Search Results (413)

Compiler optimization is a critical component for improving program performance. However, the Rustc optimization process may introduce vulnerabilities due to algorithmic flaws or issues arising from component interactions. Existing testing methods face several challenges, including high randomness in test cases, inadequate targeting of vulnerability-prone regions, and low-quality initial fuzzing seeds. This paper proposes a test case generation method based on large language models (LLMs), which utilizes prompt templates and optimization algorithms to generate a code relevant to specific optimization passes, especially for real-time control logic and safety-critical modules unique to the industrial control field. A vulnerability screening approach based on static analysis and rule matching is designed to locate potential risk points in the optimization regions of both the MIR and LLVM IR layers, as well as in unsafe code sections. Furthermore, the targeted fuzzing strategy is enhanced by designing seed queues and selection algorithms that consider the correlation between optimization areas. The implemented system, RustOptFuzz, has been evaluated on both custom datasets and real-world programs. Compared with state-of-the-art tools, RustOptFuzz improves vulnerability discovery capabilities by 16%–50% and significantly reduces vulnerability reproduction time, thereby enhancing the overall efficiency of detecting optimization-related vulnerabilities in Rustc, providing key technical support for the reliability of industrial control systems. Full article

This paper outlines a methodology to compile geo-referenced observational data of Australian birds acting as pollinators of Strelitzia sp. (Bird of Paradise) flowers and dispersers of their seeds. Given the absence of systematic published records, a crowdsourcing approach was employed, combining data from natural history platforms (e.g., iNaturalist, eBird), image hosting websites (e.g., Flickr) and, in particular, social media. Facebook emerged as the most productive channel, with 61.4% of the 301 usable observations sourced from 43 ornithology-related groups. The strategy included direct solicitation of images and metadata via group posts and follow-up communication. The holistic, snowballing search strategy yielded a unique, behavior-focused dataset suitable for analysis. While the process exposed limitations due to user self-censorship on image quality and completeness, the approach demonstrates the viability of crowdsourced behavioral ecology data and contributes a replicable methodology for similar studies in under-documented ecological contexts. Full article

Hyperspectral imaging for seed vigor detection faces the challenges of handling high-dimensional spectral data, information loss after dimensionality reduction, and low feature differentiation between vigor levels. To address the above issues, this study proposes an improved dynamic optimize MDS (DO-MDS) dimensionality reduction algorithm based on multidimensional scaling transformation. DO-MDS better preserves key features between samples during dimensionality reduction. Secondly, a dual-stream spectral collaborative attention (DSCA) module is proposed. The DSCA module adopts a dual-modal fusion approach combining global feature capture and local feature enhancement, deepening the characterization capability of spectral features. This study selected commonly used rice seed varieties in Zhejiang Province and constructed three individual spectral datasets and a mixed dataset through aging, spectral acquisition, and germination experiments. The experiments involved using the DO-MDS processed datasets with a convolutional neural network embedded with the DSCA attention module, and the results demonstrate vigor discrimination accuracy rates of 93.85%, 93.4%, and 96.23% for the Chunyou 83, Zhongzao 39, and Zhongzu 53 datasets, respectively, achieving 94.8% for the mixed dataset. This study provides effective strategies for spectral dimensionality reduction in hyperspectral seed vigor detection and enhances the differentiation of spectral information for seeds with similar vigor levels. Full article

Fall detection is a critical task in healthcare monitoring systems, especially for elderly populations, for whom timely intervention can significantly reduce morbidity and mortality. This study proposes a privacy-preserving and scalable fall-detection framework that integrates federated learning (FL) with transfer learning (TL) to train deep learning models across decentralized data sources without compromising user privacy. The pipeline begins with data acquisition, in which annotated video-based fall-detection datasets formatted in YOLO are used to extract image crops of human subjects. These images are then preprocessed, resized, normalized, and relabeled into binary classes (fall vs. non-fall). A stratified 80/10/10 split ensures balanced training, validation, and testing. To simulate real-world federated environments, the training data is partitioned across multiple clients, each performing local training using pretrained CNN models including MobileNetV2, VGG16, EfficientNetB0, and ResNet50. Two FL topologies are implemented: a centralized server-coordinated scheme and a ring-based decentralized topology. During each round, only model weights are shared, and federated averaging (FedAvg) is applied for global aggregation. The models were trained using three random seeds to ensure result robustness and stability across varying data partitions. Among all configurations, decentralized MobileNetV2 achieved the best results, with a mean test accuracy of 0.9927, F1-score of 0.9917, and average training time of 111.17 s per round. These findings highlight the model’s strong generalization, low computational burden, and suitability for edge deployment. Future work will extend evaluation to external datasets and address issues such as client drift and adversarial robustness in federated environments. Full article

