Hrčak - Portal of scientific journals of Croatia
Not a member yet
321498 research outputs found
Sort by
CoNb2O6 embedded in graphene nanosheets as an advanced intercalation anode for high-energy lithium-ion capacitors
No full textIntercalation anode materials are promising candidates for hybrid lithium-ion capacitors (LICs) owing to their excellent lithium storage capacity and cycling stability. In this study, a composite of CoNb2O6 embedded in graphene nanosheets (CoNb2O6@G) was synthesized via a two-step hydrothermal method and demonstrated for the first time as an intercalation anode material for lithium storage. The graphene sheets form a three-dimensional porous framework that provides abundant binding sites for the CoNb2O6 particles, effectively mitigating particle agglomeration and volume expansion during charge-discharge cycles. The composite with the optimal graphene content of 100 mg (CoNb2O6@G-100mg) exhibited a remarkable reversible capacity of 508.5 mA h g-1 at a current density of 50 mA g-1. Furthermore, the CoNb2O6@G-100mg//activated carbon (AC) LIC, in which CoNb2O6@G-100mg and AC are used as the anode and cathode, respectively, exhibited an energy density of 94.1W h kg-1 and a maximum power density of 8750 Wkg-1 within the voltage range of 0.0to3.5V. The device demonstrated outstanding cycling stability, with negligible capacity loss (0.00255% per cycle) over 10,000 charge-discharge cycles. These results demonstrate the potential of CoNb2O6@G as a high-performance anode material for energy-storage devices, particularly in power-oriented applications
Usporedba performansi operacija obilaska i agregacije svojstva u bazama podataka
No full textU ovom radu provedena je empirijska usporedba performansi operacija obilaska grafa i agregacije svojstava čvorova na različitim sustavima baza podataka. Operacije obilaska predstavljaju temeljne operacije nad graf strukturama te izravno utječu na primjenjivost graf baza u analitičkim i interaktivnim sustavima. Evaluirani su obilasci prve, druge i treće razine nad sintetičkim grafom zvjezdane strukture koji sadrži 1,1 milijun čvorova i bridova. Analiza se temelji na mjerenju latencije izvršavanja upita, distribucije latencija i vršne potrošnje memorije. Rezultati pokazuju značajne razlike u performansama između analiziranih arhitektura, pri čemu nativne graf baze podataka ostvaruju niže i stabilnije latencije, dok Memgraph postiže najbolje rezultate pri dubljim razinama obilaska grafa
Attitudes, Risks and Regulation: The Social Foundations of AI Adoption in Croatia
No full textThis study investigates how attitudes toward artificial intelligence (AI), levels of technological competence and patterns of trust shape AI adoption, perceived labour-market risks and support for regulatory measures among working-age adults in Croatia. The analysis draws on data from a nationally representative CAWI survey conducted within the project Artificial Intelligence and Social Change. A subsample of respondents aged 18-64 (N = 418) was used for this study. The questionnaire included measures of AI usage, perceptions of labour-market uncertainty, technological and scientific trust, AI self-efficacy and attitudes toward regulation. Composite scales were constructed using reliability analysis and principal component analysis. AI adoption was modelled with binary logistic regression. Results show that younger age, stronger trust in AI and higher AI self-efficacy significantly increase the likelihood of regular AI use. Labour-market risk perceptions were examined using a general linear model, revealing that pro-technology attitudes (reverse-coded transhumanism) and higher trust in science are associated with greater perceived job insecurity related to AI, while demographic variables exert minimal influence. Support for AI regulation was analysed using logistic regression with a binary outcome capturing consistent pro-regulatory preferences. AI optimism, perceived labour-market risks and perceived technological risks all significantly increase support for regulatory measures, whereas demographic factors play only a marginal role. Overall, the findings indicate that AI adoption, labour-market concerns and demand for regulation are driven primarily by attitudinal and perceptual mechanisms rather than socio-demographic characteristics. The study highlights the coexistence of AI optimism and regulatory caution, pointing to a societal demand for governance frameworks that balance technological innovation with social safeguards
Shortening the Test Duration of Ultrasound Penetration-Based Digital Soil Texture Analyzer
No full textIn this study, an approach is presented that compares curve fitting, support vector regression, multilayer perceptron, and long short-term memory architecture to reduce the experiment duration in the formerly proposed Ultrasound Penetration-Based Digital Soil Texture Analyzer (USTA) device, which can automatically, affordably, and effortlessly determine soil texture analysis. The primary objective is to minimize the standard 2-hour experiment time while maintaining an acceptable level of accuracy. To achieve this, signals comprising 14400 samples collected from 52 soil specimens within a 2-hour time-frame using the USTA device were utilized. First, many short variations of the signal were created by either trimming 500 samples at a time from the end of each signal or adding 500 samples from the beginning of each signal. Deviation values were then estimated by comparing these variations to the original signals using different methods. Subsequently, by comparing error values, the best shortened variation was determined. In the curve fitting method, second-degree exponential equations were selected as the best-fit curves using the R-squared method. After extensive fine-tuning and experimentation with various methods, it was found that the best results for reducing experiment duration were achieved using Long Short-Term Memory
