Mar Pollut Bull. 2025 Jan 24;212:117583. doi: 10.1016/j.marpolbul.2025.117583. Online ahead of print.

ABSTRACT

Marine pollution due to oil spills presents major risks to coastal areas and aquatic life, leading to serious environmental health concerns. Oil Spill detection using SAR data has transitioned from traditional segmentation to a variety of machine learning & deep learning models like UNET proving its efficiency for the task. This research paper proposes a GSCAT-UNET model for efficient oil spill detection and discrimination from lookalikes. The GSCAT-UNET is an advanced UNET architecture comprising of Spatial-Channel Attention Gates(SCAG), Three Level Attention Module(TLM) and Global Feature Module(GFM) for global level oil spill feature enhancement leading to effective oil spill detection and discrimination from lookalikes. Sentinel-1 Dual-Pol SAR dataset of 1112 images and respective labeled images (5 classes) including confirmed oil spills and lookalikes is used to demonstrate the efficacy of the GSCAT-UNET model. The GSCAT-UNET model significantly enhances segmentation accuracy and robustness for oil spill detection with 5% higher accuracy and 29% higher IoU i.e. 93.7% compared to the UNET segmentation model, addressing the challenges of SAR data complexities and imbalanced datasets. The strong performance of the GSCAT-UNET model demonstrates its potential as a critical tool for disaster response and environmental monitoring.

PMID:39862681 | DOI:10.1016/j.marpolbul.2025.117583

J Electrocardiol. 2025 Jan 22;89:153882. doi: 10.1016/j.jelectrocard.2025.153882. Online ahead of print.

ABSTRACT

PURPOSE OF REVIEW: WHO defines SCD as sudden unexpected death either within 1 h of symptom onset (witnessed) or within 24 h of having been observed alive and symptom-free (unwitnessed). Sudden cardiac arrest is a major cause of mortality worldwide, with survival to hospital discharge for hospital cardiac arrest and in-hospital cardiac arrest being only 9.3 % and 21.2 %, respectively, despite treatment highlighting the importance of effectively predicting and preventing cardiac arrest. This literature review aims to explore the role and application of AI (Artificial Intelligence) in predicting and preventing sudden cardiac arrest.

MATERIAL AND METHODS: Eligible studies were searched from PubMed and Web of Science. The inclusion criteria were fulfilled if sudden cardiac death prediction and prevention, artificial intelligence, machine learning, and deep learning were included.

CONCLUSIONS: Artificial intelligence, machine learning, and deep learning have shown remarkable prospects in SCA risk stratification, which can improve the survival rate from SCA. Nonetheless, they have not been adequately trained and tested, necessitating further studies with explainable techniques, larger sample sizes, external validation, more diverse patient samples, multimodal tools, ethics, and bias mitigation to unlock their full potential.

PMID:39862597 | DOI:10.1016/j.jelectrocard.2025.153882

Int J Med Inform. 2025 Jan 21;196:105793. doi: 10.1016/j.ijmedinf.2025.105793. Online ahead of print.

ABSTRACT

BACKGROUND: In the context of routine breast cancer diagnosis, the precise discrimination between benign and malignant breast masses holds utmost significance. Notably, few prior investigations have concurrently explored the integration of imaging histology features, deep learning characteristics, and clinical parameters. The primary objective of this retrospective study was to pioneer a multimodal feature fusion model tailored for the prediction of breast tumor malignancy, harnessing the potential of ultrasound images.

METHOD: We compiled a dataset that included clinical features from 1065 patients and 3315 image datasets. Specifically, we selected data from 603 patients for training our multimodal model. The comprehensive experimental workflow involves identifying the optimal unimodal model, extracting unimodal features, fusing multimodal features, gaining insights from these fused features, and ultimately generating prediction results using a classifier.

RESULTS: Our multimodal feature fusion model demonstrates outstanding performance, achieving an AUC of 0.994 (95 % CI: 0.988-0.999) and an F1 score of 0.971 on the primary multicenter dataset. In the evaluation on two independent testing cohorts (TCs), it maintains strong performance, with AUCs of 0.942 (95 % CI: 0.854-0.994) for TC1 and 0.945 (95 % CI: 0.857-1.000) for TC2, accompanied by corresponding F1 scores of 0.872 and 0.857, respectively. Notably, the decision curve analysis reveals that our model achieves higher accuracy within the threshold probability range of approximately [0.210, 0.890] (TC1) and [0.000, 0.850] (TC2) compared to alternative methods. This capability enhances its utility in clinical decision-making, providing substantial benefits.

