Article
Geography, Physical
Burak Ekim, Elif Sertel
Summary: Utilizing three different Deep Neural Network Ensemble methods can improve performance in remote sensing image classification tasks and increase accuracy. This approach enhances the generalizability of the models, generates more robust and generalizable outcomes, and promotes the widespread use of the method.
INTERNATIONAL JOURNAL OF DIGITAL EARTH
(2021)
Article
Green & Sustainable Science & Technology
Yongxiang Liu, Hongmei Zhao, Guangying Zhao, Xuelei Zhang, Aijun Xiu
Summary: Open biomass burning (OBB) has significant impacts on air pollution, human health, and climate change. This study used an optimized Fire Radiative Power (FRP) algorithm to estimate the emissions of carbonaceous gases and aerosols (CGA) from OBB in China and assessed their spatiotemporal characteristics. The results show that CGA emissions in China are concentrated in the Northeast, North, Southwest, and East regions, with a trend of initially increasing, then decreasing national emissions.
JOURNAL OF CLEANER PRODUCTION
(2022)
Article
Environmental Sciences
Jan-Niklas Weber, David Kaufholdt, Rieke Minner-Meinen, Elke Bloem, Afsheen Shahid, Heinz Rennenberg, Robert Haensch
Summary: Global climate change has led to a dramatic increase in the frequency and intensity of wildfires worldwide, affecting the entire ecosystem including plants. The study analyzed the effect of SO2 exposure on deciduous trees and found that beech and oak trees have different sulphur detoxification strategies in response to smoke exposure.
ENVIRONMENTAL POLLUTION
(2021)
Review
Environmental Sciences
Rafik Ghali, Moulay A. Akhloufi
Summary: In recent years, there has been a rise in wildland fires worldwide due to various factors. Climate change is expected to be the main driver for the increasing number of these fires in the coming years. The development of remote fire detection systems based on deep learning models and vision transformers shows promising solutions for addressing this issue. However, there is a lack of published studies on the implementation of deep learning models for wildland fire classification, detection, and segmentation tasks. This paper provides a comprehensive review and analysis of these vision methods and their performances, highlighting the superiority of deep learning approaches over traditional machine learning methods and discussing the research gaps and future directions in this field.
Article
Geochemistry & Geophysics
Yongbo Huang, Yuanpei Jin, Liqiang Zhang, Yishu Liu
Summary: Remote sensing object counting (RSOC) has various applications, and this study improves the historical method of global regression by replacing a single regressor with a deep ensemble and breaking down the problem into learning to rank (L2R) and linear transformation (LT). The study offers new theoretical insights into ensemble learning and provides a novel way of building deep regression ensembles. The proposed counting model, eFreeNet, exhibits superior performance and is more annotation-efficient than other methods on six benchmarks.
IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING
(2023)
Article
Ecology
Gensheng Hu, Pan Yao, Mingzhu Wan, Wenxia Bao, Weihui Zeng
Summary: Accurate detection and classification of pine tree diseases are crucial for monitoring tree growth and controlling forest diseases. The proposed method combines DDYOLOv5 and ResNet50 networks to improve detection accuracy using preprocessing and specific modules. Experimental results demonstrate higher precision and recall compared to other methods.
ECOLOGICAL INFORMATICS
(2022)
Article
Environmental Sciences
Yuping Tian, Zechuan Wu, Mingze Li, Bin Wang, Xiaodi Zhang
Summary: This study utilized multi-source satellite remote sensing image data to monitor and analyze wildfires in Sichuan Province. The research found that meteorological factors have the greatest impact on fire spread, and there is a correlation between vegetation coverage and fire severity. By combining multi-source remote sensing data, accurate and quantitative monitoring and analysis of wildfires can be achieved, providing important insights for fire prevention and control.
Review
Environmental Sciences
Xinglu Cheng, Yonghua Sun, Wangkuan Zhang, Yihan Wang, Xuyue Cao, Yanzhao Wang
Summary: The rapid advancement of remote sensing technology has enhanced the temporal resolution of remote sensing data, leading to the emergence of multitemporal remote sensing image classification. Deep learning methods have become prevalent in this field due to their ability to handle massive datasets. This paper provides a review and discussion on the research status and trends in multitemporal images, including retrieval statistics, dataset preparation, model overview, and application status. It also identifies current problems and proposes future prospects, aiming to help readers understand the research process and application status of this field.
