Article
Multidisciplinary Sciences
Fanglin Bao, Xueji Wang, Shree Hari Sureshbabu, Gautam Sreekumar, Liping Yang, Vaneet Aggarwal, Vishnu N. N. Boddeti, Zubin Jacob
Summary: Machine perception uses advanced sensors to collect information for situational awareness. State-of-the-art machine perception faces difficulties with increasing number of intelligent agents. Exploiting omnipresent heat signal could be a new frontier for scalable perception. The proposed heat-assisted detection and ranging (HADAR) overcomes the challenge of ghosting and shows promising results compared to AI-enhanced thermal sensing.
Article
Multidisciplinary Sciences
Renato Juliano Martins, Emil Marinov, M. Aziz Ben Youssef, Christina Kyrou, Mathilde Joubert, Constance Colmagro, Valentin Gate, Colette Turbil, Pierre-Marie Coulon, Daniel Turover, Samira Khadir, Massimo Giudici, Charalambos Klitis, Marc Sorel, Patrice Genevet
Summary: The article introduces an advanced LiDAR technology that achieves a large field of view and high frame rate by using ultrafast low FoV deflectors and large area metasurfaces, enabling simultaneous peripheral and central imaging zones. The use of this technology with advanced learning algorithms offers potential improvements in perception and decision-making processes of ADAS and robotic systems.
NATURE COMMUNICATIONS
(2022)
Article
Engineering, Marine
Chen Chen, Ying Li
Summary: A berthing information extraction system based on 3D LiDAR technology has been developed, which utilizes principal component analysis to calculate a ship's heading and normal vector, identify feature points of the bow and stern, and obtain segments passing through these points through region growing. Qualitative and quantitative analyses of the distance, velocity, and approach angle of dynamic ship targets relative to docks are conducted to ensure safe ship berthing, demonstrating the feasibility and effectiveness of the proposed method.
JOURNAL OF MARINE SCIENCE AND ENGINEERING
(2021)
Article
Geochemistry & Geophysics
Juepeng Zheng, Wenzhao Wu, Shuai Yuan, Haohuan Fu, Weijia Li, Le Yu
Summary: Providing accurate and timely oil palm information is crucial for economic development and ecological significance. However, large-scale and cross-regional oil palm tree detection is challenging due to the variety and volume of data, as well as environmental heterogeneity. This study proposes a new multisource domain generalization method that achieves promising performance in unknown target domains.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Review
Nanoscience & Nanotechnology
Inki Kim, Renato Juliano Martins, Jaehyuck Jang, Trevon Badloe, Samira Khadir, Ho-Youl Jung, Hyeongdo Kim, Jongun Kim, Patrice Genevet, Junsuk Rho
Summary: This review discusses the technological challenges of applying nanophotonics in LiDAR, the basic principles of LiDAR and overcoming hardware limitations, the characteristics of nanophotonic platforms, and the future trends in integrating nanophotonic technologies into commercially viable, fast, ultrathin, and lightweight LiDAR systems.
NATURE NANOTECHNOLOGY
(2021)
Article
Geochemistry & Geophysics
Yuanjun Xing, Jiawei Jiang, Jun Xiang, Enping Yan, Yabin Song, Dengkui Mo
Summary: Lightweight change detection models are important for industrial applications and edge devices. Developing such models with high accuracy and reduced size is a challenge. Existing methods oversimplify the model architecture, resulting in loss of information and reduced performance. To address this, we propose LightCDNet, a lightweight change detection model that effectively preserves input information. Evaluation on the LEVIR-CD dataset shows that LightCDNet achieves comparable or better performance while being much smaller in size compared to state-of-the-art models.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2023)
Article
Remote Sensing
Erzhuo Che, Michael J. Olsen, Jaehoon Jung
Summary: Mobile LiDAR technology is widely used for its efficiency in data capturing and processing, particularly in ground filtering and road detection tasks. The proposed method in this study, involving data preprocessing, trajectory reconstruction, segment-based filtering, and road detection, demonstrates robustness, effectiveness, and efficiency when tested on various datasets. Performance evaluation shows high accuracy with F (1) score and Root Mean Square Error values of over 98% for both rural and suburban scenes.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2021)
Article
Environmental Sciences
Dekker Ehlers, Chao Wang, John Coulston, Yulong Zhang, Tamlin Pavelsky, Elizabeth Frankenberg, Curtis Woodcock, Conghe Song
Summary: The majority of the aboveground biomass on the Earth's land surface is stored in forests. However, accurate estimation of forest aboveground biomass (FAGB) remains challenging. This study proposed a new conceptual model using remotely sensed data to map FAGB. The model includes height metrics as the most important variables for estimating FAGB.