CONCLUSION: The multimodal model proposed in this paper can comprehensively evaluate patients' multifaceted clinical information, achieve the prediction of benign and malignant breast ultrasound tumors, and obtain high performance indexes.

PMID:39862564 | DOI:10.1016/j.ijmedinf.2025.105793

Neural Netw. 2025 Jan 22;185:107185. doi: 10.1016/j.neunet.2025.107185. Online ahead of print.

ABSTRACT

Model-based diagnosis (MBD) is a critical problem in artificial intelligence. Recent advancements have made it possible to address this challenge using methods like deep learning. However, current approaches that use deep learning for MBD often struggle with accuracy and computation time due to the limited diagnostic information provided by a single observation. To address this challenge, we introduce two novel algorithms, Discret2DiMO (Discret2Di with Multiple Observations) and Discret2DiMO-DC (Discret2Di with Multiple Observations and Dictionary Cache), which enhance MBD by integrating multiple observations with deep learning techniques. Experimental evaluations on a simulated three-tank model demonstrate that Discret2DiMO significantly improves diagnostic accuracy, achieving up to a 685.06% increase and an average improvement of 59.18% over Discret2Di across all test cases. To address computational overhead, Discret2DiMO-DC additionally implements a caching mechanism that eliminates redundant computations during diagnosis. Remarkably, Discret2DiMO-DC achieves comparable accuracy while reducing computation time by an average of 95.74% compared to Discret2DiMO and 89.42% compared to Discret2Di, with computation times reduced by two orders of magnitude. These results indicate that our proposed algorithms significantly enhance diagnostic accuracy and efficiency in MBD compared with the state-of-the-art algorithm, highlighting the potential of integrating multiple observations with deep learning for more accurate and efficient diagnostics in complex systems.

PMID:39862533 | DOI:10.1016/j.neunet.2025.107185

Comput Methods Programs Biomed. 2025 Jan 15;261:108585. doi: 10.1016/j.cmpb.2025.108585. Online ahead of print.

ABSTRACT

BACKGROUND AND OBJECTIVES: Hypertensive Retinopathy (HR) is a retinal manifestation resulting from persistently elevated blood pressure. Severity grading of HR is essential for patient risk stratification, effective management, progression monitoring, timely intervention, and minimizing the risk of vision impairment. Computer-aided diagnosis and artificial intelligence (AI) systems play vital roles in the diagnosis and grading of HR. Over the years, very limited research has been conducted for the grading of HR. Nevertheless, there are no publicly available datasets for HR grading. Moreover, one of the key challenges observed is high-class imbalance.

METHODS: To address these issues, in this paper, we develop "HRSG: Expert-Annotated Hypertensive Retinopathy Severity Grading" dataset, classifying HR severity into four distinct classes: normal, mild, moderate, and severe. Further, to enhance the grading performance on limited datasets, this paper introduces a novel hybrid architecture that combines the strengths of pretrained ResNet-50 via transfer learning, and a modified Vision Transformer (ViT) architecture enhanced with a combination of global self-attention and locality self-attention mechanisms. The locality self-attention addresses the common issue of a lack of inductive bias in ViT architecture. This architecture effectively captures both local and global contextual information, resulting in a robust and resilient classification model. To overcome class imbalance, Decouple Representation and Classifier (DRC) - based training approach is proposed. This method improves the model's ability to learn effective features while preserving the original dataset's distribution, leading to better diagnostic accuracy.

RESULTS: Performance evaluation results show the competence of the proposed method in accurately grading the severity of HR. The proposed method achieved an average accuracy of 0.9688, sensitivity of 0.9435, specificity of 0.9766, F1-score of 0.9442, and precision of 0.9474. The comparative results indicate that the proposed method outperforms existing HR methods, state-of-the-art CNN models, and baseline pretrained ViT models. Additionally, we compared our method with a CNNViT model, which combines a shallow CNN architecture with 3 convolution blocks consisting of a convolution layer, a batch normalization layer, a max pooling layer, and lightweight ViT architecture, due to limited datasets. In comparison with the CNNViT, the proposed method achieved superior performance, demonstrating its effectiveness.

