Ahsan Shahzad

Effective interpretation of coherent and usable information from aerial images (e.g., satellite imagery or high-altitude drone photography) can greatly reduce human effort in many situations, both natural (e.g., earthquakes, forest fires, tsunamis) and man-made (e.g., highway pile-ups, traffic congestion), particularly in disaster management. This research proposes a novel encoder–decoder framework for captioning of remote sensing images that integrates self-supervised DINOv3 visual features with a hybrid Transformer–LSTM decoder. Unlike existing approaches that rely on supervised CNN-based encoders (e.g., ResNet, VGG), the proposed method leverages DINOv3’s self-supervised learning capabilities to extract dense, semantically rich features from aerial images without requiring domain-specific labeled pretraining. The proposed hybrid decoder combines Transformer layers for global context modeling with LSTM layers for sequential caption generation, producing coherent and context-aware descriptions. Feature extraction is performed using the DINOv3 model, which employs the gram-anchoring technique to stabilize dense feature maps. Captions are generated through a hybrid of Transformer with Long Short-Term Memory (LSTM) layers, which adds contextual meaning to captions through sequential hidden layer modeling with gated memory. The model is first evaluated on two traditional remote sensing image captioning datasets: RSICD and UCM-Captions. Multiple evaluation metrics like Bilingual Evaluation Understudy (BLEU), Consensus-based Image Description Evaluation (CIDEr), Recall-Oriented Understudy for Gisting Evaluation (ROUGE-L), and Metric for Evaluation of Translation with Explicit Ordering (METEOR), are used to quantify the performance and robustness of the proposed DINOv3 hybrid model. The proposed model outperforms conventional Convolutional Neural Network (CNN) and Vision Transformers (ViT)-based models by approximately 9–12% across most evaluation metrics. Attention heatmaps are also employed to qualitatively validate the proposed model when identifying and describing key spatial elements. In addition, the proposed model is evaluated on advanced remote sensing datasets, including RSITMD, DisasterM3, and GeoChat. The results demonstrate that self-supervised vision transformers are robust encoders for multi-modal understanding in remote sensing image analysis and captioning.

Early detection of Mild Cognitive Impairment (MCI), a prodromal stage of dementia, plays a pivotal role in enabling timely clinical intervention and slowing cognitive decline. This paper presents a multi-sensor balance assessment framework designed to identify MCI-related postural instabilities using a wearable inertial measurement unit (IMU) network. The proposed system employs five synchronized IMUs placed at the waist, thighs, and shanks to capture balance dynamics across four static balance tasks: Eyes-Open, Eyes-Closed, Right-Leg Lift, and Left-Leg Lift. A three-stage feature selection strategy, comprising variance and correlation pruning, univariate filtering, and embedded model selection, is implemented within a Leave-One-Subject-Out (LOSO) cross-validation scheme to extract discriminative sway features. Classification using Support Vector Machines and tree-based ensemble models consistently yields superior results, achieving accuracies between 71.7% and 79.2%, with the highest performance observed in the Eyes-Open condition. A compact 10-feature subset demonstrates stable and robust discriminative power across all tasks. Compared to a single-sensor baseline, the multi-sensor configuration provides improved classification performance, underscoring the feasibility of compact, balance-driven, non-invasive MCI screening through wearable sensor systems.

The increasing proportion of elderly individuals poses major healthcare challenges, particularly in maintaining independence and monitoring activities of daily living (ADLs). Smartphone (SP)-based human activity recognition (HAR) offers a practical and cost-effective solution, but its effectiveness is often constrained by assumptions of fixed device placement or location-agnostic modeling. In this work, we present a location-aware deep learning (DL) framework for HAR, tailored to the four most common SP carrying positions among elderly users: hand, thigh pocket, belly pouch, and shoulder bag. A novel dataset was collected from 14 elderly participants in South Korea, encompassing 15 ADLs and fall-like activities across these carrying positions. Raw inertial signals were transformed into spectrograms via short-time Fourier transform (STFT), enabling the application of advanced DL models, including convolutional neural network (CNN), EfficientNet-B0, and vision transformers (ViTs). Synthetic minority oversampling technique (SMOTE) was applied to mitigate class imbalance. Experimental results demonstrate that EfficientNet-B0 consistently achieves the highest classification performance, reaching 99% accuracy across multiple positions, and outperforming conventional architectures on both our dataset and the public HARTH dataset. The proposed framework highlights the importance of location-aware modeling and provides a scalable pathway toward robust, real-world HAR systems for elderly populations.

