This paper introduces a new approach to emotion classification utilising deep learning models, specifically the Vision Transformer (ViT) model, in the analysis of electroencephalogram (EEG) signals. A dual-feature extraction approach was implemented in our study, utilising Power Spectral Density and Differential Entropy, to analyse the SEED IV dataset. This methodology resulted in the detailed classification of four distinct emotional states. The ViT model, which was originally designed for image processing, has been successfully applied to EEG signal analysis. It demonstrated remarkable performance by attaining a test accuracy of 99.02% with little variance. Notably, it outperformed conventional models like GRUs, LSTMs, and CNNs in this context. The findings of our study indicate that the ViT model has a high level of effectiveness in accurately identifying complex patterns present in EEG data. Specifically, the precision and recall rates achieved by the model surpass 98%, while the F1 score is estimated to be about 98.9%. The results of this study not only demonstrate the efficacy of transformer-based models in analysing cognitive states, but also indicate their considerable potential in improving systems for sympathetic human-computer interaction.

Motor imagery (MI) classification using EEG signals is central to brain–computer interfaces but remains challenging due to low signal-to-noise ratio, non-stationarity, and high inter-subject variability. We introduce EEG-TriNet++, a multi-branch deep learning architecture that enhances both classification accuracy and cross-subject generalization. The model integrates three complementary components: convolutional spatial–spectral encoders for channel-wise and frequency-specific patterns, bidirectional LSTMs to model temporal dynamics, and a Transformer head for global relational reasoning. A patchwise tokenization strategy and neural architecture search optimize the trade-off between efficiency and representational capacity. To address individual differences, a model-agnostic meta-learning (MAML) module enables rapid adaptation to new users with limited data. Evaluated on two public MI datasets under within-subject and leave-one-subject-out (LOSO) protocols, EEG-TriNet++ achieves 79.1% and 78.6% accuracy in within-subject tasks, and 72.4% and 71.3% in LOSO settings. Ablation studies validate the contribution of each module, and comparisons with state-of-the-art methods demonstrate consistent performance gains under identical conditions.

Accurate crack segmentation in civil infrastructure imagery remains challenging because of the prevalence of thin, low-contrast, and spatially discontinuous defects that often appear amid textured surfaces, shadows, and acquisition noise. Although Transformer-based models improve global context modeling, many existing solutions incur substantial computational and memory overhead, which limits their use in practical, resource-constrained inspection settings. In this work, we introduce an efficient hybrid segmentation architecture that combines a convolutional encoder for high-fidelity local representation with a lightweight Transformer bottleneck for global dependency modeling, followed by a progressive decoder that restores spatial resolution through multi-level skip-feature fusion. To better accommodate severe foreground sparsity and preserve fine crack structures, the framework is trained with a composite Dice–Binary Cross-Entropy objective and employs a tokenization strategy designed to preserve fine spatial details while enabling efficient global context modeling. We validate the proposed approach on four public benchmarks, demonstrating consistent improvements over representative convolutional, Transformer-based, and hybrid baselines, while ablation studies confirm the contribution of each design component. Finally, runtime profiling shows favorable latency and memory characteristics, supporting real-time or near real-time deployment on embedded and edge inspection platforms.

Accurate detection of cardiac arrhythmias from electrocardiograms remains challenging for rare rhythm classes due to class imbalance and morphological variability. We present a hybrid deep learning framework combining per-lead convolutional encoders with a cross-lead transformer that models relationships across different lead signals through self-attention, accepting variable lead configurations. To address minority-class scarcity, a generative adversarial network synthesizes physiologically plausible beat segments for underrepresented arrhythmias. Attention-based visualizations localize influential waveform regions aligned with clinically meaningful structures. Post-training pruning and INT8 quantization enable efficient deployment with minimal performance loss. Extensive experiments on the MIT-BIH Arrhythmia Database across sixteen heartbeat classes from two-lead recordings yield exceptional results over ten independent runs: accuracy of 99.67%, F1-score of 99.66%, and AUC of 99.8%. External validation on the ECG5000 single-lead dataset and the St Petersburg INCART twelve-lead dataset confirms robust generalizability with F1-scores of 97.6% and 98% respectively. Our framework delivers accurate, interpretable, stable, and deployable arrhythmia detection across diverse clinical settings.

One of the most important problems in computer vision and remote sensing is object detection, which identifies particular categories of diverse things in pictures. Two crucial data sources for public security are the thermal infrared (TIR) remote sensing multi-scenario photos and videos produced by unmanned aerial vehicles (UAVs). Due to the small scale of the target, complex scene information, low resolution relative to the viewable videos, and dearth of publicly available labeled datasets and training models, their object detection procedure is still difficult. A UAV TIR object detection framework for pictures and videos is suggested in this study. The Forward-looking Infrared (FLIR) cameras used to gather ground-based TIR photos and videos are used to create the ``You Only Look Once'' (YOLO) model, which is based on CNN architecture. Results indicated that in the validating task, detecting human object had an average precision at IOU (Intersection over Union) = 0.5, which was 72.5\%, using YOLOv7 (YOLO version 7) state of the art model \cite{1}, while the detection speed around 161 frames per second (FPS/second). The usefulness of the YOLO architecture is demonstrated in the application, which evaluates the cross-detection performance of people in UAV TIR videos under a YOLOv7 model in terms of the various UAVs' observation angles. The qualitative and quantitative evaluation of object detection from TIR pictures and videos using deep-learning models is supported favorably by this work.