CONCLUSION: The experimental results demonstrate the efficacy of the proposed method in accurately grading HR severity.

PMID:39862474 | DOI:10.1016/j.cmpb.2025.108585

Comput Biol Med. 2025 Jan 24;186:109703. doi: 10.1016/j.compbiomed.2025.109703. Online ahead of print.

ABSTRACT

- Brain tumors (BT), both benign and malignant, pose a substantial impact on human health and need precise and early detection for successful treatment. Analysing magnetic resonance imaging (MRI) image is a common method for BT diagnosis and segmentation, yet misdiagnoses yield effective medical responses, impacting patient survival rates. Recent technological advancements have popularized deep learning-based medical image analysis, leveraging transfer learning to reuse pre-trained models for various applications. BT segmentation with MRI remains challenging despite advancements in image acquisition techniques. Accurate detection and segmentation are essential for proper diagnosis and treatment planning. This study aims to enhance BT detection and segmentation accuracy and effectiveness of categorization through the implementation of an advanced stacking ensemble learning (SEL) approach. This study explores the efficiency of SEL architecture in augmenting the precision of BT segmentation. SEL, a prominent approach within the machine learning paradigm, combines the predictions of base-level models and improves the overall performance of predictions in order to reduce the errors and biases of each model. The proposed approach involves designing a stacked DenseNet201 as the meta-model called SEL-DenseNet201, complemented by six diverse base models such as mobile network version 3 (MobileNet-v3), 3-dimensional convolutional neural network (3D-CNN), visual geometry group network with 16 and 19 layers (VGG-16 and VGG-19), residual network with 50 layers (ResNet50), and Alex network (AlexNet). The strengths of the base models are calculated to capture distinct aspects of the BT MRI, aiming for enhanced segmentation performance. The proposed SEL-DenseNet201 is trained using BT MRI datasets. The augmentation techniques are applied to MRI scans to balance and enhance the model performance through the application of image enhancement and segmentation techniques. The proposed SEL-DenseNet201 achieves impressive results with an accuracy of 99.65 % and a dice coefficient of 97.43 %. These outcomes underscore the superiority of the proposed model over existing approaches. This study holds the potential to be an initial screening approach for early BT detection, with a high success rate.

PMID:39862469 | DOI:10.1016/j.compbiomed.2025.109703

Comput Biol Med. 2025 Jan 24;186:109706. doi: 10.1016/j.compbiomed.2025.109706. Online ahead of print.

ABSTRACT

The fusion index is a critical metric for quantitatively assessing the transformation of in vitro muscle cells into myotubes in the biological and medical fields. Traditional methods for calculating this index manually involve the labor-intensive counting of numerous muscle cell nuclei in images, which necessitates determining whether each nucleus is located inside or outside the myotubes, leading to significant inter-observer variation. To address these challenges, this study proposes a three-stage process that integrates the strengths of pattern recognition and deep-learning to automatically calculate the fusion index. The experimental results demonstrate that the proposed process achieves significantly higher performance in cell nuclei detection and classification, with an F1-score of 0.953, whereas traditional object detection methods achieve less than 0.5. In addition, the fusion index obtained using the proposed method is closely aligned with the human-assessed values, showing minimal discrepancy and strong agreement with human evaluations. This process is incorporated into the development of "SEPO-FI" as public software, automating cell detection and classification to enable effective fusion index calculation and broaden access to this methodology within the scientific community.

PMID:39862466 | DOI:10.1016/j.compbiomed.2025.109706

NPJ Digit Med. 2025 Jan 26;8(1):56. doi: 10.1038/s41746-025-01454-z.

ABSTRACT

Deep-learning models have shown promise in differentiating between benign and malignant lesions. Previous studies have primarily focused on specific anatomical regions, overlooking tumors occurring throughout the body with highly heterogeneous whole-body backgrounds. Using neurofibromatosis type 1 (NF1) as an example, this study developed highly accurate MRI-based deep-learning models for the early automated screening of malignant peripheral nerve sheath tumors (MPNSTs) against complex whole-body background. In a Chinese seven-center cohort, data from 347 subjects were analyzed. Our one-step model incorporated normal tissue/organ labels to provide contextual information, offering a solution for tumors with complex backgrounds. To address privacy concerns, we utilized a lightweight deep neural network suitable for hospital deployment. The final model achieved an accuracy of 85.71% for MPNST diagnosis in the validation cohort and 84.75% accuracy in the independent test set, outperforming another classic two-step model. This success suggests potential for AI models in screening other whole-body primary/metastatic tumors.

