Article
Construction & Building Technology
Rahmat Ali, Dongho Kang, Gahyun Suh, Young-Jin Cha
Summary: An autonomous UAV system with a modified Faster R-CNN was proposed for identifying structural damage and mapping it in a GPS-denied environment. The method significantly reduced false positives, especially in detecting small cracks in blurry videos caused by UAV vibrations. In real-world tests, the autonomous UAV successfully followed desired trajectories and accurately detected defects using Faster R-CNN.
AUTOMATION IN CONSTRUCTION
(2021)
Article
Computer Science, Information Systems
Tinnakorn Keawboontan, Mason Thammawichai
Summary: In this study, a deep learning multiple object tracking model for UAV aerial videos is proposed to achieve real-time performance. The model combines detection and tracking methods using adjacent frame pairs as inputs with shared features to reduce computational time. A multi-loss function is employed to address the imbalance between challenging classes and samples. The proposed method achieves real-time tracking and outperforms state-of-the-art algorithms on the VisDrone2019 test-dev dataset.
Article
Agriculture, Multidisciplinary
Rui Gao, Penghao Chang, Dong Chang, Xin Tian, Yan Li, Zhiwen Ruan, Zhongbin Su
Summary: This study proposes an edge computing method based on deep learning and photogrammetry for real-time calculation of rice lodging areas using unmanned aerial vehicles (UAVs). The method shows high efficiency and real-time capabilities, providing assistance in rice yield measurement and disaster damage assessment.
COMPUTERS AND ELECTRONICS IN AGRICULTURE
(2023)
Article
Computer Science, Artificial Intelligence
Chuncheng Feng, Hua Zhang, Yonglong Li, Shuang Wang, Haoran Wang
Summary: The paper introduces a method of using deep learning for spillway tunnel defect detection, which shows significant improvement in efficiency and accuracy compared to traditional manual recognition methods. By collecting images of spillway tunnel using a UAV system and training with a lightweight STDD network, accurate recognition of rebar-exposed defects is achieved.
JOURNAL OF REAL-TIME IMAGE PROCESSING
(2021)
Article
Remote Sensing
Linyuan Li, Xihan Mu, Francesco Chianucci, Jianbo Qi, Jingyi Jiang, Jiaxin Zhou, Ling Chen, Huaguo Huang, Guangjian Yan, Shouyang Liu
Summary: Accurate estimation of forest crown cover is crucial for ecological studies. This study proposes a method that combines deep learning with UAV imagery and photogrammetric point clouds for canopy mapping, demonstrating its effectiveness in separating understorey and overstorey vegetation components.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2022)
Article
Computer Science, Artificial Intelligence
Siddhant Panigrahi, Prajwal Maski, Asokan Thondiyath
Summary: Ecological biodiversity is declining rapidly, leading to global efforts in biodiversity conservation. However, the lack of a feasible methodology to quantify biodiversity in real-time hinders the use of ecological data in environmental planning.
PEERJ COMPUTER SCIENCE
(2023)
Article
Computer Science, Information Systems
Luis Augusto Silva, Valderi Reis Quietinho Leithardt, Vivian Felix Lopez Batista, Gabriel Villarrubia Gonzalez, Juan Francisco De Paz Santana
Summary: This paper proposes a new automated road damage detection approach using UAV images and deep learning techniques. The manual collection of road damage data is labor-intensive and unsafe, thus UAVs and AI technologies are introduced to improve efficiency and accuracy. Three algorithms, YOLOv4, YOLOv5, and YOLOv7, are utilized for object detection and localization, achieving high mean average precision values.
Review
Remote Sensing
Zhen Cao, Lammert Kooistra, Wensheng Wang, Leifeng Guo, Joao Valente
Summary: This paper systematically reviews previous studies on UAV real-time object detection, including application scenarios, hardware selection, real-time detection paradigms, detection algorithms and their optimization technologies, and evaluation metrics. The research findings reveal that real-time object detection is more in demand in scenarios such as emergency rescue and precision agriculture. Multi-rotor UAVs and RGB images are of more interest in applications, and real-time detection mainly uses edge computing and deep learning algorithms. Optimization algorithms need to be focused on resource-limited computing platform deployment. In addition to accuracy, speed, latency, and energy are equally important evaluation metrics. Future developments in autonomous UAVs and communications may have a prospective impact on UAV real-time target detection.
Article
Environmental Sciences
Yifei Sun, Zhenbang Hao, Zhanbao Guo, Zhenhu Liu, Jiaxing Huang
Summary: This study explores the impact of sample distribution patterns on the accuracy and generalization performance of deep learning models for chestnut detection and classification. The results show that the combination of DeepLab V3 with ResNet-34 backbone performs the best, while the combination of DeepLab V3+ with ResNet-50 backbone performs the worst. Different spatial distribution patterns of chestnut planting also affect the classification accuracy. Comprehensive training data improves the generalization performance of chestnut classification with different spatial distribution patterns.
