1. Eexplaining 3D semantic segmentation through generative AI-based counterfactualsDzemail Rozajac, Niko Lukač, Stefan Schweng, Christoph Gollob, Arne Nothdurft, Karl Stampfer, Javier Del Ser, Andreas Holzinger, 2025, izvirni znanstveni članek Opis: Interpreting the predictions of deep learning models on 3D point cloud data is an important challenge for safety-critical domains such as autonomous driving, robotics and geospatial analysis. Existing counterfactual explainability methods often struggle with the sparsity and unordered nature of 3D point clouds. To address this, we introduce a generative framework for counterfactual explanations in 3D semantic segmentation models. Our approach leverages autoencoder-based latent representations, combined with UMAP embeddings and Delaunay triangulation, to construct a graph that enables geodesic path search between semantic classes. Candidate counterfactuals are generated by interpolating latent vectors along these paths and decoding into plausible point clouds, while semantic plausibility is guided by the predictions of a 3D semantic segmentation model. We evaluate the framework on ShapeNet objects, demonstrating that semantically related classes yield realistic counterfactuals with minimal geometric change, whereas unrelated classes expose sharp decision boundaries and reduced plausibility. Quantitative results confirm that the method balances defined interpretability metrics, producing counterfactuals that are both interpretable and geometrically consistent. Overall, our work demonstrates that generative counterfactuals in latent space provide a promising alternative to input-level perturbations.
Ključne besede: 3D point cloud, explainable artificial intelligence, counterfactual analysis, generative AI
Objavljeno v DKUM: 14.11.2025; Ogledov: 0; Prenosov: 6
 Celotno besedilo (27,14 MB) |
2. Dynamic modeling and experimental validation of the photovoltaic/thermal systemKlemen Sredenšek, Eva Simonič, Klemen Deželak, Marko Bizjak, Niko Lukač, Sebastijan Seme, 2025, izvirni znanstveni članek Opis: The aim of this paper is to present a novel and comprehensive methodology for the dynamic modeling and experimental validation of a photovoltaic/thermal system. The dynamic model is divided into thermal and electrical subsystems, encompassing the photovoltaic/ thermal module and the thermal energy storage. The thermal subsystem of both the photovoltaic/thermal module and the thermal energy storage is described by a one-dimensional dynamic model of heat transfer mechanisms and optical losses, while the electrical subsystem is presented as an electrical equivalent circuit of double diode solar cell. Model validation was conducted on a modern experimental photovoltaic/thermal system over an extended operational period at a five-minute resolution, with validation days classified as sunny, cloudy, or overcast based on weather conditions, thereby demonstrating an applied approach. The results demonstrate the lowest deviation values reported to date, confirmed using six quantitative indicators. The added value of the proposed methodology, not previously addressed in the literature, lies in the following contributions: (i) comprehensive modeling of the entire photovoltaic/thermal system, (ii) accurate consideration of optical losses in the photovoltaic/thermal module, and (iii) long-term experimental validation. Overall, the proposed methodology provides a reliable and efficient framework for PV/T system design, optimization, and long-term performance assessment.
Ključne besede: photovoltaic/thermal system, thermal energy storage, dynamic modeling, experimental validation, heat transfer mechanism, temperature, electrical power
Objavljeno v DKUM: 10.11.2025; Ogledov: 0; Prenosov: 5
Celotno besedilo (7,15 MB)
Gradivo ima več datotek! Več... |
3. What can artificial intelligence do for soil health in agriculture?Stefan Schweng, Luca Bernardini, Katharina Keiblinger, Peter Kaul, Iztok Fister, Niko Lukač, Javier Del Ser, Andreas Holzinger, 2025, pregledni znanstveni članek Opis: The integration of artificial intelligence (AI) into soil research presents significant opportunities to advance the understanding, management, and conservation of soil ecosystems. This paper reviews the diverse applications of AI in soil health assessment, predictive modeling of soil properties, and the development of pedotransfer functions within the context of agriculture, emphasizing AI’s advantages over traditional analytical methods. We identify soil organic matter decline, compaction, and biodiversity loss as the most frequently addressed forms of soil degradation. Strong trends include the creation of digital soil maps, particularly for soil organic carbon and chemical properties using remote sensing or easily measurable proxies, as well as the development of decision support systems for crop rotation planning and IoT-based monitoring of soil health and crop performance. While random forest models dominate, support vector machines and neural networks are also widely applied for soil parameter modeling. Our analysis of datasets reveals clear regional biases, with tropical, arid, mild continental, and polar tundra climates remaining underrepresented despite their agricultural relevance. We also highlight gaps in predictor–response combinations for soil property modeling, pointing to promising research avenues such as estimating heavy metal content from soil mineral nitrogen content, microbial biomass, or earthworm abundance. Finally, we provide practical guidelines on data preparation, feature extraction, and model selection. Overall, this study synthesizes recent advances, identifies methodological limitations, and outlines a roadmap for future research, underscoring AI’s transformative potential in soil science.