Enhanced YOLO Architecture with Attention Mechanism for Accurate Tobacco Plant Counting from UAV Images
No full text(1) Background: This study investigates the construction and optimization of the You Only Look Once (YOLO) deep learning model for high-precision identification of suitable tobacco leaves. (2) Methods: Using tobacco fields in Xiaoxin Street, Niulanjiang Town, Songming County, Kunming as the study area, a total of 1200 UAV images collected during the planting, growth, and harvesting stages were employed as the training dataset to train object detection models such as YOLO v3. After 200 training iterations, the recognition performance of each model was compared and analyzed. (3) Results: YOLO v5 and YOLO v7 were selected as baseline models, and a channel attention mechanism was integrated to develop the improved YOLO v5-EN model. Ablation experiments were conducted by incorporating the attention module, dynamic rectified linear unit (DReLu) activation function, and a feature refinement module. YOLO v7 en was designed as a backbone network, and metrics such as precision, recall, and accuracy were comprehensively evaluated to assess the performance of both the baseline and improved models in identifying the number of tobacco plants. Compared to the baseline, the improved YOLO v5 model demonstrated a 0.36% increase in precision and a 1.55% increase in recall, achieving an overall recognition accuracy of 91.41%. The improved YOLO v7 model achieved a precision of 99.16% and a mean average precision (map) of 95.86%. These results indicate that the enhanced YOLO v5 model with channel attention effectively addresses the issues of missed and false detections in tobacco plant recognition. Furthermore, the improved YOLO v7 model, integrated with collaborative optimization strategies and an enhanced backbone, significantly improves the performance and efficiency of the detection model, particularly in terms of accuracy and processing speed for complex visual tasks. (4) Conclusions: The improved YOLO models significantly enhance the accuracy of tobacco plant count recognition and offer a practical solution for efficient tobacco plant statistics, serving as a reference for intelligent agriculture
Vehicle Classification in Low-Resolution Surveillance Images Using RepViT and KernelWarehouse with Composite Loss
No full textVehicle classification within low-resolution surveillance scenarios remains a challenging task due to the subtle differences between classes and the lack of clear visual cues. This study aimed to improve vehicle classification performance under low-resolution surveillance scenarios.To this end, we proposed KRepIncep-AF, a convolutional neural network model that employed the backbone of InceptionNeXt-Tiny, RepViT modules, and a KernelWarehouse block for prioritized assimilation of spatial cues and contextual information. A compound loss function that combined linear adaptive cross-entropy and focal loss was applied to effectively address class imbalance and reinforce robustness. Comparative experiments were carried out using a vehicle dataset consisting of six classes and a resolution of 100 × 100 pixels. The proposed model attained an outstanding accuracy rate of 99.58%, with macro-average F1, precision, and recall values exceeding 99.5%, and outperformed several competitive baselines. These results demonstrate the effectiveness of the proposed architecture in constrained surveillance environments. Visual examination via heatmaps further established that the model highlighted silhouette-specific features such as bumpers and trailers. These observations indicated that improvements in model structure and the domain-specific application of loss functions could lead to considerable gains in classification accuracy, with meaningful implications for real-world traffic surveillance scenarios
Research on Personalized Recommendation Model of E-commerce Based on Multimodal Big Data Analysis
No full textE-commerce utilizes users' implicit behavior data to replace sparse ratings and strengthens the diversity optimization objective in the recommendation model to simultaneously improve accuracy and reduce the repetition rate. Firstly, a multimodal feature fusion algorithm with PSO adaptive weights is proposed. The emotional features of e-commerce are extracted by using Bi-GRU combined with attention, and the emotional features of images are extracted by using convolutional neural networks combined with attention. The shared semantic layer is studied. When conducting information fusion in the feature layer, the idea of particle swarm optimization is introduced. The multimodal emotional features are weighted and fused, and the feature vectors that have been weighted and fused through particle swarm optimization are taken as the overall emotional vector. The sentiment vector was calculated for similarity through the explicit and implicit sentiment calculation formula. Then, the collaborative filtering model based on features (integrating diversity constraints) was studied. New products were represented as feature quantities, and the score was predicted by calculating the similarity of product features. Experiments show that this method can effectively solve the problems of data sparsity, cold start of new products and high repetition rate of recommendations