Article
Agronomy
Sheng Jiang, Ziyi Liu, Jiajun Hua, Zhenyu Zhang, Shuai Zhao, Fangnan Xie, Jiangbo Ao, Yechen Wei, Jingye Lu, Zhen Li, Shilei Lyu
Summary: This study introduces a real-time detection and maturity classification method for loofah, which includes a one-stage instance segmentation model called LuffaInst and a machine learning-based maturity classification model. Experimental results show that LuffaInst has lower parameter weights and higher accuracy than other prevalent instance segmentation models. A random forest model relying on color and texture features is also developed for three maturity classifications of loofah fruit instances. The research results have important implications for loofah fruit maturity detection.
Article
Chemistry, Analytical
Krzysztof Gromada, Barbara Siemiatkowska, Wojciech Stecz, Krystian Plochocki, Karol Wozniak
Summary: The article presents real-time object detection and classification methods using unmanned aerial vehicles (UAVs) equipped with synthetic aperture radar (SAR). The research introduces a new method that combines YOLOv5 with post-processing using classic image analysis, resulting in improved classification accuracy and object location. The algorithms were implemented and tested on a mobile platform installed on a military-class UAV, reducing the size of the scans sent to the ground control station.
Article
Remote Sensing
Vivien Zahs, Katharina Anders, Julia Kohns, Alexander Stark, Bernhard Hoefle
Summary: Automatic damage assessment using UAV-derived 3D point clouds provides fast information on post-earthquake damage. This study presents a change-based approach that uses a machine learning model trained on virtual laser scanning data to assess multi-class building damage from real-world point clouds. The approach achieves high classification accuracies for three damage grades in different geographic regions. The method has the potential to provide timely information on damage situations when real-world training data is limited.
INTERNATIONAL JOURNAL OF APPLIED EARTH OBSERVATION AND GEOINFORMATION
(2023)
Article
Chemistry, Multidisciplinary
Zhonghua Hong, Yahui Yang, Jun Liu, Shenlu Jiang, Haiyan Pan, Ruyan Zhou, Yun Zhang, Yanling Han, Jing Wang, Shuhu Yang, Changyue Zhong
Summary: This paper introduces a deep-learning-based 3D reconstruction method for building damage assessment after an earthquake. The method reconstructs the 3D model of buildings using multi-view UAV images and utilizes the model for damage assessment. The results of the experiment demonstrate the effectiveness and accuracy of the method.
APPLIED SCIENCES-BASEL
(2022)
Article
Engineering, Aerospace
E. Cetin, C. Barrado, E. Pastor
Summary: This study proposes a low-cost solution for counter-drone using state-of-the-art object detection algorithms and transfer learning to improve existing models for real-time drone detection. Training data is generated from a realistic flight simulator, resulting in a 22% accuracy improvement compared to current models.
AERONAUTICAL JOURNAL
(2021)
Article
Environmental Sciences
Stefan Wolf, Lars Sommer, Arne Schumann
Summary: The article discusses the task of designing an on-board aerial object detector and presents a design process. By optimizing the baseline model and proposing a fast detection head, significant improvement in runtime is achieved while maintaining accuracy. In addition, several aspects to consider during deployment and in the runtime environment are discussed.
Article
Geochemistry & Geophysics
Jyun-Ping Jhan, Norman Kerle, Jiann-Yeou Rau
Summary: The effectiveness of damaged building investigation relies on rapid data collection, using UAV and a backpack panoramic imaging system for recording damage status comprehensively. Integrating these for generating complete 3D point clouds is crucial for 3D measurement of damaged areas. This study evaluates the impact of using panoramic images and multiview aerial images for 3D mapping, highlighting the necessity of integrating both for rapid and complete point cloud generation.
IEEE GEOSCIENCE AND REMOTE SENSING LETTERS
(2022)
Article
Geography, Physical
Ning Zhang, Francesco Nex, Norman Kerle, George Vosselman
Summary: This paper presents a novel cascade network for studying semantic segmentation in low-light indoor environments, utilizing real and rendered images datasets. The proposed method achieves high accuracy in segmentation by decomposing low-light images and incorporating illumination invariant features. The results also demonstrate the importance of semantic information in enhancing reflectance restoration and segmentation accuracy.