Article
Geochemistry & Geophysics
Xudong Zhao, Ran Tao, Wei Li, Wilfried Philips, Wenzhi Liao
Summary: This article proposes a fractional Gabor convolutional network (FGCN) for efficient feature fusion and comprehensive feature extraction. The FGCN uses Octave convolution layers for multisource information fusion and fractional Gabor convolutional (FGC) layers for extracting multiscale, multidirectional, and semantic change features. Experimental results demonstrate the effectiveness and superiority of the proposed FGCN in multisource classification tasks.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2022)
Article
Environmental Sciences
Jianhui Luo, Qiang Chen, Lei Wang, Yixiao Huang
Summary: A novel multi-difference image fusion change detection method based on a visual attention model (VA-MDCD) is proposed for very-high-resolution (VHR) remote sensing images. The method constructs difference images, calculates difference saliency images, fuses saliency images, and applies threshold segmentation to obtain the final change detection map. Experimental results show that the proposed method outperforms classical methods in terms of missed alarms and false alarms, demonstrating its strong robustness and generalization ability.
Article
Chemistry, Multidisciplinary
Qian Gao, Lei Fan, Siyu Wei, Yishun Li, Yuchuan Du, Chenglong Liu
Summary: This study utilized high-precision LiDAR technology to obtain three-dimensional point cloud data for a 25 km road section in Shanghai. Variance analysis and the Kruskal-Wallis test were conducted to evaluate the differences in the distribution and variability of pavement roughness. The findings revealed significant differences in the international roughness index (IRI) among the survey lines within the road section. Additionally, it was recommended to consider the lateral distribution of pavement roughness and regulate the detection length in road performance evaluations to enhance reliability and facilitate more accurate maintenance decision making.
APPLIED SCIENCES-BASEL
(2023)
Article
Environmental Studies
Randhi Atiqi, Muhammad Dimyati, Ahmad Gamal, Rizki Pramayuda
Summary: Economic growth and demographic advantages have contributed to the high rate of urbanization in Indonesia, although property tax revenues are currently lower than those of G20 countries. This can be partially attributed to the limited capacity of local governments in determining building values for tax calculations. The use of LIDAR mapping can help improve local tax performance by providing quick estimates of building prices.
Article
Engineering, Electrical & Electronic
Yong Wang, Xinhui Liu, Quanxiao Zhao, Haiteng He, Zongwei Yao
Summary: This article presents a research on road target detection based on deep learning by combining image data of vision with point cloud data from light detection and ranging (LiDAR). The proposed approach utilizes a MY3Net network, which integrates Mobilenet v2 and YOLO v3, to detect RGB images and densified depth maps. A decision-level fusion model is proposed to integrate the detection results of RGB images and depth maps. Experimental results show high detection accuracy even under complex illumination conditions.
IEEE SENSORS JOURNAL
(2023)
Article
Computer Science, Artificial Intelligence
Xinyu Zhang, Zhiwei Li, Zhenhong Zou, Xin Gao, Yijin Xiong, Dafeng Jin, Jun Li, Huaping Liu
Summary: Noise is always a significant issue in object detection, causing confusion in model reasoning and reducing the informativeness of data. To address this, we propose a universal uncertainty-aware multimodal fusion model that adaptsively selects valid information from both point clouds and images. Our model reduces randomness and generates reliable output, as demonstrated by experiments on the KITTI 2-D object detection dataset.
IEEE TRANSACTIONS ON NEURAL NETWORKS AND LEARNING SYSTEMS
(2023)
Article
Chemistry, Analytical
Francisco Soares Neves, Rafael Marques Claro, Andry Maykol Pinto
Summary: A perception module is crucial for a modern robotic system, with the most common sensor choices being vision, radar, thermal, and LiDAR for environmental awareness. Relying on a single source of information is susceptible to specific environmental conditions, so using different sensors is essential for robustness. This paper proposes an early fusion module that combines visual, infrared, and LiDAR modalities to detect offshore maritime platforms for UAV landing. The early fusion-based detector achieves high detection recalls of up to 99% in all cases of sensor failure and extreme weather conditions, with an inference duration below 6 ms.