Nested named entity recognition (NER) is a task that identifies hierarchically structured entities, where one entity can contain other entities within its span. This study introduces a nested NER model for few-shot learning environments, addressing the difficulty of building extensive datasets for general named entities. We enhance the Biaffine nested NER model by modifying its output layer to incorporate label semantic information through a novel label description embedding (LDE) approach, improving performance with limited training data. Our method replaces the traditional biaffine classifier with a label attention mechanism that leverages comprehensive natural language descriptions of entity types, encoded using BERT to capture rich semantic relationships between labels and input spans. We conducted comprehensive experiments on four benchmark datasets: GENIA (nested NER), ACE 2004 (nested NER), ACE 2005 (nested NER), and CoNLL 2003 English (flat NER). Performance was evaluated across multiple few-shot scenarios (1-shot, 5-shot, 10-shot, and 20-shot) using F1-measure as the primary metric, with five different random seeds to ensure robust evaluation. We compared our approach against strong baselines including BERT-LSTM-CRF with nested tags, the original Biaffine model, and recent few-shot NER methods (FewNER, FIT, LPNER, SpanNER). Results demonstrate significant improvements across all few-shot scenarios. On GENIA, our LDE model achieves 45.07% F1 in five-shot learning compared to 30.74% for the baseline Biaffine model (46.4% relative improvement). On ACE 2005, we obtain 44.24% vs. 32.38% F1 in five-shot scenarios (36.6% relative improvement). The model shows consistent gains in 10-shot (57.19% vs. 49.50% on ACE 2005) and 20-shot settings (64.50% vs. 58.21% on ACE 2005). Ablation studies confirm that semantic information from label descriptions is the key factor enabling robust few-shot performance. Transfer learning experiments demonstrate the model’s ability to leverage knowledge from related domains. Our findings suggest that incorporating label semantic information can substantially enhance NER models in low-resource settings, opening new possibilities for applying NER in specialized domains or languages with limited annotated data. Full article

There is a lack of systematic research on the comprehensive regulatory effects of urea and organic fertilizer application on soil quality and cotton yield in summer direct-seeded cotton fields in the Yangtze River Basin. Additionally, there is a redundancy of indicators in the cotton field soil quality evaluation system and a lack of reports on constructing a minimum dataset to evaluate the soil quality status of cotton fields. We aim to accurately and efficiently evaluate soil quality in cotton fields and screen nitrogen application measures that synergistically improve soil quality, cotton yield, and nitrogen fertilizer utilization efficiency. Taking the summer live broadcast cotton field in Jiangxi Province as the research object, four treatments, including CK without nitrogen application, CF with conventional nitrogen application, N1 with nitrogen reduction, and N2 with nitrogen reduction and organic fertilizer application, were set up for three consecutive years from 2022 to 2024. A total of 15 physical, chemical, and biological indicators of the 0–20 cm plow layer soil were measured in each treatment. A minimum dataset model was constructed to evaluate and verify the soil quality status of different nitrogen application treatments and to explore the physiological mechanisms of nitrogen application on yield performance and stability from the perspectives of cotton source–sink relationship, nitrogen use efficiency, and soil quality. The minimum dataset for soil quality evaluation in cotton fields consisted of five indicators: soil bulk density, moisture content, total nitrogen, organic carbon, and carbon-to-nitrogen ratio, with a simplification rate of 66.67% for the evaluation indicators. The soil quality index calculated based on the minimum dataset (MDS) was significantly positively correlated with the soil quality index of the total dataset (TDS) (R² = 0.904, p < 0.05). The model validation parameters RMSE was 0.0733, nRMSE was 13.8561%, and the d value was 0.9529, all indicating that the model simulation effect had reached a good level or above. The order of soil quality index based on MDS and TDS for CK, CF, N1, and N2 treatments was CK < N1 < CF < N2. The soil quality index of N2 treatment under MDS significantly increased by 16.70% and 26.16% compared to CF and N1 treatments, respectively. Compared with CF treatment, N2 treatment significantly increased nitrogen fertilizer partial productivity by 27.97%, 31.06%, and 21.77%, respectively, over a three-year period while maintaining the same biomass, yield level, yield stability, and yield sustainability. Meanwhile, N1 treatment had the risk of significantly reducing both boll density and seed cotton yield. Compared with N1 treatment, N2 treatment could significantly increase the biomass of reproductive organs during the flower and boll stage by 23.62~24.75% and the boll opening stage by 12.39~15.44%, respectively, laying a material foundation for the improvement in yield and yield stability. Under CF treatment, the cotton field soil showed a high degree of soil physical property barriers, while the N2 treatment reduced soil barriers in indicators such as bulk density, soil organic carbon content, and soil carbon-to-nitrogen ratio by 0.04, 0.04, 0.08, and 0.02, respectively, compared to CF treatment. In summary, the minimum dataset (MDS) retained only 33.3% of the original indicators while maintaining high accuracy, demonstrating the model’s efficiency. After reducing nitrogen by 20%, applying 10% total nitrogen organic fertilizer could substantially improve cotton biomass, cotton yield performance, yield stability, and nitrogen partial productivity while maintaining soil quality levels. This study also assessed yield stability and sustainability, not just productivity alone. The comprehensive nitrogen fertilizer management (reducing N + organic fertilizer) under the experimental conditions has high practical applicability in the intensive agricultural system in southern China. Full article

