Multimodal Driver Condition Monitoring System Operating in the Far-Infrared Spectrum

Electronics, Год журнала: 2024, Номер 13(17), С. 3502 - 3502

Опубликована: Сен. 3, 2024

Monitoring the psychophysical conditions of drivers is crucial for ensuring road safety. However, achieving real-time monitoring within a vehicle presents significant challenges due to factors such as varying lighting conditions, vibrations, limited computational resources, data privacy concerns, and inherent variability in driver behavior. Analyzing states using visible spectrum imaging particularly challenging under low-light at night. Additionally, relying on single behavioral indicator often fails provide comprehensive assessment driver’s condition. To address these challenges, we propose system that operates exclusively far-infrared spectrum, enabling detection critical features yawning, head drooping, pose estimation regardless scenario. It integrates channel fusion module assess state more accurately underpinned by our custom-developed annotated datasets, along with modified deep neural network designed facial feature thermal spectrum. Furthermore, introduce two modules synthesizing events into coherent state: one based simple machine another combines modality encoder large language model. This latter approach allows generation responses queries beyond system’s explicit training. Experimental evaluations demonstrate high accuracy detecting responding signs fatigue distraction.

Язык: Английский

---

A Noisy Sample Selection Framework Based on a Mixup Loss and Recalibration Strategy

Mathematics, Год журнала: 2024, Номер 12(15), С. 2389 - 2389

Опубликована: Июль 31, 2024

Deep neural networks (DNNs) have achieved breakthrough progress in various fields, largely owing to the support of large-scale datasets with manually annotated labels. However, obtaining such is costly and time-consuming, making high-quality annotation a challenging task. In this work, we propose an improved noisy sample selection method, termed “sample framework”, based on mixup loss recalibration strategy (SMR). This framework enhances robustness generalization abilities models. First, introduce robust function pre-train two models identical structures separately. approach avoids additional hyperparameter adjustments reduces need for prior knowledge noise types. Additionally, use Gaussian Mixture Model (GMM) divide entire training set into labeled unlabeled subsets, followed by using semi-supervised learning (SSL) techniques. Furthermore, cross-entropy (CE) prevent from converging local optima during SSL process, thus further improving performance. Ablation experiments CIFAR-10 50% symmetric 40% asymmetric demonstrate that modules introduced paper improve accuracy baseline (i.e., DivideMix) 1.5% 0.5%, respectively. Moreover, experimental results multiple benchmark our proposed method effectively mitigates impact labels significantly performance DNNs datasets. For instance, WebVision dataset, improves top-1 0.7% 2.4% compared method.

Язык: Английский

---

Fractional Adaptive Resonance Theory (FRA-ART): An Extension for a Stream Clustering Method with Enhanced Data Representation

Mathematics, Год журнала: 2024, Номер 12(13), С. 2049 - 2049

Опубликована: Июнь 30, 2024

Clustering data streams has become a hot topic and been extensively applied to many real-world applications. Compared with traditional clustering, stream clustering is more challenging. Adaptive Resonance Theory (ART) powerful (online) method, it can automatically adjust learn both abstract concrete information, respond arbitrarily large non-stationary databases while having fewer parameters, low computational complexity, less sensitivity noise, but its limited feature representation hinders application complex streams. In this paper, considering advantages disadvantages, we present flexible extension for called fractional adaptive resonance theory (FRA-ART). FRA-ART enhances by fractionally exponentiating input features using self-interactive basis functions (SIBFs) incorporating interaction through cross-interactive (CIBFs) at the cost only of introducing an additionally adjustable order. Both SIBFs CIBFs be precomputed existing algorithms, making easily adaptable any ART variant. Finally, comparative experiments on five datasets, including artificial demonstrate FRA-ART’s superior robustness comparable or improved performance in terms accuracy, normalized mutual rand index, cluster stability compared state-of-the-art G-Stream algorithm.

Язык: Английский

---

Research on Default Classification of Unbalanced Credit Data Based on PixelCNN-WGAN

Electronics, Год журнала: 2024, Номер 13(17), С. 3419 - 3419

Опубликована: Авг. 28, 2024

Personal credit assessment plays a crucial role in the financial system, which not only relates to activities of individuals but also affects overall system and economic health society. However, current problem data imbalance affecting classification results field personal has been fully solved. In order solve this better, we propose data-enhanced algorithm based on Pixel Convolutional Neural Network (PixelCNN) Generative Adversarial (Wasserstein GAN, WGAN). Firstly, historical containing borrowers’ borrowing information are transformed into grayscale maps; then, enhancement default images is performed using improved PixelCNN-WGAN model; finally, expanded image dataset inputted CNN, AlexNet, SqueezeNet, MobileNetV2 for classification. The real LendingClub show that designed paper improves accuracy four algorithms by 1.548–3.568% compared with original dataset, can effectively improve effect data, certain extent, it provides new idea task assessment.