Article
Environmental Sciences
Getachew Workineh Gella, Lorenz Wendt, Stefan Lang, Dirk Tiede, Barbara Hofer, Yunya Gao, Andreas Braun
Summary: This study investigates the use of a deep convolutional neural network-based Mask R-CNN model for dwelling extractions in IDP/refugee settlements. The model was trained using transfer learning from historical images, and showed better performance compared to training from scratch.
Article
Computer Science, Information Systems
Martin Sudmanns, Hannah Augustin, Brian Killough, Gregory Giuliani, Dirk Tiede, Alex Leith, Fang Yuan, Adam Lewis
Summary: This article explores the technology landscape for managing big Earth observation data and the importance and challenges of local EO data cubes. By analyzing examples of global and local EO data cubes, it is found that local EO data cubes can benefit various stakeholders but require technical developments such as establishing global and cloud-native EO data streaming mechanisms. The article argues that blurring the dichotomy between global and local aligns with the vision of the Digital Earth.
Article
Computer Science, Information Systems
Andrea Baraldi, Luca D. Sapia, Dirk Tiede, Martin Sudmanns, Hannah Augustin, Stefan Lang
Summary: This paper focuses on the convergence between Earth observation Big Data and Artificial General Intelligence. It compares existing EO optical sensory image-derived Level 2/Analysis Ready Data (ARD) products and processes and proposes new requirements for harmonization and standardization. The paper presents original contributions in semantic-enriched ARD co-product pair requirements, ARD process requirements, ARD processing system design, and computer vision subsystem design.
Article
Computer Science, Artificial Intelligence
Anh Vu Vo, Michela Bertolotto, Ulrich Ofterdinger, Debra F. Laefer
Summary: Street view imagery databases, such as Google Street View, Mapillary, and Karta View, coupled with computer vision algorithms, can be used to analyze urban environments at scale. This paper investigates the use of street view imagery data to identify building features indicating vulnerability to flooding, such as basements and semi-basements. It discusses building features, data sources, computer vision algorithms, and methods for reconstructing geometry representations from images. Preliminary experiments using Mapillary images confirm their usability for detecting basement features and geolocating them.
KUNSTLICHE INTELLIGENZ
(2023)
Article
Geography, Physical
Wufan Zhao, Hu Ding, Jiaming Na, Mengmeng Li, Dirk Tiede
Summary: This study proposes a gradient-based self-supervised learning network to extract geometric information from non-labeled images, and uses novel local implicit constraint layers to refine high-resolution features in height estimation. Experimental evaluation shows that the proposed method outperforms other baseline networks in height map estimation.
INTERNATIONAL JOURNAL OF DIGITAL EARTH
(2023)
Article
Environmental Sciences
Alex Levering, Diego Marcos, Jasper van Vliet, Devis Tuia
Summary: Remote sensing images can be used to predict the liveability of neighborhoods on a large scale by predicting intermediate domain scores. Domains directly visible in aerial images (physical environment, buildings) are easier to generalize than those predicted through proxies (population, safety, amenities). Our model is able to predict the liveability of different types of neighborhoods with varying accuracy.
REMOTE SENSING OF ENVIRONMENT
(2023)
Article
Biodiversity Conservation
Ilan Havinga, Diego Marcos, Patrick Bogaart, Dario Massimino, Lars Hein, Devis Tuia
Summary: Social media provides new opportunities to map cultural ecosystem services (CES) related to biodiversity at large scales, but it is still challenging to understand people's preferences in relation to these CES using these novel data.
Article
Chemistry, Analytical
Siyuan Chen, Xiangding Zeng, Debra F. Laefer, Linh Truong-Hong, Eleni Mangina
Summary: Imagery from Unmanned Aerial Vehicles (UAVs) can generate 3D point cloud models, but the final data quality is influenced by various factors, such as flight altitude, camera angle, overlap rate, and data processing strategies. A set of seven metrics was proposed to analyze the relationship between input resources and output quality, including total points, average point density, uniformity, yield rate, coverage, geometry accuracy, and time efficiency. The study found that UAV altitude and camera angle had the strongest impact on data density and uniformity, and a 66% overlapping rate was needed for successful 3D reconstruction.