Ključne besede: artificial intelligence, machine learning, agriculture, soil health, soil parameter modeling, regional data bias
Objavljeno v DKUM: 17.10.2025; Ogledov: 0; Prenosov: 4
Celotno besedilo (4,22 MB) |
4. Razvoj, avtomatizacija in optimizacija spletne trgovine na platformi WooCommerceLuka Gril, 2025, magistrsko delo Opis: Cilj magistrske naloge je predstaviti trende spletne prodaje v zadnjem obdobju ter prikazati razvoj osnovne spletne trgovine z uporabo platform WordPress in WooCommerce. Predstavili bomo osnovne funkcionalnosti, ki jih nudita tehnologiji, ter opisali možnosti nadgradnje s prilagoditvami in razširitvami, ki prispevajo k boljši prodaji in enostavnejšemu upravljanju trgovine. Poleg tega bomo prikazali rešitve za avtomatizacijo določenih procesov z namenom večje časovne učinkovitosti ter optimizacijo hitrosti delovanja spletne trgovine, ki pomembno vpliva na uporabniško izkušnjo.
Ključne besede: WooCommerce, spletna trgovina, optimizacija, avtomatizacija
Objavljeno v DKUM: 04.09.2025; Ogledov: 0; Prenosov: 8
Celotno besedilo (6,29 MB) |
5. Spletni priporočilni sistem za potrošniške medijske vsebine : diplomsko deloFilip Duvnjak, 2025, diplomsko delo Opis: V teoretičnem delu so predstavljeni priporočilni sistemi, primerjava pristopa KNN in Slope One, prikaz slednjih v praktični implementaciji spletnega priporočilnega sistema za potrošniške medijske vsebine ter implementacija spletne strani. Ob tem so podrobno podani tudi razlogi za izbrani način dela. V praktičnem delu so predstavljeni končni rezultati dela in vpliv, ki ga imajo na uporabnika. Tako je razviden celotni vpliv na proces razvijanja spletnega produkta, ki uporablja spletne priporočilne sisteme.
Ključne besede: KNN, Slope One, algoritem, priporočilni sistem, spletna stran
Objavljeno v DKUM: 13.08.2025; Ogledov: 0; Prenosov: 22
Celotno besedilo (2,49 MB) |
6. Contextualized spatio-temporal graph-based method for forecasting sparse geospatial sensor networksNiko Uremović, Domen Mongus, Aleksander Pur, Niko Lukač, 2025, izvirni znanstveni članek Opis: Spatio-temporal forecasting is a rapidly evolving field, accelerated by the increasing accessibility of sensoring infrastructure and computational hardware, capable of processing the large amount of sampled data. Applications of spatio-temporal forecasts range from traffic, weather, air pollution forecasting and others. Emerging technologies employ deep learning architectures, such as graph, convolutional, recurrent and transformer neural networks. While the state-of-the-art methods provide accurate time series predictions, they are typically limited to providing forecasts only for the direct locations of sampling, whereas coverage of the entire area is often desired by the applications. In this work, we propose a method that addresses this challenge and improves on the shortcomings of related works, which have already tackled the task. The proposed graph convolutional recurrent neural network based method provides forecasts for arbitrary geolocations without available measurement data, formulating predictions based on contextual information of target geolocations and the time series data of nearby measurement geolocations. We evaluate the method on three real-world datasets from meteorological, traffic and air pollution domains, and gauge its performance against the state-of-the-art spatio-temporal forecasting methods. The proposed method achieves 12.26 %, 66.97 % and 42.89 % improvements in the mean absolute percentage errors on the three aforementioned datasets, compared to the best performing state-of-the-art method GConvGRU.
