Claim Missing Document
Check
Articles

Found 3 Documents
Search

Edge-Deployable Multi-Class Surveillance Video Anomaly Classification Using a MobileNetV2–LSTM Spatial–Temporal Framework S Aasha Nandhini; R Karthick Manoj; Vasantakumaren S R
Journal of Applied Data Sciences Vol 7, No 3: September 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i3.1471

Abstract

Automated surveillance systems require models that can recognize abnormal events accurately while maintaining efficient inference on resource-constrained hardware. This study proposes a lightweight spatial–temporal classification framework that integrates MobileNetV2 and Long Short-Term Memory networks for identifying five surveillance activity classes: Normal, Fighting, Burglary, Road Accident, and Explosion. Surveillance videos were preprocessed through frame sampling, resizing, normalization, and fixed-length sequence construction. MobileNetV2 was employed to extract compact frame-level spatial features, while the LSTM modeled temporal dependencies across consecutive frames. The framework was evaluated using accuracy, class-wise precision, recall, F1-score, macro F1-score, weighted F1-score, confusion-matrix analysis, and probability-based anomaly scoring. Experimental results on 463 video sequences show that the proposed model achieved an accuracy of 97.41%, a macro F1-score of 0.9612, and a weighted F1-score of 0.9744. Most classification errors occurred between visually similar human-centered anomaly classes, particularly Fighting and Burglary, while Normal and Explosion sequences were recognized with very high reliability. The trained model was converted into TensorFlow Lite format and deployed on a Raspberry Pi 4, achieving approximately 6–8 frames per second with an average latency of 120–150 ms per frame. These findings indicate that the proposed MobileNetV2–LSTM framework provides an effective balance between multi-class event recognition, temporal representation learning, and near-real-time edge inference. However, further validation on larger and more diverse surveillance datasets is required to confirm its generalizability under unconstrained operational conditions.
Edge-Deployable Multi-Class Surveillance Video Anomaly Classification Using a MobileNetV2–LSTM Spatial–Temporal Framework S Aasha Nandhini; R Karthick Manoj; Vasantakumaren S R
Journal of Applied Data Sciences Vol 7, No 3: September 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i3.1471

Abstract

Automated surveillance systems require models that can recognize abnormal events accurately while maintaining efficient inference on resource-constrained hardware. This study proposes a lightweight spatial–temporal classification framework that integrates MobileNetV2 and Long Short-Term Memory networks for identifying five surveillance activity classes: Normal, Fighting, Burglary, Road Accident, and Explosion. Surveillance videos were preprocessed through frame sampling, resizing, normalization, and fixed-length sequence construction. MobileNetV2 was employed to extract compact frame-level spatial features, while the LSTM modeled temporal dependencies across consecutive frames. The framework was evaluated using accuracy, class-wise precision, recall, F1-score, macro F1-score, weighted F1-score, confusion-matrix analysis, and probability-based anomaly scoring. Experimental results on 463 video sequences show that the proposed model achieved an accuracy of 97.41%, a macro F1-score of 0.9612, and a weighted F1-score of 0.9744. Most classification errors occurred between visually similar human-centered anomaly classes, particularly Fighting and Burglary, while Normal and Explosion sequences were recognized with very high reliability. The trained model was converted into TensorFlow Lite format and deployed on a Raspberry Pi 4, achieving approximately 6–8 frames per second with an average latency of 120–150 ms per frame. These findings indicate that the proposed MobileNetV2–LSTM framework provides an effective balance between multi-class event recognition, temporal representation learning, and near-real-time edge inference. However, further validation on larger and more diverse surveillance datasets is required to confirm its generalizability under unconstrained operational conditions.
Digital Twin–Driven AI Framework for Predictive Maintenance of Marine Diesel Engines Using Multi-Sensor IoT Data R Karthick Manoj; Gokulnath Anandakumar; Aasha Nandhini S; Vasantakumaren S R
Journal of Applied Data Sciences Vol 7, No 4: December 2026
Publisher : Bright Publisher

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.47738/jads.v7i4.1470

Abstract

Marine diesel engines are critical components of ship propulsion systems, and unexpected failures can lead to operational disruptions, increased maintenance costs, and reduced reliability. Conventional maintenance approaches may not adequately capture complex degradation patterns under varying operating conditions. This study proposes a Digital Twin–driven artificial intelligence (AI) framework for predictive maintenance of marine diesel engines using multi-sensor Internet of Things (IoT) data. The framework integrates engine sensor measurements, Digital Twin-derived information, and AI-based analytics to support fault detection and remaining useful life (RUL) estimation. Sensor data are processed to obtain relevant operating features, while residuals between measured and Digital Twin-predicted variables can provide additional information about deviations from expected engine behaviour. These features are used as inputs for fault classification and RUL prediction models. Based on the reported results, the proposed integrated framework achieved a fault detection accuracy of 93.8%, compared with 89.7% for the AI-based sensor data approach and 86.4% for the baseline approach. The reported RUL prediction error decreased to 9.2 hours, compared with 14.2 hours and 18.5 hours for the respective comparison approaches. These findings suggest that combining multi-sensor IoT data with Digital Twin-derived information may improve engine condition assessment and RUL estimation. However, the reliability of these findings depends on the underlying dataset, experimental design, and consistency of the evaluation results. Further validation using real-world marine engine data and diverse operating conditions is necessary to assess the framework's generalizability and practical applicability. The proposed approach offers a potential foundation for data-driven maintenance planning and improved operational reliability in marine engineering.