Proceedings Paper
Environmental Sciences
Khishma Modoosoodun Nicolas, Lucas Drumetz, Sebastien Lefevre, Dirk Tiede, Touria Bajjouk, Jean-Christophe Burnel
Summary: Bathymetry studies are important for monitoring coastal topographies, updating navigation charts, and understanding the marine environment dynamics. This study explores the possibility of using deep learning with multispectral satellite data to predict bathymetry around Europa Island. The model shows good accuracy in predicting depth values and has the potential to be incorporated into electronic navigational charts.
EUROPEAN SPATIAL DATA FOR COASTAL AND MARINE REMOTE SENSING, EUCOMARE 2022
(2023)
Proceedings Paper
Computer Science, Artificial Intelligence
Diego Marcos, Jana Kierdorf, Ted Cheeseman, Devis Tuia, Ribana Roscher
Summary: Explainable machine learning and uncertainty quantification have emerged as promising approaches in understanding decision processes. In this paper, a landmark-based approach using heatmapping techniques is proposed to derive sensitivity and uncertainty information for monitoring whales. Experimental results show that this method is more accurate compared to traditional methods.
XXAI - BEYOND EXPLAINABLE AI: International Workshop, Held in Conjunction with ICML 2020, July 18, 2020, Vienna, Austria, Revised and Extended Papers
(2022)
Proceedings Paper
Geosciences, Multidisciplinary
John E. Vargas-Munoz, Diego Schibli, Devis Tuia
Summary: This article proposes a method that uses a Random Forest classifier to sort the detection errors of coconut trees for efficient analysis and correction. Experimental results in the Kingdom of Tonga demonstrate the feasibility of this method.
2022 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2022)
(2022)
Proceedings Paper
Geosciences, Multidisciplinary
Marc Russwurm, Sherrie Wang, Devis Tuia
Summary: Learning to predict accurately from a few data samples is a challenging task in machine learning. While human vision performs better than deep learning approaches on few-shot learning with natural images, we argue that aerial and satellite images are more difficult for the human eye. Our study compares model-agnostic meta-learning algorithms with human participants on few-shot land cover classification using Sentinel-2 imagery. We find that the categorization of land cover from globally distributed regions is challenging for participants, who perform less accurately and with varying success rates compared to the MAML-trained model. This suggests that labeling land cover directly on Sentinel-2 imagery may not be optimal for tackling new land cover classification problems, and using a trained meta-learning model with a few labeled images could lead to more accurate and consistent solutions compared to manual labeling by multiple individuals.
2022 IEEE INTERNATIONAL GEOSCIENCE AND REMOTE SENSING SYMPOSIUM (IGARSS 2022)
(2022)
Proceedings Paper
Geography, Physical
Wei-Hsin Tseng, Hoang-An Le, Alexandre Boulch, Sebastien Lefevre, Dirk Tiede
Summary: This paper investigates the localization of ground-based LiDAR point cloud on remote sensing imagery. A contrastive learning-based method is proposed, which trains on a digital elevation model (DEM) and high-resolution optical imagery. Experimental results show that the method achieves high scores and has the potential for feature learning and localization.
XXIV ISPRS CONGRESS IMAGING TODAY, FORESEEING TOMORROW, COMMISSION II
(2022)
Article
Geosciences, Multidisciplinary
Matthias Themessl, Katharina Enigl, Stefan Reisenhofer, Judith Koberl, Dominik Kortschak, Steffen Reichel, Marc Ostermann, Stefan Kienberger, Dirk Tiede, David N. Bresch, Thomas Roosli, Dagmar Lehner, Chris Schubert, Andreas Pichler, Markus Leitner, Maria Balas
Summary: Loss and damage databases are crucial for informed decision-making in disaster risk management. Despite data-rich countries like Austria, a consistent and curated multi-hazard database is unavailable. This study presents a demonstrator for a national event-based loss and damage database that fulfills United Nations and European Union requirements. By combining existing data and incorporating earth observation and weather data, the demonstrator provides valuable insights and information on hazards, losses, and damages. The findings highlight the feasibility and added value of a curated event-based database compared to single national datasets or existing international databases.