PMID:39863790 | DOI:10.1038/s41746-025-01454-z

Sci Rep. 2025 Jan 25;15(1):3273. doi: 10.1038/s41598-025-87803-0.

ABSTRACT

We aimed to build a robust classifier for the MGMT methylation status of glioblastoma in multiparametric MRI. We focused on multi-habitat deep image descriptors as our basic focus. A subset of the BRATS 2021 MGMT methylation dataset containing both MGMT class labels and segmentation masks was used. A comprehensive mask fusion approach was developed to select relevant image crops of diseased tissue. These fusion masks, which were guided by multiple sequences, helped collect information from the regions that seem disease-free to radiologists in standard MRI sequences while harboring pathology. Integrating the information in different MRI sequences and leveraging the high entropic capacity of deep neural networks, we built a 3D ROI-based custom CNN classifier for the automatic prediction of MGMT methylation status of glioblastoma in multi-parametric MRI. Single sequence-based classifiers reached intermediate predictive performance with 0.65, 0.71, 0.77, and 0.82 accuracy for T1W, T2W, T1 contrast-enhanced, and FLAIR sequences, respectively. The multiparametric classifier using T1 contrast-enhanced and FLAIR images reached 0.88 accuracy. The accuracy of the four-input model that used all sequences was 0.81. The best model reached 0.90 ROC AUC value. Integrating human knowledge in the form of relevant target selection was a useful approach in MGMT methylation status prediction in MRI. Exploration of means to integrate radiology knowledge into the models and achieve human-machine collaboration may help to develop better models. MGMT methylation status of glioblastoma is an important prognostic marker and is also important for treatment decisions. The preoperative non-invasive predictive ability and the explanation tools of the developed model may help clinicians to better understand imaging phenotypes of MGMT methylation status of glial tumors.

PMID:39863759 | DOI:10.1038/s41598-025-87803-0

Sci Rep. 2025 Jan 25;15(1):3226. doi: 10.1038/s41598-025-87966-w.

ABSTRACT

Exploring the potential of advanced artificial intelligence technology in predicting microsatellite instability (MSI) and Ki-67 expression of endometrial cancer (EC) is highly significant. This study aimed to develop a novel hybrid radiomics approach integrating multiparametric magnetic resonance imaging (MRI), deep learning, and multichannel image analysis for predicting MSI and Ki-67 status. A retrospective study included 156 EC patients who were subsequently categorized into MSI and Ki-67 groups. The hybrid radiomics model (HMRadSum) was developed by extracting quantitative imaging features and deep learning features from multiparametric MRI using emerging attention mechanism. Tumor markers were subsequently predicted utilizing an XGBoost classifier. Model performance and interpretability were evaluated using standard classification metrics, Gradient-weighted Class Activation Mapping (Grad-CAM), and SHapley Additive exPlanations (SHAP) techniques. For the MSI prediction task, the HMRadSum model achieved area-under-curve (AUC) value of 0.945 (95% CI 0.862-1.000) and accuracy of 0.889. For the Ki-67 prediction task, the AUC and accuracy of HMRadSum model was 0.888 (95% CI 0.743-1.000) and 0.810. This hybrid radiomics model effectively extracted features associated with EC gene expression, providing potential clinical implications for personalized diagnosis, treatment, and treatment strategy optimization.

PMID:39863695 | DOI:10.1038/s41598-025-87966-w

Sci Rep. 2025 Jan 25;15(1):3204. doi: 10.1038/s41598-025-87450-5.

