24, 2025

Grassland ecosystems provide a range of services in semi-arid and arid regions. However, they have significantly declined due to overgrazing desertification. In the current study, we employed Landsat time series data (TM, OLI, OLI-2) spanning from 1990 2024, combined with vegetation indices such as NDVI SAVI, along NDWI digital elevation models (DEMs), analyze land cover dynamics Ugii Lake watershed area, Mongolia. By integrating multisource remote sensing into advanced XGBoost (extreme gradient boosting) machine learning algorithm, achieved high classification accuracy, overall accuracies exceeding 94% Kappa coefficients greater than 0.92. The results revealed decline montane grasslands (−6.2%) an increase other grassland types, suggesting ecosystem redistribution influenced by climatic anthropogenic factors. Cropland exhibited resilience, recovering significant 1990s moderate growth 2024. Our findings highlight stability barren underscore pressures ecological degradation human activities. This study provides up-to-date statistical support decision-making conservation sustainable management Mongolia under changing conditions.

Язык: Английский

---

SIDEST: A sample-free framework for crop field boundary delineation by integrating super-resolution image reconstruction and dual edge-corrected Segment Anything model

Computers and Electronics in Agriculture, Год журнала: 2025, Номер 230, С. 109897 - 109897

Опубликована: Янв. 10, 2025

Язык: Английский

---