Article
Environmental Sciences
Zhuangzhuang Shao, Bo Tan, Tianze Li, Meiyan Guo, Ruili Hu, Yan Guo, Haiyan Wang, Jun Yan
Summary: The impact of gas released from coal fire combustion on the spatial-temporal distribution of CO2 and CH4 and other greenhouse gas emissions was studied in Xinjiang. Using Landsat 8 and GOSAT satellite data, the impact of regional coal fire on CO2 and CH4 emission flux was comprehensively evaluated. The results showed that CO2 and CH4 emissions in Xinjiang were generally dispersed and locally concentrated, with higher intensity in coal fire concentrated areas. The results provide a reference for coal fire control and carbon emission reduction.
ENVIRONMENTAL SCIENCE AND POLLUTION RESEARCH
(2023)
Article
Environmental Sciences
Jessica Castagna, Alfonso Senatore, Guido Pellis, Marina Vitullo, Mariantonia Bencardino, Giuseppe Mendicino
Summary: In the context of climate change impacts, it is crucial to quantitatively estimate wildfire emissions and evaluate their uncertainties for mitigation and adaptation purposes. However, global atmospheric emission models relying on remote sensing fire datasets are affected by significant uncertainties. This study compares satellite-based and ground-based wildfire emissions data for the Calabria region in southern Italy, highlighting the overestimation of burned areas by satellite and the underestimation of dry matter and emissions by forest and grassland wildfires. Additionally, land cover information helps assess uncertainties in ground-based emission inventories.
AIR QUALITY ATMOSPHERE AND HEALTH
(2023)
Article
Geography, Physical
Foroogh Golkar, Seyed Mohsen Mousavi
Summary: This study investigates anthropogenic CO2 emissions in the Middle East using satellite observations. By analyzing the anomaly of CO2 concentration, the major sources of CO2 emissions and growing seasons can be identified. The study explores the relationship between CO2 emissions and human and natural driving factors using open-source data and primary productivity data. The Delta XCO2 maps are capable of detecting CO2 emission fluctuations in defined periods, providing valuable information for controlling CO2 emissions in critical regions.
INTERNATIONAL JOURNAL OF DIGITAL EARTH
(2022)
Article
Environmental Sciences
Rong Song, Tijian Wang, Juncai Han, Beiyao Xu, Danyang Ma, Ming Zhang, Shu Li, Bingliang Zhuang, Mengmeng Li, Min Xie
Summary: Warm and dry climate conditions favor forest fires, leading to the release of pollutants and trace gases that degrade air quality and impact human health. This study developed a forest fire emission inventory for China in 2020 using high-resolution data, and analyzed the variations at the provincial and seasonal level.
ATMOSPHERIC ENVIRONMENT
(2022)
Article
Engineering, Aerospace
M. Shaygan, M. Mokarram
Summary: The purpose of this study is to investigate the air pollution levels in Tehran, Isfahan, Semnan, Mashhad, Golestan, and Shiraz during the COVID-19 era and compare it to the pre-pandemic period. The study analyzes the concentration of various pollutants and examines the relationship between greenhouse gases, air inversion, air temperature, and pollutant indices. The findings show a decrease in pollution caused by pollutants during the COVID-19 era and higher pollution levels in Tehran and Isfahan.
ADVANCES IN SPACE RESEARCH
(2023)
Article
Environmental Sciences
Ri Jin, Kyoo-Seock Lee
Summary: Forest fires cause significant damage to property and the environment worldwide. North Korea suffers from fires annually, and its political isolation makes it difficult to study. Remote sensing techniques and digital topographic data can be used to analyze fire characteristics. The study reveals differences in the distribution and size of fires in North Korea, with coniferous forests being more susceptible to damage.
