Browse Articles

An improved UAV image object detection algorithm combining multi-scale feature fusion and receptive-field attention-based convolution

Taxonomical loss for weed seedlings image classification

Abstract Accurate classification of weed seedlings is a key challenge to address in order to advance precision weed management and reduce pesticide use. In this study, an original image dataset, the Weed Phenological Dataset (WPD), with annotated early growth stages is introduced. Furthermore, a novel deep learning taxonomic loss function for few-shot learning was evaluated. Using hierarchical structures to direct the classification process, this taxonomic loss function introduces dynamic margins during the computation. In experiments using the taxonomic loss using the ResNet-50 architectures across different plant image datasets, this taxonomic approach allowed better clustering according to the Silhouette scores when compared with triplet loss using 100 images per class. It also led to better identification of weed seedlings at early growth stages. Although the new taxonomic loss did not consistency improve classification results for all the deep learning model architectures, it open new research avenues for the robotization of agriculture.

Phytochemical characterisation and antifungal activities of five botanicals used by subsistence farmers to manage plant diseases

Research on the structural parameters of loess based on shear strength

Effects of pulsating flow frequency and dimensionless amplitude on the thermal performance of SEGS LS-2 parabolic trough solar collector

Trade-off between canonical and unusual recombination sites promotes diversity and stability of gene cassette arrays of mobile integrons

Abstract Mobile integrons are the most efficient mechanism of Gram-negative bacteria to resist antimicrobial changing pressures in the nosocomial niche. Integron’s integrases mediate site-specific recombination of distinct DNA structures such as attI , attC and attG sites. Here, we identified novel Δ attI sites as part of 26 unusual Δ attI -type gene cassettes conferring resistance to four different antibiotic families. Since scarce data are found related to their functionality, we investigated site-specific recombination mediated by IntI1 of attI1 - aadB - attI1 , attI1 - aadB -Δ attI1 − 11 , attC aadA1 - aadB -Δ attI2 − 238 , and attC aadA1 - ybeA -Δ attI2 − 11 genetic architectures. All proved to be functional with some displaying excision and insertion rates equal to canonical gene cassettes. Interestingly, gene cassettes with the same recombination site upstream and downstream of the ORF, i.e. either two equal attI1 or two attC sites, had an excision frequency of more than 97%, which outlines a scenario in which the canonical gene cassette, once inserted, is almost instantaneously excised. These findings evidence that a trade-off between different recombination sites, including attI , attC and Δ attI sites in canonical and unusual Δ attI -type gene cassettes, is necessary to maintain a stable and diverse gene cassette array within the variable region of mobile integrons.

Comparative evaluation of HTG and TempO Seq targeted transcriptome profiling methods

Latency and energy-aware adaptive service migration in mobile edge computing

Determinants of moderate-intensity physical activity during pregnancy based on the COM-B model

Improving free-space continuous variable quantum key distribution with adaptive optics

Abstract A significant performance inhibitor of free-space continuous variable quantum key distribution (CVQKD) is turbulence, which gives rise to wavefront phase and amplitude aberrations. We demonstrate that in a turbulent channel, during coherent state transmissions from a continuous-wave laser, that the interferometric visibility between the local oscillator (LO) and quantum signal decreases. A solution to this is incorporating adaptive optics at the receiver to correct phase and amplitude aberrations in the wavefronts of the received quantum signal. We demonstrate the increased interferometric visibility and decrease in its fluctuations in a 60 cm and 30 m turbulent channel when using adaptive optics through channel characterisation. In an ideal CVQKD system, we show that this leads to more precise and larger positive secret key rates, improving the performance of free-space CVQKD in turbulent channels.

