Abstract Human pose estimation is a fundamental task in computer vision with widespread applications in human–computer interaction, sports analytics, and healthcare. While convolutional neural networks (CNNs) and Transformers have achieved notable success, they often struggle to capture structured body relationships, handle occlusions, and generalize effectively across diverse environments. Graph Neural Networks (GNNs), which represent human poses as structured graphs, offer a compelling alternative by explicitly modeling spatial and temporal dependencies among body joints. This survey provides a comprehensive review of GNN-based pose estimation approaches, encompassing spatial GCNs, spatiotemporal models, graph–Transformer hybrids, and hypergraph frameworks. We analyze these methods along key dimensions, including graph construction, learning paradigms, attention mechanisms, and computational efficiency, using standard benchmarks such as Human3.6 M, COCO, and MPI-INF-3DHP. Our review identifies several emerging trends and critical limitations. These include high computational cost, limited generalization to unconstrained scenarios, and inconsistent evaluation protocols. To advance the field, we outline future research directions, such as hybrid GNN–Transformer architectures, lightweight models for edge deployment, multi-modal fusion, and self-supervised learning strategies aimed at reducing annotation dependency and improving cross-domain robustness.

During experimentation, it is found that the EfficientNetB3 model outperforms in accuracy, loss, as well as root mean square error by obtaining 99.96%, 0.149, and 0.386 respectively. However, other models also show the good performance in terms of precision, recall, and F1 score, with high scores close to 0.98 or 1.00, except for VGG19. The findings of the paper emphasize the prospective of deep transfer learning as a viable technique for cotton plant disease diagnosis by providing a cost-effective and efficient solution for crop disease monitoring and management. This strategy can also help to improve agricultural practices by ensuring sustainable cotton farming and increased crop output.

The human body is designed to experience stress and react to it, and experiencing challenges causes our body to produce physical and mental responses and also helps our body to adjust to new situations. However, stress becomes a problem when it continues to remain without a period of relaxation or relief. When a person has long-term stress, continued activation of the stress response causes wear and tear on the body. Chronic stress results in cancer, cardiovascular disease, depression, and diabetes, and thus is deeply detrimental to our health. Previous researchers have performed a lot of work regarding mental stress, using mainly machine-learning-based approaches. However, most of the methods have used raw, unprocessed data, which cause more errors and thereby affect the overall model performance. Moreover, corrupt data values are very common, especially for wearable sensor datasets, which may also lead to poor performance in this regard. This paper introduces a deep-learning-based method for mental stress detection by encoding time series raw data into Gramian Angular Field images, which results in promising accuracy while detecting the stress levels of an individual. The experiment has been conducted on two standard benchmark datasets, namely WESAD (wearable stress and affect detection) and SWELL. During the studies, testing accuracies of 94.8% and 99.39% are achieved for the WESAD and SWELL datasets, respectively. For the WESAD dataset, chest data are taken for the experiment, including the data of sensor modalities such as three-axis acceleration (ACC), electrocardiogram (ECG), body temperature (TEMP), respiration (RESP), etc.

Path planning is a fundamental component in robotics and game artificial intelligence that considerably influences the motion efficiency of robots and unmanned aerial vehicles, as well as the realism and immersion of virtual environments. However, traditional algorithms are often limited to single-objective optimization and lack real-time adaptability to dynamic environments. This study addresses these limitations through a proposed real-time dynamic multiobjective (RDMO) path-planning algorithm based on an enhanced A* framework. The proposed algorithm employs a queue-based structure and composite multiheuristic functions to dynamically manage game tasks and compute optimal paths under changing-map-connectivity conditions in real time. Simulation experiments are conducted using real-world road network data and benchmarked against mainstream hybrid approaches based on genetic algorithms (GAs) and simulated annealing (SA). The results show that the computational speed of the RDMO algorithm is 88 and 73 times faster than that of the GA- and SA-based solutions, respectively, while the total planned path length is reduced by 58% and 33%, respectively. In addition, the RDMO algorithm also shows excellent responsiveness to dynamic changes in map connectivity and can achieve real-time replanning with a minimal computational overhead. The research results prove that the RDMO algorithm provides a robust and efficient solution for multiobjective path planning in games and robotics applications and has a great application potential in improving system performance and user experience in related fields in the future.