Article
Environmental Sciences
Yixin Zhao, Yajun Huang, Xupeng Sun, Guanyu Dong, Yuanqing Li, Mingguo Ma
Summary: This study used various satellite images to extract the burned area of forest fires that occurred in Chongqing, China in August 2022. The results of three monitoring methods were consistent, with a coefficient of determination R-2 > 0.96. Fire severity was analyzed using a threshold method based on the extracted dNBR, revealing that moderate-severity fires accounted for the majority (58.05%). Different topographic factors influenced the severity of the forest fires, with high elevation, steep slopes, and northwestern aspect having the largest burned area.
Article
Computer Science, Artificial Intelligence
Blaz Skrlj, Saso Dzeroski, Nada Lavrac, Matej Petkovic
Summary: This paper explores how to adapt the Relief algorithm to benefit from (Riemannian) manifold-based embeddings of instance and target spaces, proposing the faster and better feature ranking ReliefE algorithm. Evaluation on 20 real-life data sets shows the utility of ReliefE for high-dimensional data sets.
Article
Biology
Bijit Roy, Tomaz Stepis, The Pooled Resource Open-Access Als Clinical Trials Consortium, Celine Vens, Saso Dzeroski
Summary: This article treats the prediction of time-to-event as a multi-target regression task and models censored observations as partially labeled examples. By applying semi-supervised learning, a method with better predictive performance and smaller models is proposed. Additionally, the informative feature selection mechanism of the method is illustrated in the context of predicting survival for amyotrophic lateral sclerosis patients.
COMPUTERS IN BIOLOGY AND MEDICINE
(2022)
Article
Ornithology
Urska Ratajc, Martin Breskvar, Saso Dzeroski, Al Vrezec
Summary: A long-term study in central and southern Europe found that fluctuations in small mammal populations in montane temperate forests have significant impacts on owl predators. The Yellow-necked Mouse plays a key role in determining owl populations and their breeding performance.
Article
Computer Science, Artificial Intelligence
Jasmin Bogatinovski, Ljupco Todorovski, Saso Dzeroski, Dragi Kocev
Summary: In this study, a comprehensive meta-learning analysis of data sets and methods for multilabel classification was conducted. The results showed that meta features describing the label space were the most important, and meta features describing label relationships occurred more frequently than those describing label distributions. Furthermore, optimizing hyperparameters can improve predictive performance, although the extent of improvement may not always be justified by resource utilization.
INTERNATIONAL JOURNAL OF INTELLIGENT SYSTEMS
(2022)
Article
Computer Science, Artificial Intelligence
Jure Brence, Jovan Tanevski, Jennifer Adams, Edward Malina, Saso Dzeroski
Summary: This study presents the development of surrogate models for two radiative transfer models (RTMs), which can speed up the simulation process and accurately emulate satellite observations. The surrogate models show good performance in simulating Sentinel 5P spectra and exhibit broad applicability in different parameter sets and applications.
Article
Engineering, Industrial
Yang Jiao, Saso Dzeroski, Ales Jurca
Summary: This study investigates the variation in toe shape, as measured by the hallux valgus angle, and finds that it has a normal distribution in the general population. Females have larger angles compared to males, and people from Asia have larger angles compared to those from North America and Europe.
Article
Environmental Sciences
Marjan Stoimchev, Dragi Kocev, Saso Dzeroski
Summary: Images are now being generated at an unprecedented rate, and remote sensing images have attracted considerable research attention in image classification. Recently, the task of assigning multiple semantic categories to an image, known as multi-label classification, has become increasingly complex. This work explores different strategies for model training using pre-trained convolutional neural network architectures and traditional tree ensemble methods for multi-label classification, and conducts extensive experimental analysis on publicly available remote sensing image datasets.
Article
Chemistry, Analytical
Csaba Voros, David Bauer, Ede Migh, Istvan Grexa, Attila Gergely Vegh, Balazs Szalontai, Gastone Castellani, Tivadar Danka, Saso Dzeroski, Krisztian Koos, Filippo Piccinini, Peter Horvath
Summary: Nowadays, morphology and molecular analyses at the single-cell level have a fundamental role in understanding biology better. These methods are utilized for cell phenotyping and in-depth studies of cellular processes, such as mitosis. Fluorescence microscopy and Raman microscopy techniques provide detailed morphological information and spatial distribution of molecular fingerprints, respectively, enabling the study of cellular stages at the single-cell level. Artificial intelligence can characterize the molecular backgrounds of phenotypes and biological processes by analyzing spectral maps, making it a valuable solution for interpreting complex subcellular spectral maps.
