KTU Open Journal Systems (Kaunas University of technology)
Not a member yet
13395 research outputs found
Sort by
Yolov5-based Intelligent Detection Method for Retail Goods
No full textIn the current context, intelligent unmanned retail checkout systems offer the prospect of efficient and innovative development. This study proposes an enhanced lightweight YOLOv5 merchandise detection and recognition method. The method introduces SELayer and a multi-headed self-attentive module of Transformer in YOLOv5 to enable the network to focus more on essential factors such as commodities when performing retail merchandise detection, and improve the recognition performance of the model. Also, the Ghost module is introduced to reduce network parameters and computation, increase computation speed and reduce latency. We validated the performance of the approach on a public dataset. Compared with the existing YOLOv5 model, the model achieves a 0.9% improvement in detection accuracy and a 27.7% reduction in GFLOPs. With this study, we optimise the problem of small batch identification of retail goods, providing a basis for automated processing of intelligent retail supply and marketing systems with practical implications.
Enhancing Open-Set Few-Shot Object Detection with Limited Visual Prompts
No full textThe text-prompt-based open-vocabulary object detection model effectively encapsulates the abstract concepts of common objects, thereby overcoming the limitations of pre-trained models, which are restricted to detecting a fixed, predefined set of categories. However, due to data scarcity and the constraints of textual descriptions, representing rare or complex objects solely through text remains challenging. In this study, we propose an open-set detection model that supports both visual and textual prompt queries (VTP-OD) to enhance few-shot object detection. A small number of visual prompts not only provide rich class-wise visual features, which enhance class textual representations, but also enable flexible extension to new classes for different downstream tasks. Specifically, we incorporate two adaptation modules based on cross-attention to adapt the pre-trained vision-language model, allowing it to support both text and visual queries. These modules facilitate (i) visual fusion between a limited number of visual prompts and query images and (ii) visual-language fusion between class-aware visual features and textual representations of the classes. Subsequently, the model undergoes prompt tuning using the available few-shot downstream data to adapt to target detection tasks. Experimental results demonstrate that our model outperforms the pre-trained model on the LVIS and COCO benchmarks. Furthermore, we validate its effectiveness on the real-world CoalMine dataset.
Dense-Attention CNN with Spatial-Attention Fusion for Robust Facial Expression Recognition
No full textCurrently, facial expression recognition technology has been gradually applied in fields such as intelligent healthcare, online education, and assisted driving. However, traditional Convolutional Neural Network (CNN) lack attention to facial local regions related to emotions, and classic loss functions cannot handle intra-class variability in facial expressions. This paper establishes a facial expression recognition model combining deep learning and attention mechanisms for both static and dynamic facial expressions. By extracting image features, it obtains rich multi-scale information flow and controls the number of model parameters. It constructs a spatial attention unit to focus on information with significant emotional intensity, and combines an intra-class distance penalty term and classification loss to supervise the network learning process. This approach addresses the issue of CNN paying insufficient attention to regions of interest while reducing the variability among facial expressions of the same class. Experimental results show that the accuracy of this model has increased by 1.1% and 2.7% on the CK+ and FER2013 public datasets, respectively
Influence Mechanisms and Spatial Spillover Effects of Technological Innovation on New Urbanization
No full textSustainable and high-level new urbanization cannot be achieved without the drive of technological innovation. By collecting data from 282 prefecture-level cities in China from 2007 to 2020, this paper used a two-way fixed effects model and spatial Durbin model to analyze the influence mechanisms and spatial spillover effects of technological innovation on new urbanization respectively. The research conclusions are as follows: (1) Technological innovation affects new urbanization positively and heterogeneously due to geographical location, city ranking, and city size. The threshold test reflects that the impact of technological innovation on new urbanization has the characteristic of weakening along with crossing double thresholds. (2) Technological and financial constraints are two critical variables that positively moderate how technological innovation affects new urbanization. (3) The spatial effect of technological innovation on new urbanization is significant. However, the impact intensity is inversely proportional to the geographical distance between cities, with an estimated spatial attenuation boundary of approximately 350 kilometers. Therefore, this paper proposes adhering to innovation-driven development, synchronizing the technological market with new urbanization, formulating differentiated policies in different regions, and using locational advantages well
Deep Learning Model for Estimation Market Risk in Insurance Sector
No full textThe paper developed a model for market risk assessment based on deep learning in combination with non-parametric, parametric and semi-parametric VaR and ES models. We presented the ANN-GRU model more precisely. It is intended for insurance companies operating in emerging markets because the model was developed to cover all the characteristics of emerging markets. The research was conducted on the example of 8 optimal investment portfolios for insurance companies operating in the Balkan countries. The portfolios were calibrated at the daily level and calculated for the period from 1 January 2020 to 31 December 2023. The first 500 data were used to estimate the calibration of the VaR model, the other 250 to estimate the validity of the VaR model, and the last 250 to test the validity of the VaR/ES-GRU-DL market risk estimates in accordance with Directive II. Conditional and unconditional coverage tests were used to test the validity of VaR estimates, while Berkowitz\u27s ES test was used to test the validity of ES estimates. Due to the limitations of these tests, the validity of the backtesting VaR estimate was performed using Dufour Monte Carlo simulations, while the validity of the backtesting ES estimate used the Bootstrap procedure. The backtesting results, as well as the results of the validity of the backtesting results, show that the model generates reliable estimates of VaR and ES in accordance with the Solvency II directive as well as produces better estimates compared to the popular and widely used VaR and ES models