ABSTRACT

In order to solve the limitations of flipped classroom in personalized teaching and interactive effect improvement, this paper designs a new model of flipped classroom in colleges and universities based on Virtual Reality (VR) by combining the algorithm of Contrastive Language-Image Pre-Training (CLIP). Through cross-modal data fusion, the model deeply combines students' operation behavior with teaching content, and improves teaching effect through intelligent feedback mechanism. The test data shows that the similarity between video and image modes reaches 0.89, which indicates that different modal information can be effectively integrated to ensure the semantic consistency and intuitive understanding of teaching content. The minimum Kullback-Leibler (KL) divergence is 0.12, which ensures the stability of data distribution and avoids information loss. The accuracy of automatically generating feedback reaches 93.72%, which significantly improves the efficiency of personalized learning guidance. In the adaptability test of virtual scene, the frequency of scene adjustment is 2.5 times/minute, and the consistency score is stable above 8.6, ensuring the consistency of teaching goals under complex interaction. This paper aims to enhance personalized learning experience, improve teaching efficiency and autonomous learning effect through VR technology and intelligent feedback, and promote the innovation of interactive teaching mode.

PMID:39863690 | DOI:10.1038/s41598-025-87450-5

NPJ Precis Oncol. 2025 Jan 26;9(1):27. doi: 10.1038/s41698-025-00808-w.

ABSTRACT

Patients with High-Grade Serous Ovarian Cancer (HGSOC) exhibit varied responses to treatment, with 20-30% showing de novo resistance to platinum-based chemotherapy. While hematoxylin-eosin (H&E)-stained pathological slides are used for routine diagnosis of cancer type, they may also contain diagnostically useful information about treatment response. Our study demonstrates that combining H&E-stained whole slide images (WSIs) with proteomic signatures using a multimodal deep learning framework significantly improves the prediction of platinum response in both discovery and validation cohorts. This method outperforms the Homologous Recombination Deficiency (HRD) score in predicting platinum response and overall patient survival. Our study suggests that histology and proteomics contain complementary information about biological processes determining response to first line platinum treatment in HGSOC. This integrative approach has the potential to improve personalized treatment and provide insights into the therapeutic vulnerabilities of HGSOC.

PMID:39863682 | DOI:10.1038/s41698-025-00808-w

Sci Data. 2025 Jan 25;12(1):149. doi: 10.1038/s41597-025-04447-5.

ABSTRACT

With climate extremes' rising frequency and intensity, robust analytical tools are crucial to predict their impacts on terrestrial ecosystems. Machine learning techniques show promise but require well-structured, high-quality, and curated analysis-ready datasets. Earth observation datasets comprehensively monitor ecosystem dynamics and responses to climatic extremes, yet the data complexity can challenge the effectiveness of machine learning models. Despite recent progress in deep learning to ecosystem monitoring, there is a need for datasets specifically designed to analyse compound heatwave and drought extreme impact. Here, we introduce the DeepExtremeCubes database, tailored to map around these extremes, focusing on persistent natural vegetation. It comprises over 40,000 globally sampled small data cubes (i.e. minicubes), with a spatial coverage of 2.5 by 2.5 km. Each minicube includes (i) Sentinel-2 L2A images, (ii) ERA5-Land variables and generated extreme event cube covering 2016 to 2022, and (iii) ancillary land cover and topography maps. The paper aims to (1) streamline data accessibility, structuring, pre-processing, and enhance scientific reproducibility, and (2) facilitate biosphere dynamics forecasting in response to compound extremes.

PMID:39863624 | DOI:10.1038/s41597-025-04447-5

Nat Commun. 2025 Jan 25;16(1):1031. doi: 10.1038/s41467-025-56412-w.

ABSTRACT

Compute-in-memory based on resistive random-access memory has emerged as a promising technology for accelerating neural networks on edge devices. It can reduce frequent data transfers and improve energy efficiency. However, the nonvolatile nature of resistive memory raises concerns that stored weights can be easily extracted during computation. To address this challenge, we propose RePACK, a threefold data protection scheme that safeguards neural network input, weight, and structural information. It utilizes a bipartite-sort coding scheme to store data with a fully on-chip physical unclonable function. Experimental results demonstrate the effectiveness of increasing enumeration complexity to 5.77 × 1075 for a 128-column compute-in-memory core. We further implement and evaluate a RePACK computing system on a 40 nm resistive memory compute-in-memory chip. This work represents a step towards developing safe, robust, and efficient edge neural network accelerators. It potentially serves as the hardware infrastructure for edge devices in federated learning or other systems.

