Smitha Kurian
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Smitha Kurian has a total of 15 co-authors. Here are the top co-authors by number of shared publications:
A. Suresh
Department of Computer Science and Engineering, GRT Institute of Engineering and Technology, Tiruttani, India
1 shared publication- Advaith Dinesan Pudussery
Computer Science & Engineering HKBK College of Engineering Bangalore India
1 shared publication - Alen Roy
Computer Science & Engineering HKBK College of Engineering Bangalore India
1 shared publication - D. Loganathan
Cambridge Institute of Technology
1 shared publication - Husna Tabassum
HKBK College of Engineering
1 shared publication - Ihsan Rafeeque Sainudheen
Computer Science & Engineering HKBK College of Engineering Bangalore India
1 shared publication
Latest Publications
- Open Access
- 8 October 2025
- Published by EDP Sciences in ITM Web of Conferences
- Vol. 79, 01041
U-Net++ and ConvLSTM based Semantic Event Stream Processing for Real-Time Flood Monitoring
Currently, Floods are the most popular and devastating natural hazards, and their effects on economic stability and social well-being are continuously increasing. Recent methods have focused on fast and accurate detection of submerged regions to improve emergency planning and damage estimation in spatial and temporal dimensions. Satellite Multispectral Images have limited spectral bands and low spatial resolution, which restrict the depth of analysis. Image processing methods use intensity-hue-saturation and statistical features to enhance the spatial and spectral data before segmentation and classification. The proposed UConvFloodNet starts by removing noise for sharper inputs using Wiener filtering, and increases the spatial contrast by converting pixel intensities. With these enhanced images, U-Net++ segments key areas such as flood zones, water bodies, and land covers and tracks how flooding evolves over time, and ConvLSTM adds a temporal layer, capturing changes frame by frame. Together, these steps form a streamlined outline for real-time flood monitoring, which combines image enhancement, spatial enhancement, precise segmentation, and dynamic tracking. The UConvFloodNet performs an Accuracy of 99.31 and 99.06 for Sen1Floods11 and S1GFloods datasets which is better than existing Compact Convolutional Tokenizer integrating U-Net and Vision Transformer (CCT-U-ViT).
- 8 October 2025
- Published by EDP Sciences in ITM Web of Conferences
- Vol. 79, 01041
Open AccessU-Net++ and ConvLSTM based Semantic Event Stream Processing for Real-Time Flood Monitoring
Currently, Floods are the most popular and devastating natural hazards, and their effects on economic stability and social well-being are continuously increasing. Recent methods have focused on fast and accurate detection of submerged regions to improve emergency planning and damage estimation in spatial and temporal dimensions. Satellite Multispectral Images have limited spectral bands and low spatial resolution, which restrict the depth of analysis. Image processing methods use intensity-hue-saturation and statistical features to enhance the spatial and spectral data before segmentation and classification. The proposed UConvFloodNet starts by removing noise for sharper inputs using Wiener filtering, and increases the spatial contrast by converting pixel intensities. With these enhanced images, U-Net++ segments key areas such as flood zones, water bodies, and land covers and tracks how flooding evolves over time, and ConvLSTM adds a temporal layer, capturing changes frame by frame. Together, these steps form a streamlined outline for real-time flood monitoring, which combines image enhancement, spatial enhancement, precise segmentation, and dynamic tracking. The UConvFloodNet performs an Accuracy of 99.31 and 99.06 for Sen1Floods11 and S1GFloods datasets which is better than existing Compact Convolutional Tokenizer integrating U-Net and Vision Transformer (CCT-U-ViT).
- Open Access
- 12 June 2025
- Published by International Research Publication and Journals in International Journal on Science and Technology
- Vol. 16 (2)
Computer Vision-Based Anomaly Detection in Industrial Components
Anomaly detection in industrial components is essential for ensuring product quality and operational efficiency by identifying defects early. Advances in computer vision and deep learning have transformed traditional inspection, enabling automated systems to detect subtle anomalies with high accuracy. This survey reviews recent approaches, including supervised and unsupervised methods, focusing on deep neural networks, memory-augmented models, and feature clustering techniques that reduce reliance on labeled anomaly data. Advanced preprocessing, feature extraction, and patch-wise analysis enhance detection sensitivity and localization in high-resolution images. Real-time frameworks and lightweight architectures allow millisecond-level inference, suitable for production lines. The fusion of multiple data modalities, like RGB and depth data, further improves robustness in complex environments. However, challenges remain in handling data imbalance, distribution shifts, and generalizing models across diverse industrial settings. This survey discusses benchmark datasets, evaluation metrics, and future research directions, such as explainable AI, hybrid learning, and scalable adaptive systems for next-generation industrial anomaly detection.