Ključne besede: spatio-temporal forecasting, graph recurrent neural networks, sparse geospatial sensor networks
Objavljeno v DKUM: 25.07.2025; Ogledov: 0; Prenosov: 2
Celotno besedilo (5,19 MB) |
7. Detection and optimization of photovoltaic arrays’ tilt angles using remote sensing dataNiko Lukač, Sebastijan Seme, Klemen Sredenšek, Gorazd Štumberger, Domen Mongus, Borut Žalik, Marko Bizjak, 2025, izvirni znanstveni članek Opis: Maximizing the energy output of photovoltaic (PV) systems is becoming increasingly important. Consequently, numerous approaches have been developed over the past few years that utilize remote sensing data to predict or map solar potential. However, they primarily address hypothetical scenarios, and few focus on improving existing installations. This paper presents a novel method for optimizing the tilt angles of existing PV arrays by integrating Very High Resolution (VHR) satellite imagery and airborne Light Detection and Ranging (LiDAR) data. At first, semantic segmentation of VHR imagery using a deep learning model is performed in order to detect PV modules. The segmentation is refined using a Fine Optimization Module (FOM). LiDAR data are used to construct a 2.5D grid to estimate the modules’ tilt (inclination) and aspect (orientation) angles. The modules are grouped into arrays, and tilt angles are optimized using a Simulated Annealing (SA) algorithm, which maximizes simulated solar irradiance while accounting for shadowing, direct, and anisotropic diffuse irradiances. The method was validated using PV systems in Maribor, Slovenia, achieving a 0.952 F1-score for module detection (using FT-UnetFormer with SwinTransformer backbone) and an estimated electricity production error of below 6.7%. Optimization results showed potential energy gains of up to 4.9%.
Ključne besede: solar energy, photovoltaics, semantic segmentation, optimization, LiDAR, VHR imagery
Objavljeno v DKUM: 22.07.2025; Ogledov: 0; Prenosov: 9
Celotno besedilo (11,60 MB)
Gradivo ima več datotek! Več... |
8. LLM in the loop: a framework for contextualizing counterfactual segment perturbations in point cloudsVeljka Kočić, Niko Lukač, Dzemail Rozajac, Stefan Schweng, Christoph Gollob, Arne Nothdurft, Karl Stampfer, Javier Del Ser, Andreas Holzinger, 2025, izvirni znanstveni članek Opis: Point Cloud Data analysis has seen a major leap forward with the introduction of PointNet algorithms, revolutionizing how we process 3D environments. Yet, despite these advancements, key challenges remain, particularly in optimizing segment perturbations to influence model outcomes in a controlled and meaningful way. Traditional methods struggle to generate realistic and contextually appropriate perturbations, limiting their effectiveness in critical applications like autonomous systems and urban planning. This paper takes a bold step by integrating Large Language Models into the counterfactual reasoning process, unlocking a new level of automation and intelligence in segment perturbation. Our approach begins with semantic segmentation, after which LLMs intelligently select optimal replacement segments based on features such as class label, color, area, and height. By leveraging the reasoning capabilities of LLMs, we generate perturbations that are not only computationally efficient but also semantically meaningful. The proposed framework undergoes rigorous evaluation, combining human inspection of LLM-generated suggestions with quantitative analysis of semantic classification model performance across different LLM variants. By bridging the gap between geometric transformations and high-level semantic reasoning, this research redefines how we approach perturbation generation in Point Cloud Data analysis. The results pave the way for more interpretable, adaptable, and intelligent AI-driven solutions, bringing us closer to realworld applications where explainability and robustness are paramount.