Due to the limitations of the sowing machine performance and rice seed germination rates, missing seedlings inevitably occur after rice is sown in large fields. This phenomenon has a direct impact on the rice yield. In the field environment, the existing methods for detecting missing seedlings based on unmanned aerial vehicle (UAV) remote sensing images often have unsatisfactory effects. Therefore, to enable the fast and accurate detection of missing rice seedlings and facilitate subsequent reseeding, this study proposes a UAV remote-sensing-based method for detecting missing rice seedlings in large fields. The proposed method uses an improved PCERT-DETR model to detect rice seedlings and missing seedlings in UAV remote sensing images of large fields. The experimental results show that PCERT-DETR achieves an optimal performance on the self-constructed dataset, with an mean average precision (mAP) of 81.2%, precision (P) of 82.8%, recall (R) of 78.3%, and F₁-score (F₁) of 80.5%. The model’s parameter count is only 21.4 M and its FLOPs reach 66.6 G, meeting real-time detection requirements. Compared to the baseline network models, PCERT-DETR improves the P, R, F₁, and mAP by 15.0, 1.2, 8.5, and 6.8 percentage points, respectively. Furthermore, the performance evaluation experiments were carried out through ablation experiments, comparative detection model experiments and heat map visualization analysis, indicating that the model has a strong detection performance on the test set. The results confirm that the proposed model can accurately detect the number of missing rice seedlings. This study provides accurate information on the number of missing seedlings for subsequent reseeding operations, thus contributing to the improvement of precision farming practices. Full article

Accurate liver segmentation and volume estimation from CT images are critical for diagnosis, surgical planning, and treatment monitoring. This paper proposes a GPU-accelerated voxel-level cropping method that localizes the liver region in a single pass, significantly reducing unnecessary computation and memory transfers. We integrate this pre-processing step into two segmentation pipelines: a traditional Chan-Vese model and a deep learning U-Net trained on the LiTS dataset. After segmentation, a seeded region growing algorithm is used for 3D liver volume assessment. Our method reduces unnecessary image data by an average of 90%, speeds up segmentation by 1.39× for Chan-Vese, and improves dice scores from 0.938 to 0.960. When integrated into U-Net pipelines, the post-processed dice score rises drastically from 0.521 to 0.956. Additionally, the voxel-based cropping approach achieves a 2.29× acceleration compared to state-of-the-art slice-based methods in 3D volume assessment. Our results demonstrate high segmentation accuracy and precise volume estimates with errors below 2.5%. This proposal offers a scalable, interpretable, efficient liver segmentation and volume assessment solution. It eliminates unwanted artifacts and facilitates real-time deployment in clinical environments where transparency and resource constraints are critical. It is also tested in other anatomical structures such as skin, lungs, and vessels, enabling broader applicability in medical imaging. Full article