Article
Biology
Nina Omejc, Manca Peskar, Aleksandar Miladinovic, Voyko Kavcic, Saso Dzeroski, Uros Marusic
Summary: The use of non-invasive EEG as an input sensor in brain-computer interfaces is common, but the collected EEG data face challenges, such as age-related variability of ERPs. To assess the effects of aging, a study was conducted with young and older individuals using EEG. Two types of EEG datasets were created for classifier training, and linear classifiers performed best. It was found that classification performance differs between dataset types, and using temporal features resulted in higher and more consistent performance scores. The effect of aging on classification performance depends on the classifier and its feature ranking, highlighting the importance of careful feature extraction and selection to avoid age-related performance degradation in practice.
Article
Computer Science, Information Systems
Jure Brence, Saso Dzeroski, Ljupco Todorovski
Summary: This paper proposes using attribute grammars to ensure the dimensional consistency of the induced equations, which can combine cross-domain knowledge and domain-specific knowledge effectively. The study also demonstrates that attribute grammars can be transformed into probabilistic context-free grammars for equation discovery efficiently. Furthermore, empirical evidence shows that attribute grammars ensuring dimensional consistency of equations can significantly improve the performance of equation discovery on the standard set of a hundred Feynman benchmarks.
INFORMATION SCIENCES
(2023)
Article
Engineering, Aerospace
Bozhidar Stevanoski, Dragi Kocev, Aljaz Osojnik, Ivica Dimitrovski, Saso Dzeroski
Summary: The Mars Express spacecraft, operated by the European Space Agency, has provided unprecedented scientific data about Mars but also needs accurate power modeling due to degradation. This pilot study predicts the thermal power consumption of the spacecraft using telemetry data, employing multi-target regression and considering both local and global approaches.
Article
Multidisciplinary Sciences
Ziga Kokalj, Saso Dzeroski, Ivan Sprajc, Jasmina Stajdohar, Andrej Draksler, Maja Somrak
Summary: This study aims to collect a multimodal annotated dataset suitable for deep learning in remote sensing of Maya archaeology. The dataset covers the area around Chactun, one of the largest ancient Maya urban centres in the central Yucatan Peninsula. It includes raster visualisations and canopy height models from airborne laser scanning, satellite data from Sentinel-1 and Sentinel-2, as well as manual data annotations representing different types of ancient Maya structures.
Article
Microbiology
Monika Novak Babic, Gregor Marolt, Jernej Imperl, Martin Breskvar, Saso Dzeroski, Nina Gunde-Cimerman
Summary: The presence and abundance of fungi in water are influenced by the source of water, water cleaning methods, and the materials in contact with water. Chlorination reduces fungal numbers, but its effect diminishes with longer water networks. Different fungi are observed on different materials, with plastic materials being more susceptible to colonization by basidiomycetous fungi.
Proceedings Paper
Computer Science, Artificial Intelligence
Ana Nikolikj, Saso Dzeroski, Mario Andres Munoz, Carola Doerr, Peter Korosec, Tome Eftimov
Summary: In black-box optimization, understanding the behavior of an algorithm instance is crucial. We propose a methodology that formulates an algorithm instance footprint by identifying easy and difficult problem instances. This methodology uses meta-representations to link algorithm performance and landscape properties, and clustering to detect regions of poor and good algorithm performance.
PROCEEDINGS OF THE 2023 GENETIC AND EVOLUTIONARY COMPUTATION CONFERENCE, GECCO 2023
(2023)
Article
Computer Science, Artificial Intelligence
Viktor Andonovikj, Pavle Boskoski, Saso Dzeroski, Biljana Mileva Boshkoska
Summary: This study addresses the problem of estimating the time-to-employment of a jobseeker using survival analysis and oblique predictive clustering tree. The approach treats censored data as missing data and shows its effectiveness on jobseekers' personal and professional characteristics.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