Research on the Heat Transfer Characteristics and Energy-Saving Performance Prediction Model of Vacuum Insulation Board Composite Wall
No full textThis study conducts a quantitative analysis of the thermal bridge effect caused by the joints during the construction of vacuum insulation panels, revealing a significant positive correlation between the width of the joints and the increase in the heat transfer coefficient of the wall. Calculations show that when the width of the board joint increases to the upper limit of 20 mm, the deterioration rate of the heat transfer coefficient of the wall can reach 40% (i.e., the insulation performance will decline by approximately 33%). Based on this, a dynamic correction model for thermal conductivity is proposed. This model achieves the engineering quantitative correction of the thermal bridge effect by introducing the panel gap attenuation factor, providing a key parameter basis for the thermal design of the insulation system. At the same time, energy consumption calculations are conducted for passive buildings using composite insulation boards. The impact of vacuum insulation board failure on building energy consumption before and after is analyzed, as well as the methods for ensuring the operation of the building in the later stage, providing a reference for the subsequent engineering design of the built-in insulation system using composite insulation boards
Transfer Entropy-Based Research on Vibration Source Tracing and Key Paths in a Hydropower House
No full textThe vibration sources of power house of hydroelectric power station are complex, and are prone to undesirable vibrations. Among them, the transfer path of hydraulic vibration is an important part of vibration control and operation monitoring. Accurately separating hydraulic vibration sources is crucial to identify the transfer path. To accurately identify the transfer path of the hydraulic vibration source in the structure of the power house of hydroelectric power station, a method combining CEEMDAN-SVD and TE is proposed. The rationality is verified through cross-correlation simulation signals, which are applied to the analysis of the vibration transfer path of the power house of hydroelectric power station. Firstly, based on the vibration test data, CEEMDAN-SVD is used to adaptively perform eigenvector decomposition by the signal energy. Secondly, based on the decomposed vibration signals, the TE (Transfer Entropy) is utilized to identify and analyze the transfer paths of partial vibration source of hydroelectric power station structure. Finally, the applicability of this method to the vibration transfer path of the factory building structure was verified by the quantitative index - information transmission rate. Research shows that the main transfer path of the vibration caused by the vortex rope is as follows: draft tube → undercarriage or stator base → upper floor structure of the power house. The research results are significant for the vibration control and operation monitoring of hydropower station, which provide new ideas for vibration analysis in the engineering field
Robust and High-Precision Harmonic Estimation of Shaking Table Vibrations Using Accelerometer and Cheetah Optimization
No full textThe accuracy of shaking tables in seismic simulations is limited by harmonic distortions in acceleration signals resulting from the inherent nonlinearity of the system. This study presents a novel approach based on the Cheetah Optimization (CO) algorithm for estimating the amplitude and phase components of these harmonics. The algorithm's performance was first validated against other metaheuristic methods in the literature on a standard test signal and was found to be successful. Subsequently, the CO algorithm was applied to data obtained from a real-time shaking table experimental setup. These experiments included testing the system under loaded and unloaded conditions, at three different displacement levels (3.55 mm, 5 mm, 10 mm) and with three different waveforms: sinusoidal, triangular, and square. The results showed that the CO algorithm exhibited very high estimation accuracy for sinusoidal signals in both load cases. However, the algorithm's performance degraded under load on triangular signals and struggled to accurately model the signal in all scenarios due to the sharp transitions and high harmonic content of square waveforms. These findings suggest that while CO has proven to be an effective method for shaking table vibration analysis, particularly for sinusoidal signals, additional model improvements are necessary for more complex waveforms