Browse Articles
Discover research articles across all indexed journals
Numerical computation of the stochastic hepatitis B model using feed forward neural network and real data
ELDGG: an end-to-end LiDAR-dynamic-guided GAN for hyperspectral image hierarchical reconstruction and classification
Abstract To address the prevalent issues in the classification of hyperspectral image (HSI) and light detection and ranging (LiDAR) data fusion, such as insufficient dynamic adaptive interaction of cross-modal features, and difficulties in high-fidelity spatial detail reconstruction, this paper proposes an end-to-end LiDAR-dynamic-guided GAN for hyperspectral image hierarchical reconstruction and classification (ELDGG). The core framework of the network consists of a guided hierarchical reconstruction generator (GHR-Generator) and a perception-enhanced spectral regularization discriminator (PSR-Discriminator). First, we propose the cross-modal parameter-adaptive fusion module (CPAF-Module), which leverages the global context of LiDAR data to generate dynamic convolutional operators tailored for HSI features, addressing the limitations of static fusion methods. Second, to enhance the reconstruction quality of spatial details, we design the LiDAR-guided neural implicit field reconstruction unit (L-GNIF Unit). By learning a continuous mapping from coordinates to features, it achieves high-fidelity and artifact-free feature space reconstruction. Furthermore, we innovatively integrate spectral normalization constraints with a multi-level feature matching mechanism to construct the PSR-Discriminator. This discriminator provides more comprehensive perceptual signals across three scales: shallow textures, mid-level structures, and deep semantics. The entire framework is optimized through end-to-end training and a joint multi-task optimization loss function, ensuring that the generated fused features exhibit both authenticity and class discriminability. On this basis, we further design a spatial-spectral refinement classifier (SSR-Classifier) to accurately decode the deeply optimized feature maps, ultimately producing high-precision land cover classification results. Experiments demonstrate ELDGG’s superiority over state-of-the-art methods in both fusion quality and classification accuracy.
Effect of short dentin etching and water storage on bonding of resin composite to dentin with universal and two-step self-etch adhesive systems
Abstract To evaluate the effect of short dentin etching and 6-month water storage on the microtensile bond strength (µTBS) of universal and two-step self-etch adhesives to dentin. Mid-coronal dentin specimens obtained from 56 third molars were assigned into two groups according to the adhesive type ( n = 28); universal adhesive (Scotchbond Universal Plus Adhesive, 3 M Oral Care) and two-step self-etch adhesive (Clearfil SE Bond, Kuraray Noritake). Each group was further divided into two subgroups ( n = 14) based on the application protocol: self-etch (SE) and etch-and-rinse with short dentin etching (E&R/SDE) for 3 s. After bonding and composite application, half of the specimens were stored in 37 ± 1 °C distilled water for 24 h (immediate), while the other half were stored for 6 months (aged). Thereafter, specimens were cut into 1 mm² beams using a slow-speed diamond saw under copious water cooling. The beams were then subjected to a tensile force at a cross-head speed of 0.5 mm/min in a universal testing machine until failure occurred. The µTBS was then calculated in megapascals (MPa) by dividing the load at failure by the cross-sectional area of each beam. The values of five beams were averaged to obtain one µTBS value per tooth, and accordingly, data were analyzed using three-way ANOVA and Tukey HSD post-hoc tests ( p < 0.05). Failure modes were recorded. The application protocol and the storage time significantly influenced the µTBS results ( p < 0.001). Regarding the application protocol, E&R/SDE for 3 s yielded significantly higher bond strength values than the SE for both adhesives ( p < 0.05). In terms of storage-time, all groups exhibited a statistically significant reduction in bond strength after 6-month water storage ( p < 0.05). On the contrary, no statistically significant difference was detected between the two adhesives irrespective of the application protocol or the storage time ( p > 0.05). The predominant failure mode observed for immediate groups was mixed failure, while adhesive failure was the most frequently noted after 6 months. Despite the beneficial effect of E&R/SDE for 3 s in improving the bond strength of universal and 2-step self-etch adhesives to dentin, the 6-month water storage negatively affected the bonding performance of both adhesives. Clinical relevance: E&R/SDE enhanced μTBS at both storage times and may contribute to better bond stability, although all groups exhibited degradation after 6-month water storage, which necessitates further clinical validation.
A lightweight cross-scale EDS-DETR model for hazard detection in transmission corridors
Retraction Note: Counter-ion dependent, longitudinal unzipping of multi-walled carbon nanotubes to highly conductive and transparent graphene nanoribbons
In vitro and in silico analysis of anticancer and antioxidant potential of camphor derivatives
Qwen TextCNN and BERT models for enhanced multilabel news classification in mobile apps
Abstract Mobile news classification systems face significant challenges due to their large scale and complexity. In this paper, we perform a comprehensive comparative study between traditional classification models, such as TextCNN and BERT based models and Large Language Models (LLMs), for the purpose of multi-label news categorization in mobile apps about the Chinese mobile news application. We evaluated the performance of conventional techniques, including a BERT model, along with Qwen models that have been tuned with instruction and fine-tuned using the LoRA technique, to optimize their effectiveness while preserving classification accuracy. Our experimental results show that BERT models perform best for multi-label classification with balanced datasets, while textCNN performs better for binary classification tasks. Our results also reveal that the LSTM and MLP classifiers consistently achieve the highest accuracy with text instruction prompts, while random embeddings achieve competitive accuracy. Furthermore, despite the low macro F1 scores due to class imbalance, consistent relative performance confirms the validity of our analysis. Our research reveals crucial information about the classification of automotive news, highlighting the importance of weighing technical prowess against deployment constraints when choosing model architectures.