The reliance on imported seeds for grassland rehabilitation in Mexico has led to increased costs and other difficulties in implementing grassland rehabilitation programs. Varieties need to be generated from local ecotypes that are outstanding in forage production and their response to rehabilitation programs. However, the scarcity of occurrence records is often a deterrent to niche and distribution modeling, hence the need for an approach that overcomes such limitations. The objectives of this study were (1) to develop a geographic information system (GIS)-based approach to determining the population distribution of a promising ecotype of Bouteloua curtipendula (Michx.) Torr. for grassland rehabilitation in the Chihuahuan Desert, Mexico; (2) to identify the edaphoclimatic variables that define the ecotype’s distribution; and (3) to develop models to determine the potential area for the use of the ecotype in grassland rehabilitation. The challenge for the present study was that only one georeferenced collection site of the ecotype in Chihuahua was available for use in the construction and calibration of the models. GIS software 10.3 was used to develop two potential distribution models: Model A, with variables obtained directly from a vector climate dataset, and Model B, with derived variables. A field work methodology was developed for the validation process using a georeferenced digital mesh and the nested sampling method modified by Whittaker. The information was analyzed with 10 non-parametric statistical tests. The two models had an overall accuracy and sensitivity level greater than 70% and a positive predictive power greater than 80%. The predicted population distribution areas in Chihuahua (18,158 ha) in the form of discontinuous patches cohered with those in previous reports on the distribution form of B. curtipendula. The edaphoclimatic variables influencing ecotype distribution were soil type, average minimum and maximum temperature in January, average maximum temperature in June, average minimum temperature in July, and average precipitation in August. The sensitivity analysis showed soil type as an important variable in defining the ecotype’s distribution. Considering soil as the main predictor variable, the potential rehabilitation area where the ecotype may be used was estimated at 7,181,735 ha in the Chihuahuan Desert region. The study developed and validated an approach to modeling the ecological niche of an ecotype of commercial interest, despite severe limitations in the number of georeferenced sites available for modeling. Further study is needed to explore its applicability to grassland rehabilitation in the Chihuahuan Desert and the study of rare and understudied ecotypes or species in other settings. Full article

Backdoor attacks in self-supervised learning pose an increasing threat. Recent studies have shown that knowledge distillation can mitigate these attacks by altering feature representations. In response, we propose BASK, a novel backdoor attack that remains effective after distillation. BASK uses feature weighting and representation alignment strategies to implant persistent backdoors into the encoder’s feature space. This enables transferability to student models. We evaluated BASK on the CIFAR-10 and STL-10 datasets and compared it with existing self-supervised backdoor attacks under four advanced defenses: SEED, MKD, Neural Cleanse, and MiMiC. Our experimental results demonstrate that BASK maintains high attack success rates while preserving downstream task performance. This highlights the robustness of BASK and the limitations of current defense mechanisms. Full article

Understanding the influence of environmental factors on cotton performance is crucial for enhancing yield and fiber quality in the context of climate change. This study investigates genotype-by-environment (G×E) interactions in cotton, using data from 250 recombinant inbred lines (CCRI70 RILs) cultivated across 14 diverse environments in China’s major cotton cultivation areas. Our findings reveal that environmental effects predominantly influenced yield-related traits (boll weight, lint percentage, and the seed index), contributing to 34.7% to 55.7% of their variance. In contrast fiber quality traits showed lower environmental sensitivity (12.3–27.0%), with notable phenotypic plasticity observed in the boll weight, lint percentage, and fiber micronaire. Employing six machine learning models, Random Forest demonstrated superior predictive ability (R² = 0.40–0.72; predictive Pearson correlation = 0.63–0.86). Through SHAP-based interpretation and sliding-window regression, we identified key environmental drivers primarily active during mid-to-late growth stages. This approach effectively reduced the number of influential input variables to just 0.1–2.4% of the original dataset, spanning 2–9 critical time windows per trait. Incorporating these identified drivers significantly improved cross-environment predictions, enhancing Random Forest accuracy by 0.02–0.15. These results underscore the strong potential of machine learning to uncover critical temporal environmental factors underlying G×E interactions and to substantially improve predictive modeling in cotton breeding programs, ultimately contributing to more resilient and productive cotton cultivation. Full article

EEG-based emotion recognition (EEG-ER) through deep learning models has gained more attention in recent years, with more researchers focusing on architecture, feature extraction, and generalisability. This paper presents a novel end-to-end deep learning framework for EEG-ER, combining temporal feature extraction, self-attention mechanisms, and adversarial domain adaptation. The architecture entails a multi-stage 1D CNN for spatiotemporal features from raw EEG signals, followed by a transformer-based attention module for long-range dependencies, and a domain-adversarial neural network (DANN) module with gradient reversal to enable a powerful subject-independent generalisation by learning domain-invariant features. Experiments on benchmark datasets (DEAP, SEED, DREAMER) demonstrate that our approach achieves a state-of-the-art performance, with a significant improvement in cross-subject recognition accuracy compared to non-adaptive frameworks. The architecture tackles key challenges in EEG emotion recognition, including generalisability, inter-subject variability, and temporal dynamics modelling. The results highlight the effectiveness of combining convolutional feature learning with adversarial domain adaptation for robust EEG-ER. Full article