PMID:39863590 | DOI:10.1038/s41467-025-56412-w

Am J Orthod Dentofacial Orthop. 2025 Feb;167(2):242-249. doi: 10.1016/j.ajodo.2024.09.006.

ABSTRACT

INTRODUCTION: This study aimed to assess the precision of an open-source, clinician-trained, and user-friendly convolutional neural network-based model for automatically segmenting the mandible.

METHODS: A total of 55 cone-beam computed tomography scans that met the inclusion criteria were collected and divided into test and training groups. The MONAI (Medical Open Network for Artificial Intelligence) Label active learning tool extension was used to train the automatic model. To assess the model's performance, 15 cone-beam computed tomography scans from the test group were inputted into the model. The ground truth was obtained from manual segmentation data. Metrics including the Dice similarity coefficient, Hausdorff 95%, precision, recall, and segmentation times were calculated. In addition, surface deviations and volumetric differences between the automated and manual segmentation results were analyzed.

RESULTS: The automated model showed a high level of similarity to the manual segmentation results, with a mean Dice similarity coefficient of 0.926 ± 0.014. The Hausdorff distance was 1.358 ± 0.466 mm, whereas the mean recall and precision values were 0.941 ± 0.028 and 0.941 ± 0.022, respectively. There were no statistically significant differences in the arithmetic mean of the surface deviation for the entire mandible and 11 different anatomic regions. In terms of volumetric comparisons, the difference between the 2 groups was 1.62 mm³, which was not statistically significant.

CONCLUSIONS: The automated model was found to be suitable for clinical use, demonstrating a high degree of agreement with the reference manual method. Clinicians can use open-source software to develop custom automated segmentation models tailored to their specific needs.

PMID:39863342 | DOI:10.1016/j.ajodo.2024.09.006

Health Informatics J. 2025 Jan-Mar;31(1):14604582251315594. doi: 10.1177/14604582251315594.

ABSTRACT

Purpose: In the context of Chinese clinical texts, this paper aims to propose a deep learning algorithm based on Bidirectional Encoder Representation from Transformers (BERT) to identify privacy information and to verify the feasibility of our method for privacy protection in the Chinese clinical context. Methods: We collected and double-annotated 33,017 discharge summaries from 151 medical institutions on a municipal regional health information platform, developed a BERT-based Bidirectional Long Short-Term Memory Model (BiLSTM) and Conditional Random Field (CRF) model, and tested the performance of privacy identification on the dataset. To explore the performance of different substructures of the neural network, we created five additional baseline models and evaluated the impact of different models on performance. Results: Based on the annotated data, the BERT model pre-trained with the medical corpus showed a significant performance improvement to the BiLSTM-CRF model with a micro-recall of 0.979 and an F1 value of 0.976, which indicates that the model has promising performance in identifying private information in Chinese clinical texts. Conclusions: The BERT-based BiLSTM-CRF model excels in identifying privacy information in Chinese clinical texts, and the application of this model is very effective in protecting patient privacy and facilitating data sharing.

PMID:39862116 | DOI:10.1177/14604582251315594

Viruses. 2024 Dec 25;17(1):14. doi: 10.3390/v17010014.

ABSTRACT

Human respiratory syncytial virus (RSV) remains a significant global health threat, particularly for vulnerable populations. Despite extensive research, effective antiviral therapies are still limited. To address this urgent need, we present AVP-GPT2, a deep-learning model that significantly outperforms its predecessor, AVP-GPT, in designing and screening antiviral peptides. Trained on a significantly expanded dataset, AVP-GPT2 employs a transformer-based architecture to generate diverse peptide sequences. A multi-modal screening approach, incorporating Star-Transformer and Vision Transformer, enables accurate prediction of antiviral activity and toxicity, leading to the identification of potent and safe candidates. SHAP analysis further enhances interpretability by explaining the underlying mechanisms of peptide activity. Our in vitro experiments confirmed the antiviral efficacy of peptides generated by AVP-GPT2, with some exhibiting EC50 values as low as 0.01 μM and CC50 values > 30 μM. This represents a substantial improvement over AVP-GPT and traditional methods. AVP-GPT2 has the potential to significantly impact antiviral drug discovery by accelerating the identification of novel therapeutic agents. Future research will explore its application to other viral targets and its integration into existing drug development pipelines.