Characterization of biomass based novel microcrystalline cellulose from Eucalyptus teriticornis leaf

Experimental and theoretical evaluation of geometry-dependent doxorubicin loading onto cerium oxide nanoparticles via van der Waals interaction modeling

Water quality index prediction via a robust machine learning model using oxygen-related indices for river water quality monitoring

Abstract Rivers face increasing pollution, requiring accurate water quality assessment tools. Existing indices like the Water Quality Index (WQI) often overlook the integration of oxygen-related parameters critical to aquatic health. Here, we develop a machine learning model using Support Vector Regression (SVR) to predict the Water Quality Index (WQI OIs ) by integrating key oxygen-related parameters, including Biological Oxygen Demand (BOD), Chemical Oxygen Demand (COD), Dissolved Oxygen (DO), and the reaeration coefficients (K 1 , K 2 ). Applied to three rivers in Iran, the model demonstrated high accuracy, with a cross-validated R² > 0.95 and root mean squared error (RMSE) of 0.92 for the Haraz River and 1.41 for the Simineh River. Predictions showed strong correlation ( r  = 0.98) with standard indices, and feature importance analysis revealed DO as the most influential parameter. The model’s generalizability was confirmed through validation on independent river datasets, highlighting its robustness across diverse hydrological conditions. This approach offers a scalable, interpretable framework for continuous water quality monitoring, enabling more precise and data-driven management of aquatic ecosystems, particularly in regions with varying environmental factors.

Patterns and associations of summer thermal comfort and students’ physical activity in campus green spaces

HIV treatment cascade and associated factors among men who have sex with men in Brazil: a cross-sectional study

Dynamic damage and crack propagation of granite under thermal shock: DEM modeling insights

Structure optimization design for supporting ring of casing head hanger in extra deep oil wells

SHP2 promotes osteosarcoma via regulating STAT3/TET3/HOXB2 signaling

An integrative approach to identify novel miRNA-mRNA interaction networks in LMNA-cardiomyopathy

Abstract Dilated cardiomyopathy caused by variants in the LMNA gene leads to malignant arrhythmogenic events, faster phenotype progression and high risk of sudden cardiac death. The pathophysiological mechanisms triggering disease progression remains poorly understood. We investigated the mRNA and miRNA transcriptome in the myocardial tissue of 50-week-old LMNA R249W mice developing dilated cardiomyopathy. We found 2148 genes and 53 miRNAs that were differentially expressed in LMNA R249W hearts. Gene ontology and pathway enrichments showed that differentially expressed genes were enriched mainly for fatty acid metabolism, muscle contraction, cell adhesion and dilated cardiomyopathy pathways. The miRNA-mRNA interactions analysis identified 2197 miRNA-target pairs with an anti-correlation between differentially expressed genes and miRNAs. Gene ontology and pathway enrichments revealed that the most significant functions of miRNA targets are mainly related to heart development, cardiac muscle contraction, fatty acid β-oxidation, cell adhesion and calcium binding pathways, among others. Our study provides new insights into the molecular mechanisms that determine dilated cardiomyopathy due to pathogenic variants in the LMNA gene, and identified several target pairs that are of potential interest for further studies.

Sickle cell disease detection in low-resource conditions using transfer-learning and contrastive-learning coupled with XAI

Abstract Sickle cell disease (SCD) is a severe hereditary blood disorder that affects millions worldwide, necessitating early and accurate detection to improve patient outcomes. State-of-the-art approaches for automatic detection of SCD use deep learning (DL) based models, which require a large amount of training data for efficient training. However, such large training datasets are often not available, significantly limiting the efficiency of DL-based models. In this paper, we propose different approaches to address this issue. Firstly, we propose to use a transfer-learning based approach, where we use pre-trained models like ResNet-50, DenseNet-121, and EfficientNet-B0 and fine-tune them for SCD detection. To further enhance the efficiency of the models, we then propose to include contrastive-learning-based approach using triplet loss. We also use focal loss to handle class imbalance. Additionally, we integrate Explainable Artificial Intelligence (XAI) methodologies to interpret and explain the model’s predictions, ensuring transparency and trustworthiness in clinical settings. Experiments on a publicly available SCD image dataset show that models trained with transfer learning and triplet loss outperform those trained with binary cross-entropy or focal loss.