Ključne besede: explainable AI, point cloud data, counterfactual reasoning, LiDAR, 3D point cloud data, interpretability, human-centered AI, large language models, K-nearest neighbors
Objavljeno v DKUM: 19.05.2025; Ogledov: 0; Prenosov: 3
Celotno besedilo (7,24 MB) |
9. Optimization-based downscaling of satellite-derived isotropic broadband albedo to high resolutionNiko Lukač, Domen Mongus, Marko Bizjak, 2025, izvirni znanstveni članek Opis: In this paper, a novel method for estimating high-resolution isotropic broadband albedo is proposed, by downscaling satellite-derived albedo using an optimization approach. At first, broadband albedo is calculated from the lower-resolution multispectral satellite image using standard narrow-to-broadband (NTB) conversion, where the surfaces are considered Lambertian with isotropic reflectance. The high-resolution true orthophoto for the same location is segmented with the deep learning-based Segment Anything Model (SAM), and the resulting segments are refined with a classified digital surface model (cDSM) to exclude small transient objects. Afterwards, the remaining segments are grouped using K-means clustering, by considering orthophoto-visible (VIS) and near-infrared (NIR) bands. These segments present surfaces with similar materials and underlying reflectance properties. Next, the Differential Evolution (DE) optimization algorithm is applied to approximate albedo values to these segments so that their spatial aggregate matches the coarse-resolution satellite albedo, by proposing two novel objective functions. Extensive experiments considering different DE parameters over an 0.75 km2 large urban area in Maribor, Slovenia, have been carried out, where Sentinel-2 Level-2A NTB-derived albedo was downscaled to 1 m spatial resolution. Looking at the performed spatiospectral analysis, the proposed method achieved absolute differences of 0.09 per VIS band and below 0.18 per NIR band, in comparison to lower-resolution NTB-derived albedo. Moreover, the proposed method achieved a root mean square error (RMSE) of 0.0179 and a mean absolute percentage error (MAPE) of 4.0299% against ground truth broadband albedo annotations of characteristic materials in the given urban area. The proposed method outperformed the Enhanced Super-Resolution Generative Adversarial Networks (ESRGANs), which achieved an RMSE of 0.0285 and an MAPE of 9.2778%, and the Blind Super-Resolution Generative Adversarial Network (BSRGAN), which achieved an RMSE of 0.0341 and an MAPE of 12.3104%.
Ključne besede: isotropic broadband albedo, high-resolution albedo, Sentinel-2 albedo, true orthophoto, anything model, differential evolution
Objavljeno v DKUM: 23.04.2025; Ogledov: 0; Prenosov: 2
Celotno besedilo (20,47 MB) |
10. A hierarchical universal algorithm for geometric objects’ reflection symmetry detectionBorut Žalik, Damjan Strnad, Štefan Kohek, Ivana Kolingerová, Andrej Nerat, Niko Lukač, David Podgorelec, 2022, izvirni znanstveni članek Opis: A new algorithm is presented for detecting the global reflection symmetry of geometric
objects. The algorithm works for 2D and 3D objects which may be open or closed and may or may
not contain holes. The algorithm accepts a point cloud obtained by sampling the object’s surface at
the input. The points are inserted into a uniform grid and so-called boundary cells are identified.
The centroid of the boundary cells is determined, and a testing symmetry axis/plane is set through
it. In this way, the boundary cells are split into two parts and they are faced with the symmetry
estimation function. If the function estimates the symmetric case, the boundary cells are further split
until a given threshold is reached or a non-symmetric result is obtained. The new testing axis/plane
is then derived and tested by rotation around the centroid. This paper introduces three techniques to
accelerate the computation. Competitive results were obtained when the algorithm was compared
against the state of the art.
Ključne besede: computer science, computational geometry, uniform subdivision, centroids
Objavljeno v DKUM: 01.04.2025; Ogledov: 0; Prenosov: 8
Celotno besedilo (2,99 MB)
Gradivo ima več datotek! Več... |