PMID:39861804 | DOI:10.3390/v17010014

Sensors (Basel). 2025 Jan 20;25(2):589. doi: 10.3390/s25020589.

ABSTRACT

With the rapid development of AI algorithms and computational power, object recognition based on deep learning frameworks has become a major research direction in computer vision. UAVs equipped with object detection systems are increasingly used in fields like smart transportation, disaster warning, and emergency rescue. However, due to factors such as the environment, lighting, altitude, and angle, UAV images face challenges like small object sizes, high object density, and significant background interference, making object detection tasks difficult. To address these issues, we use YOLOv8s as the basic framework and introduce a multi-level feature fusion algorithm. Additionally, we design an attention mechanism that links distant pixels to improve small object feature extraction. To address missed detections and inaccurate localization, we replace the detection head with a dynamic head, allowing the model to route objects to the appropriate head for final output. We also introduce Slideloss to improve the model's learning of difficult samples and ShapeIoU to better account for the shape and scale of bounding boxes. Experiments on datasets like VisDrone2019 show that our method improves accuracy by nearly 10% and recall by about 11% compared to the baseline. Additionally, on the AI-TODv1.5 dataset, our method improves the mAP50 from 38.8 to 45.2.

PMID:39860960 | DOI:10.3390/s25020589

Sensors (Basel). 2025 Jan 20;25(2):580. doi: 10.3390/s25020580.

ABSTRACT

The Internet of Things (IoT) has emerged as a crucial element in everyday life. The IoT environment is currently facing significant security concerns due to the numerous problems related to its architecture and supporting technology. In order to guarantee the complete security of the IoT, it is important to deal with these challenges. This study centers on employing deep learning methodologies to detect attacks. In general, this research aims to improve the performance of existing deep learning models. To mitigate data imbalances and enhance learning outcomes, the synthetic minority over-sampling technique (SMOTE) is employed. Our approach contributes to a multistage feature extraction process where autoencoders (AEs) are used initially to extract robust features from unstructured data on the model architecture's left side. Following this, long short-term memory (LSTM) networks on the right analyze these features to recognize temporal patterns indicative of abnormal behavior. The extracted and temporally refined features are inputted into convolutional neural networks (CNNs) for final classification. This structured arrangement harnesses the distinct capabilities of each model to process and classify IoT security data effectively. Our framework is specifically designed to address various attacks, including denial of service (DoS) and Mirai attacks, which are particularly harmful to IoT systems. Unlike conventional intrusion detection systems (IDSs) that may employ a singular model or simple feature extraction methods, our multistage approach provides more comprehensive analysis and utilization of data, enhancing detection capabilities and accuracy in identifying complex cyber threats in IoT environments. This research highlights the potential benefits that can be gained by applying deep learning methods to improve the effectiveness of IDSs in IoT security. The results obtained indicate a potential improvement for enhancing security measures and mitigating emerging threats.

PMID:39860948 | DOI:10.3390/s25020580

Sensors (Basel). 2025 Jan 19;25(2):561. doi: 10.3390/s25020561.

ABSTRACT

Real-time and accurate traffic forecasting aids in traffic planning and design and helps to alleviate congestion. Addressing the negative impacts of partial data loss in traffic forecasting, and the challenge of simultaneously capturing short-term fluctuations and long-term trends, this paper presents a traffic forecasting model, D-MGDCN-CLSTM, based on Multi-Graph Gated Dilated Convolution and Conv-LSTM. The model uses the DTWN algorithm to fill in missing data. To better capture the dual characteristics of short-term fluctuations and long-term trends in traffic, the model employs the DWT for multi-scale decomposition to obtain approximation and detail coefficients. The Conv-LSTM processes the approximation coefficients to capture the long-term characteristics of the time series, while the multiple layers of the MGDCN process the detail coefficients to capture short-term fluctuations. The outputs of the two branches are then merged to produce the forecast results. The model is tested against 10 algorithms using the PeMSD7(M) and PeMSD7(L) datasets, improving MAE, RMSE, and ACC by an average of 1.38% and 13.89%, 1% and 1.24%, and 5.92% and 1%, respectively. Ablation experiments, parameter impact analysis, and visual analysis all demonstrate the superiority of our decompositional approach in handling the dual characteristics of traffic data.

PMID:39860932 | DOI:10.3390/s25020561