Maize is a vital global crop, serving as a fundamental component of global food security. To support sustainable maize production, the accurate classification of maize seeds—particularly distinguishing haploid from diploid types—is essential for enhancing breeding efficiency. Conventional methods relying on manual inspection or simple machine learning are prone to errors and unsuitable for large-scale data. To overcome these limitations, we propose Slot-Maize, a novel deep learning architecture that integrates Convolutional Neural Networks (CNN), Slot Attention, Gated Recurrent Units (GRU), and Long Short-Term Memory (LSTM) layers. The Slot-Maize model was evaluated using two datasets: the Maize Seed Dataset and the Maize Variety Dataset. The Slot Attention module improves feature representation by focusing on object-centric regions within seed images. The GRU captures short-term sequential patterns in extracted features, while the LSTM models long-range dependencies, enhancing temporal understanding. Furthermore, Grad-CAM was utilized as an explainable AI technique to enhance the interpretability of the model’s decisions. The model demonstrated an accuracy of 96.97% on the Maize Seed Dataset and 92.30% on the Maize Variety Dataset, outperforming existing methods in both cases. These results demonstrate the model’s robustness, generalizability, and potential to accelerate automated maize breeding workflows. In conclusion, the Slot-Maize model provides a robust and interpretable solution for automated maize seed classification, representing a significant advancement in agricultural technology. By combining accuracy with explainability, Slot-Maize provides a reliable tool for precision agriculture. Full article

Seed microbes play crucial roles in plant health, but studying their diversity is challenging due to host DNA contamination. This study aimed to optimise methodologies for investigating seed microbiomes across diverse plant species, focusing on the efficacy of peptide nucleic acid (PNA) clamps to reduce host DNA amplification. We tested PNA clamps on three plant species: Melaleuca quinquenervia (tree), Microlaena stipoides, and Themeda triandra (grasses). The effectiveness of PNA clamps was assessed through in silico analysis, axenic tissue culture, and metabarcoding techniques. In silico analysis confirmed the specificity of PNA clamps to the 16S rRNA gene V4 region of chloroplasts in the grass species. Axenic tissue culture experiments showed that applying PNA clamps at both 1 µM and 0.25 µM concentrations significantly reduced plant DNA amplification. Metabarcoding analyses further confirmed that PNA clamps effectively suppressed host DNA, enhancing microbial diversity estimates across all three species while preserving core microbial taxa. The efficacy of the clamps varied among host species, with T. triandra exhibiting the highest blocking efficacy, and chloroplast clamps outperforming mitochondrial ones. This study demonstrates that PNA clamps are a useful for improving seed endophyte metabarcoding datasets, although they require optimisation for some plant species. This knowledge will contribute to enhancing our understanding of seed microbiome diversity and its ecological implications. Full article

Keeping drinking water safe is a critical aspect of protecting public health. Traditional laboratory-based methods for evaluating water potability are often time-consuming, costly, and labour-intensive. This paper presents a comparative analysis of four transformer-based deep learning models in the development of automatic classification systems for water potability based on physicochemical attributes. The models examined include the enhanced tabular transformer (ETT), feature tokenizer transformer (FTTransformer), self-attention and inter-sample network (SAINT), and tabular autoencoder pretraining enhancement (TAPE). The study utilized an open-access water quality dataset that includes nine key attributes such as pH, hardness, total dissolved solids (TDS), chloramines, sulphate, conductivity, organic carbon, trihalomethanes, and turbidity. The models were evaluated under a unified protocol involving 70–15–15 data partitioning, five-fold cross-validation, fixed random seed, and consistent hyperparameter settings. Among the evaluated models, the enhanced tabular transformer outperforms other models with an accuracy of 95.04% and an F1 score of 0.94. ETT is an advanced model because it can efficiently model high-order feature interactions through multi-head attention and deep hierarchical encoding. Feature importance analysis consistently highlighted chloramines, conductivity, and trihalomethanes as key predictive features across all models. SAINT demonstrated robust generalization through its dual-attention mechanism, while TAPE provided competitive results with reduced computational overhead due to unsupervised pretraining. Conversely, FTTransformer showed limitations, likely due to sensitivity to class imbalance and hyperparameter tuning. The results underscore the potential of transformer-based models, especially ETT, in enabling efficient, accurate, and scalable water quality monitoring. These findings support their integration into real-time environmental health systems and suggest approaches for future research in explainability, domain adaptation, and multimodal fusion. Full article