The proliferation of AI-generated medical deepfakes, such as tumor insertions or removals in diagnostic scans, threatens patient safety and healthcare integrity. Existing detection methods often lack robustness against adversarial attacks or fail to integrate multimodal feature representations. To address these gaps, we propose AFFETDS (Adversarial Feature Fusion Enhanced Tumor Detection System), a novel ensemble framework combining adversarial training, feature fusion, and weighted voting. AFFETDS leverages adversarial attack methods (PGD, FGSM) to harden the model, fuses high-level ResNet50 features with handcrafted HOG descriptors, and employs an SVM-based ensemble classifier. Evaluated on a curated dataset of 1378 MRI scans (774 real, 604 manipulated) from TCIA and ADNI repositories, AFFETDS achieves state-of-the-art performance with 91.5% accuracy, 90.7% precision, and 91.2% recall, outperforming baseline models (SVM: 86.2%, CNN: 88.4%). The framework's ROC-AUC (0.80) and calibrated confidence scores demonstrate superior generalization across diverse imaging conditions. The ability of combining the adversarial techniques with multimodal feature fusion, our proposed AFFETDS framework improves the detection of subtle tumor manipulations, presenting an important safeguard to maintain the authenticity of medical images. The findings of research work underscore the urgent need of proactive defenses against growing deepfake threats in healthcare applications.

Semantic segmentation is critical in remote sensing applications such as urban planning, disaster management, and environmental monitoring. However, segmenting complex satellite images remains challenging due to varied textures and overlapping boundaries. This study proposes a PSO-optimized ensemble framework integrating DeepLabV3+ and UNet to enhance segmentation precision and reliability. DeepLabV3+ captures global contextual information, while UNet preserves fine spatial details, enabling the ensemble to achieve superior segmentation outcomes. The model is trained using the Semantic Segmentation of Aerial Imagery dataset and optimized with Particle Swarm Optimization (PSO) for adaptive learning rate tuning, ensuring convergence stability and performance maximization. Experimental evaluations demonstrate the ensemble’s ability to outperform individual models, achieving a Dice score of 0.9201 compared to DeepLabV3+ (0.8538) and UNet (0.7877). Further, cross-dataset validation on the Bhuvan Satellite Data confirms its generalization capability with an unprecedented Dice score of 0.9999, indicating robust segmentation across diverse landscapes. These findings highlight the effectiveness of PSO-driven ensemble learning for high-precision satellite image analysis, offering a scalable solution for real-world geospatial applications. Our approach sets a new benchmark in segmentation accuracy and cross-domain adaptability, enhancing the practical utility of semantic segmentation in remote sensing tasks. In this context, PSO is introduced as a metaheuristic strategy to optimize the learning rate of the ensemble model adaptively. Unlike static or heuristically chosen learning rates, PSO dynamically explores the search space based on segmentation loss feedback, enabling improved convergence and reduced training instability. This adaptive mechanism is particularly valuable in cross-dataset semantic segmentation tasks, where variability in resolution, scene complexity, and geographic distribution demands robust generalization and precise learning rate adjustment.

The rapid evolution of virtual reality (VR) and augmented reality (AR) technologies has significantly transformed human-computer interaction, with applications spanning entertainment, education, healthcare, industry, and remote collaboration. A central challenge in these immersive systems lies in enabling intuitive, efficient, and natural interactions. Hand gesture recognition offers a compelling solution by leveraging the expressiveness of human hands to facilitate seamless control without relying on traditional input devices such as controllers or keyboards, which can limit immersion. However, achieving robust gesture recognition requires overcoming challenges related to accurate hand tracking, complex environmental conditions, and minimizing system latency. This study proposes an artificial intelligence (AI)-driven framework for recognizing both static and dynamic hand gestures in VR and AR environments using skeleton-based tracking compliant with the OpenXR standard. Our approach employs a lightweight neural network architecture capable of real-time classification within approximately 1.3 ms while maintaining average accuracy of 95%. We also introduce a novel dataset generation method to support training robust models and demonstrate consistent classification of diverse gestures across widespread commercial VR devices. This work represents one of the first studies to implement and validate dynamic hand gesture recognition in real time using standardized VR hardware, laying the groundwork for more immersive, accessible, and user-friendly interaction systems. By advancing AI-driven gesture interfaces, this research has the potential to broaden the adoption of VR and AR across diverse domains and enhance the overall user experience.

The evolution of educational environments has seen a shift from conventional classrooms to technology-enhanced smart classrooms, driven by the rapid advancement of digital tools. The integration of traditional art education and modern technologies lacks interactivity and personalized feedback, which limits student engagement and creative progression. The objective of this research is to assess how AI and AR can be combined to improve student engagement, creativity, academic performance, and aesthetic understanding in art education. Data were collected from smart classroom sessions involving educational videos and interactive AR applications focused on photography. The pre-processing stage automatically filters low-quality images, retaining those with high saliency and clarity scores to ensure meaningful input for analysis. Using a TensorFlow-based experimental framework, a Deep Recurrent Neural Network (DRNN) algorithm was employed for intelligent image synthesis and feedback, allowing real-time analysis of composition and augmented visual storytelling. Results indicated notable improvements in student, Accuracy (97.18%), precision (97.33%), recall (96.95%), F1 score (97%). Students responded positively to the immersive experience, showing increased appreciation for cultural and visual diversity. In conclusion, the study demonstrates that integrating AI and AR in smart classroom environments can redefine art education by fostering experiential learning and providing dynamic, student-centered educational opportunities.