Dementia is characterized by a progressive loss of cognitive abilities, and diagnosing its early stages Mild Cognitive Impairment (MCI), is difficult since it is a transitory state that is different from total cognitive collapse. Recent clinical research studies have identified that balance impairments can be a significant indicator for predicting dementia in older adults. Accordingly, the current research focuses on finding innovative postural balance-based digital biomarkers by using wearable inertial sensors and pre-screening of MCI in home settings using machine learning techniques. For this research, sixty subjects (30 cognitively normal and 30 MCI) with waist-mounted inertial sensor performed balance tasks in four different standing postures: eyes-open, eyes-closed, right-leg-lift, and left-leg-lift. The significant balance biomarkers for MCI identification are discovered by our research, demonstrating specific characteristics in each of these four states. A robust feature selection approach is ensured by the multi-step methodology that combines the strengths of Filter techniques, Wrapper methods, and SHAP (Shapley Additive exPlanations) technique. The proposed balance biomarkers have the potential to detect MCI (with 75.8% accuracy), as evidenced by the results of machine learning algorithms for classification. This work adds to the growing body of literature targeted at enhancing understanding and proactive management of cognitive loss in older populations and lays the groundwork for future research efforts aimed at refining digital biomarkers, validating findings, and exploring longitudinal perspectives.

The mild symptoms in Mild Cognitive Impairment (MCI), a precursor of dementia, often go unnoticed and are assumed as normal aging signs. Such negligence result in late visits which consequently, lead to the diagnosis and progression of dementia. An instrumented gait assessment in home settings may facilitate the detection of subtle MCI-related motor deficits thus, allowing early diagnosis and intervention. This paper investigates potential gait biomarkers derived from shank mounted inertial sensors signals under normal and dual-task walking conditions using data collected from thirty MCI and thirty cognitively normal (CN) subjects. To identify potential gait biomarkers for MCI screening, we assess the variance and predictive power of each feature. Moreover, multiple classification models using different machine learning and feature selection techniques are built to automate MCI detection by leveraging the gait biomarkers. Statistical analysis reveal multiple gait parameters that are significantly different under both single and dual-task settings. However, we show that dual-task walking provides better distinction between MCI and CN subjects. The machine learning model employed for MCI pre-screening based on the inertial sensor-derived gait biomarkers achieves accuracy and sensitivity of 71.67% and 83.33%, respectively.

Traditional handcrafted crowd-counting techniques in an image are currently transformed via machine-learning and artificial-intelligence techniques into intelligent crowd-counting techniques. This paradigm shift offers many advanced features in terms of adaptive monitoring and the control of dynamic crowd gatherings. Adaptive monitoring, identification/recognition, and the management of diverse crowd gatherings can improve many crowd-management-related tasks in terms of efficiency, capacity, reliability, and safety. Despite many challenges, such as occlusion, clutter, and irregular object distribution and nonuniform object scale, convolutional neural networks are a promising technology for intelligent image crowd counting and analysis. In this article, we review, categorize, analyze (limitations and distinctive features), and provide a detailed performance evaluation of the latest convolutional-neural-network-based crowd-counting techniques. We also highlight the potential applications of convolutional-neural-network-based crowd-counting techniques. Finally, we conclude this article by presenting our key observations, providing strong foundation for future research directions while designing convolutional-neural-network-based crowd-counting techniques. Further, the article discusses new advancements toward understanding crowd counting in smart cities using the Internet of Things (IoT).

Common fall occurrences in the elderly population pose dramatic challenges in public healthcare domain. Adoption of an efficient and yet highly reliable automatic fall detection system may not only mitigate the adverse effects of falls through immediate medical assistance, but also profoundly improve the functional ability and confidence level of elder people. This paper presents a pervasive fall detection system developed on smart phones, namely, FallDroid that exploits a two-step algorithm proposed to monitor and detect fall events using the embedded accelerometer signals. Comprising of the threshold-based method and multiple kernel learning support vector machine, the proposed algorithm uses novel techniques to effectively identify fall-like events (such as lying on a bed or sudden stop after running) and reduce false alarms. In addition to user convenience and low power consumption, experimental results reveal that the system detects falls with high accuracy (97.8% and 91.7%), sensitivity (99.5% and 95.8%), and specificity (95.2% and 88.0%) when placed around the waist and thigh, respectively. The system also achieves the lowest false alarm rate of 1 alarm per 59 h of usage, which is best till date.