- 12 June 2025
- Published by International Research Publication and Journals in International Journal on Science and Technology
- Vol. 16 (2)
Open AccessComputer Vision-Based Anomaly Detection in Industrial Components
Anomaly detection in industrial components is essential for ensuring product quality and operational efficiency by identifying defects early. Advances in computer vision and deep learning have transformed traditional inspection, enabling automated systems to detect subtle anomalies with high accuracy. This survey reviews recent approaches, including supervised and unsupervised methods, focusing on deep neural networks, memory-augmented models, and feature clustering techniques that reduce reliance on labeled anomaly data. Advanced preprocessing, feature extraction, and patch-wise analysis enhance detection sensitivity and localization in high-resolution images. Real-time frameworks and lightweight architectures allow millisecond-level inference, suitable for production lines. The fusion of multiple data modalities, like RGB and depth data, further improves robustness in complex environments. However, challenges remain in handling data imbalance, distribution shifts, and generalizing models across diverse industrial settings. This survey discusses benchmark datasets, evaluation metrics, and future research directions, such as explainable AI, hybrid learning, and scalable adaptive systems for next-generation industrial anomaly detection.
- 21 December 2024
- Published by Springer Nature
- p. 243-256
A Literature Survey—Application of Image Processing for the Effective Diagnosis of Rod Cone Dystrophy Using OCT
- 21 December 2024
- Published by Springer Nature
- p. 243-256
A Literature Survey—Application of Image Processing for the Effective Diagnosis of Rod Cone Dystrophy Using OCT
- 23 March 2021
- Published by Springer Nature in Wireless Networks
- Vol. 27 (4) , 2497-2508
Novel AODV based service discovery protocol for MANETS
- 23 March 2021
- Published by Springer Nature in Wireless Networks
- Vol. 27 (4) , 2497-2508
Novel AODV based service discovery protocol for MANETS
- 30 November 2019
- Published by Blue Eyes Intelligence Engineering and Sciences Engineering and Sciences Publication (BEIESP) in International Journal of Recent Technology and Engineering (IJRTE)
- Vol. 8 (4) , 4068-4072
AODV based Service Discovery Protocol with Two Hop Neighbor Information
AdHoc networks are self-sustained infrastructure less networks which survives on mutual cooperation by sharing resources. To share the resources as services it is important to be able to advertise available services and discover the required services. Service discovery that is performed in application layer can be benefitted when performed at network layer since it reduces the control message. Therefore the message format of AODV protocol was extended to include service discovery and service advertisements. We propose a method to extend the hello messages of AODV protocol further to include the service information available with the 2-Hop neighbors. To reduce the storage overhead every node would only store the availability of a particular service. Thereby to access the services available at a 2- hop distance the process of service discovery is not required only a uncast message can be sent. This approach has been able to reduce the service acquisition latenc.y
- 30 November 2019
- Published by Blue Eyes Intelligence Engineering and Sciences Engineering and Sciences Publication (BEIESP) in International Journal of Recent Technology and Engineering (IJRTE)
- Vol. 8 (4) , 4068-4072
AODV based Service Discovery Protocol with Two Hop Neighbor Information
AdHoc networks are self-sustained infrastructure less networks which survives on mutual cooperation by sharing resources. To share the resources as services it is important to be able to advertise available services and discover the required services. Service discovery that is performed in application layer can be benefitted when performed at network layer since it reduces the control message. Therefore the message format of AODV protocol was extended to include service discovery and service advertisements. We propose a method to extend the hello messages of AODV protocol further to include the service information available with the 2-Hop neighbors. To reduce the storage overhead every node would only store the availability of a particular service. Thereby to access the services available at a 2- hop distance the process of service discovery is not required only a uncast message can be sent. This approach has been able to reduce the service acquisition latenc.y