Spillover Effects of Cryptocurrency Volatility on Green Finance
No full textThis study investigates the risk spillover between clean and dirty cryptocurrencies and their impact on green finance indexes (solar, wind, and nuclear energy) and regional economic indexes (Baltic Dry Index and CRB Index), with data processed using the diagonal BEKK model. The results identify several dirty cryptocurrencies such as: Ethereum Cash (ETC), Litecoin (LTC), and Bitcoin (BIT) as potential diversifiers and hedges with specific green energy and economic indexes. Our findings show that news from the cryptocurrency markets predominantly have a positive, significant effect on the covariance with green finance indices. The study also presents the covolatility spillover effect, showcasing the impact of a return shock in one market, such as the cryptocurrency market or the green finance market, on the co-volatility between markets, including regional economic indices like the Baltic Dry Index and CRB Index. The analysis reveals differential spillover patterns between clean and dirty cryptocurrencies and various green finance indices, highlighting the complexity of their interactions and the varying degrees of influence on regional economic indicators
Rheology, Morphology and Mechanical Properties for Mixtures of Multicomponent Waste from Polymer Composites as Secondary Raw Materials
No full textThe article presents a study of the rheological properties and morphology of a mixture of multicomponent waste polymer compositions of polyamide 6 (PA6GF30) and polycarbonate (PC) from automotive parts. A comparative analysis of the melt flow rate of the obtained mixture was carried out depending on the content of its components. The compatibility of the components in the mixture and their distribution in the obtained polymer composition were studied. The effect of the composition of the secondary multicomponent mixture on the physical and mechanical properties is demonstrated. The possibility of reusing multicomponent polymer waste with the predicted main parameters of the technological process of injection molding is shown
The Effect of Ce and Nd Addition on Dry/Corrosive Wear Behavior of Newly Developed EZ43 Grade Mg Alloys
No full textThis study is concerned with the production of EZ43A and EZ43B alloys by induction melting/casting method and their microstructure, mechanical, and dry/corrosive wear properties. The investigated EZ43A and EZ43B alloys were alloyed with 1 % Ce – 2 % Nd and 2 % Ce – 1 % Nd, respectively. According to the XRD results, ternary Mg0.97Zn0.03/Mg41Nd5/Mg3Gd eutectic phases were present in the microstructure of EZ43A alloy, while quaternary Mg0.97Zn0.03/Mg41Nd5/Mg3Gd/Mg17Ce2 eutectic phase was formed in the microstructure of EZ43B alloy. The hardness increased from 60.41 to 63.34 by increasing Ce from 1 % to 2 %, representing a 5 % improvement in hardness. As the Ce/Nd ratio increased, the yield/tensile strength and modulus of elasticity increased by 2 %, 6 %, and 3 %, respectively. With the increase in Ce, the relative dry wear loss of EZ43A was 1.00, while that of EZ43B was 0.93, and the corrosive relative wear loss of EZ43A was 2.44 and that of EZ43B was 1.56. Accordingly, the dry and corrosive wear resistance of EZ43B increased by 7 % and 36 %, respectively, compared to that of EZ43A
On the Influence of Microstructure and Properties of Direct Laser Deposited 50Cr6Ni2Y Coatings with Different TiC Contents
No full textThree 50Cr6Ni2Y + x wt.% TiC (x = 1.5, 3.0, 4.5) alloy steel coatings were prepared using direct laser deposition (DLD) technology. The microstructure, microhardness, and wear resistance of DLD samples were studied. The results indicate that DLD coatings were composed of α-Fe (Fe-Cr-Ti), γ-Fe (Fe-Ni), and TiC. When the added TiC content was 3.0 wt.%, the DLD coating without cracks was fabricated, and TiC particles were well embedded in the sample. In addition, the coating demonstrated the best performance, with a microhardness of 758 ± 23.3 HV0.2, an average friction coefficient of 0.58, and a wear rate of 0.37 %. The addition of an appropriate amount of TiC as a reinforcing phase, on the one hand, had played a role in the second phase strengthening. On the other hand, the diffusion interfaces formed between TiC particles and the matrix allowed some Ti elements to melt into the matrix and formed a solid solution, playing a role in solid solution strengthening. The results could provide a reference for the preparation and repair of laser additive manufacturing high-performance wear-resistant parts
Energy Absorption and Damage Analysis of Glass Fibre Reinforced Polymer Spherical Core Sandwich Structures
No full textThe present study describes the energy absorption and damage analysis of the spherical core sandwich structures (SCSS) fabricated using woven Glass Fibre Reinforced Plastic (GFRP) by hand- layup method. Based on the core orientation, the spherical cores are categorized as stagger (S), regular (R), inverted (I), and interlock (L). The pitch distance and diameter of the models considered for the study are 24 mm and 16 mm, respectively. The specimens are subjected to a low velocity impact test (LVIT) at three different energy levels 9.9, 27.5, and 53.9 J respectively. Evaluations are carried out on the different kind of parameters namely coefficient of restitution (COR),energy absorption ratio, and energy loss percentage maximum displacement, maximum force, absorbed energy, and rebound energy. Among the models at every impact velocity it is found that the model R sustains a maximum force of 3078 N at 7 m/s impact velocity. The stagger model has recorded a maximum displacement of 34.4 mm among all velocities, whereas the regular model reveals a minimum displacement of 4.9 mm based on the analysis of maximum displacement. Similarly, the regular model has a maximum energy absorption ratio at 5 and 7 m/s respectively, whereas at 3 m/s the interlock model absorbs more energy. The failure pattern of the specimens is analyzed through visual inspection and ultrasound testing. Matrix cracking and fibre breakage are the typical failures seen in the model at 3 m/s, while core crushing and perforation are seen at 5 and 7 m/s impact velocities. The damage area is minimum for the interlock model whereas it is maximum for the stagger model.