ISPRS JOURNAL OF PHOTOGRAMMETRY AND REMOTE SENSING
(2022)
Article
Computer Science, Artificial Intelligence
Simone Borsci, Ville V. Lehtola, Francesco Nex, Michael Ying Yang, Ellen-Wien Augustijn, Leila Bagheriye, Christoph Brune, Ourania Kounadi, Jamy Li, Joao Moreira, Joanne Van der Nagel, Bernard Veldkamp, Duc Le, Mingshu Wang, Fons Wijnhoven, Jelmer M. Wolterink, Raul Zurita-Milla
Summary: The article reviews the EU Commission's whitepaper on Artificial Intelligence and highlights potential conflicts with current societal, technical, and methodological constraints. The lack of a coherent EU vision and methods to support sustainable AI diffusion are identified as main obstacles. The article recommends complementary rules and compensatory mechanisms to avoid market fragmentation, as well as research to address technical and methodological open questions for the sustainable development of human-AI co-action.
Article
Environmental Sciences
Ning Zhang, Francesco Nex, George Vosselman, Norman Kerle
Summary: This research focuses on using deep learning to detect victims in disaster debris, proposes a method to generate harmonious composite images for training, and significantly improves detection accuracy.
Article
Environmental Sciences
Sofia Tilon, Francesco Nex, George Vosselman, Irene Sevilla de la Llave, Norman Kerle
Summary: This paper introduces a versatile UAV system that can carry out multiple road infrastructure monitoring tasks simultaneously in real-time. The system design considers computational strain and latency, and it is deployed on a unique typology of UAV. It includes important modules such as vehicle detection, scene segmentation, and 3D scene reconstruction, and has a good performance.
Article
Remote Sensing
Bashar Alsadik, Fabio Remondino, Francesco Nex
Summary: This paper investigates a multi-view camera system integrated with a multi-beam LiDAR to build an efficient UAV hybrid system. Two types of cameras, MAPIR Survey and SenseFly SODA, integrated with a multi-beam digital Ouster OS1-32 LiDAR sensor, are proposed and examined. The results show that with appropriate conditions, high-density facade coverage can be achieved.
Article
Remote Sensing
Samer Karam, Francesco Nex, Bhanu Teja Chidura, Norman Kerle
Summary: This article presents a low-cost SLAM-based drone for creating exploration maps of building interiors. The experimental results indicate that the system is capable of creating quality exploration maps of small indoor spaces and handling the loop-closure problem.
Article
Environmental Studies
Diogo Duarte, Cidalia Fonte, Hugo Costa, Mario Caetano
Summary: This study compares the differences between a global land cover map and a national use/land cover map of Portugal. The results show differences in land use, classification, and focus between the two maps. It is found that the global land cover map is better at distinguishing artificial surfaces and grasslands within urban environments, but often fails to differentiate sparse/individual trees from neighboring cover in the Portuguese landscape.
Proceedings Paper
Geography, Physical
S. Karam, F. Nex, O. Karlsson, J. Rydell, E. Bilock, M. Tulldahl, M. Holmberg, N. Kerle
Summary: This paper presents the operations and mapping techniques of two drones used for mapping indoor spaces. Both the Crazyflie and MAX drones are capable of mapping cluttered indoor environments and providing sufficient point clouds for quick exploration. The results show that the LIDAR scanner-based system can handle larger office environments with minimal drift, while the Crazyflie drone performs well given its limited sensor configuration.
XXIV ISPRS CONGRESS: IMAGING TODAY, FORESEEING TOMORROW, COMMISSION I
(2022)
Proceedings Paper
Geography, Physical
Karen K. Mwangangil, P. O. Mc'okeyo, S. J. Oude Elberink, F. Nex
Summary: This research explores the potential of using UAV image data for 3D buildings reconstruction, and analyzes the optimal parameter settings. The results show that proper segmentation and detection methods can improve the 3D building modeling from UAV point clouds.
XXIV ISPRS CONGRESS IMAGING TODAY, FORESEEING TOMORROW, COMMISSION II
(2022)
Proceedings Paper
Geography, Physical
N. Zhang, F. Nex, G. Vosselman, N. Kerle
Summary: This paper addresses the issue of deep detection networks in detecting buried victims. By generating realistic images and using an unsupervised generative adversarial network for harmonization, the accuracy of victim detection can be effectively improved.
XXIV ISPRS CONGRESS: IMAGING TODAY, FORESEEING TOMORROW, COMMISSION III
(2022)
Proceedings Paper
Geography, Physical
D. Duarte, C. C. Fonte, J. Patriarca, I Jesus
Summary: This study evaluates the geographical transferability of satellite image-based segmentation models trained with OpenStreetMap (OSM) derived data through a series of experiments. The results show that models trained in different locations can improve the mapping of certain classes.
XXIV ISPRS CONGRESS: IMAGING TODAY, FORESEEING TOMORROW, COMMISSION III
(2022)