Язык: Английский

---

Improved Generalized-Pinball-Loss-Based Laplacian Twin Support Vector Machine for Data Classification

Symmetry, Год журнала: 2024, Номер 16(10), С. 1373 - 1373

Опубликована: Окт. 15, 2024

Nowadays, unlabeled data are abundant, while supervised learning struggles with this challenge as it relies solely on labeled data, which costly and time-consuming to acquire. Additionally, real-world often suffer from label noise, degrades the performance of models. Semi-supervised addresses these issues by using both data. This study extends twin support vector machine generalized pinball loss function (GPin-TSVM) into a semi-supervised framework incorporating graph-based methods. The assumption is that connected points should share similar labels, mechanisms handle noisy labels. Laplacian regularization ensures uniform information spread across graph, promoting balanced assignment. By leveraging term, two quadratic programming problems formulated, resulting in LapGPin-TSVM. Our proposed model reduces impact noise improves classification accuracy. Experimental results UCI benchmarks image demonstrate its effectiveness. Furthermore, addition accuracy, also measured Matthews Correlation Coefficient (MCC) score, experiments analyzed through statistical

Язык: Английский

---

Deep Learning-Driven Virtual Furniture Replacement Using GANs and Spatial Transformer Networks

Mathematics, Год журнала: 2024, Номер 12(22), С. 3513 - 3513

Опубликована: Ноя. 11, 2024

This study proposes a Generative Adversarial Network (GAN)-based method for virtual furniture replacement within indoor scenes. The proposed addresses the challenge of accurately positioning new in an space by combining image reconstruction with geometric matching through spatial transformer networks and GANs. system leverages deep learning architectures like Mask R-CNN executing segmentation generating masks, it employs DeepLabv3+, EdgeConnect algorithms, ST-GAN carrying out replacement. With system, shoppers can obtain shopping experience, providing easier way to understand aesthetic effects rearrangement without putting effort physically move furniture. has practical applications furnishing industry interior design practices, cost-effective efficient alternative physical results indicate that achieves accurate scenes minimal distortion or displacement. is limited 2D front-view images Future work would involve synthesizing 3D expanding replace photographed from different angles. enhance efficiency practicality

Язык: Английский

---

Mix-layers semantic extraction and multi-scale aggregation transformer for semantic segmentation

Complex & Intelligent Systems, Год журнала: 2024, Номер 11(1)

Опубликована: Ноя. 14, 2024

Recently, a number of vision transformer models for semantic segmentation have been proposed, with the majority these achieving impressive results. However, they lack ability to exploit intrinsic position and channel features image are less capable multi-scale feature fusion. This paper presents method that successfully combines attention multiscale representation, thereby enhancing performance efficiency. represents significant advancement in field. Multi-layers extraction aggregation decoder (MEMAFormer) is which consists two components: mix-layers dual module (MDCE) pyramid pooling (SAPPM). The MDCE incorporates multi-layers cross (MCAM) an efficient (ECAM). In MCAM, horizontal connections between encoder stages employed as queries module. hierarchical maps derived from different integrated into key value. To address long-term dependencies, ECAM selectively emphasizes interdependent by integrating relevant across all channels. adaptability reduced pooling, reduces amount computation without compromising performance. SAPPM comprised several distinct pooled kernels extract context deeper flow information, forming various sizes. MEMAFormer-B0 model demonstrates superior compared SegFormer-B0, exhibiting gains 4.8%, 4.0% 3.5% on ADE20K, Cityscapes COCO-stuff datasets, respectively.

Язык: Английский

---

An improved sample selection framework for learning with noisy labels

PLoS ONE, Год журнала: 2024, Номер 19(12), С. e0309841 - e0309841

Опубликована: Дек. 5, 2024

Deep neural networks have powerful memory capabilities, yet they frequently suffer from overfitting to noisy labels, leading a decline in classification and generalization performance. To address this issue, sample selection methods that filter out potentially clean labels been proposed. However, there is significant gap size between the filtered, possibly subset unlabeled subset, which becomes particularly pronounced at high-noise rates. Consequently, results underutilizing label-free samples methods, leaving room for performance improvement. This study introduces an enhanced framework with oversampling strategy (SOS) overcome limitation. leverages valuable information contained instances enhance model by combining SOS state-of-the-art methods. We validate effectiveness of through extensive experiments conducted on both synthetic datasets real-world such as CIFAR, WebVision, Clothing1M. The source code will be made available https://github.com/LanXiaoPang613/SOS.

Язык: Английский

---

Semi-supervised learning-based virtual adversarial training on graph for molecular property prediction

Alexandria Engineering Journal, Год журнала: 2024, Номер unknown

Опубликована: Дек. 1, 2024

Язык: Английский

---