B. Subbulakshmi
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Research Interests
B. Subbulakshmi has a total of 27 co-authors. Here are the top co-authors by number of shared publications:
- M. Nirmala Devi
Thiagarajar College of Engineering
10 shared publications - Deisy Chelliah
Thiagarajar College of Engineering
3 shared publications - Sivakumar Muthuramalingam
Thiagarajar College of Engineering
3 shared publications - A. Rakshana Malya
Mepco Schlenk Engineering College
2 shared publications - S. Mahesh
Thiagarajar College of Engineering,Computer Science and Engineering,Madurai,India
2 shared publications - Thanusri V
Thiagrajar College of Engineering,Department of Computer Science and Engineering,Madurai,India
2 shared publications
Latest Publications
- 1 April 2026
- Published by Springer Nature
- p. 252-266
Boosting-Based Prediction and Auto-Optimization Models for Irrigation Scheduling, Soil Deficiency Analysis
- 1 April 2026
- Published by Springer Nature
- p. 252-266
Boosting-Based Prediction and Auto-Optimization Models for Irrigation Scheduling, Soil Deficiency Analysis
- 1 April 2026
- Published by Springer Nature
- p. 203-216
A Comprehensive Review of Methods and Approaches for Clickbait Detection
- 1 April 2026
- Published by Springer Nature
- p. 203-216
A Comprehensive Review of Methods and Approaches for Clickbait Detection
- 13 January 2026
- Published by Springer Nature
- p. 491-504
GAE-MedBERT: Leveraging Graph Neural Networks to Optimize Chronic Disease Risk Prediction from Patient Records
- 13 January 2026
- Published by Springer Nature
- p. 491-504
GAE-MedBERT: Leveraging Graph Neural Networks to Optimize Chronic Disease Risk Prediction from Patient Records
- 20 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1957-1962
Enhanced Tuberculosis Detection using a Dual-Path Vision Model with Feature-Space Augmentation and Progressive Fine-Tuning
Tuberculosis (TB) remains a major global health threat, necessitating rapid and accurate diagnostic tools. Conventional Chest X-ray (CXR) interpretation is often hampered by resource constraints and human variability. This paper presents DPVM-MS (Dual-Path Vision Model with Mirage Synthesizer), a novel hybrid architecture for pulmonary tuberculosis (PB) detection from chest radiography (CXR) images, that integrates both ResNet50 for local image features and ViT-Base Transformer for the global context in the image via Multihead Attention mechanism. The architecture was deliberately designed to enhance generalization and regularization by introducing the novel Mirage Synthesizer (MS), a learnable and perturbatively designed feature-space augmentation module, along with a stabilizing Progressive Fine-Tuning (PFT) regimen. The evaluation included strict patient-wise stratification and cross-dataset validation across four cohorts: Shenzhen (N=662) and Montgomery (N=138) and DATASET A (N=156) and DATASET B (N=122) (from the National Institute of Tuberculosis and Respiratory Diseases, New Delhi). DPVM-MS achieved high validation accuracies of 90.62% to 96.00% and significantly outperformed the single-path baselines; and for balanced generalization, it achieved perfect Specificity (1.0000) in DATASET B and perfect Sensitivity (1.0000) in the Montgomery dataset, which demonstrated it was a highly accurate and robust positioning it as a resilient diagnostic tool suitable for WHO screening and triage recommendations in low-resource settings.
- 20 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1957-1962
Enhanced Tuberculosis Detection using a Dual-Path Vision Model with Feature-Space Augmentation and Progressive Fine-Tuning
Tuberculosis (TB) remains a major global health threat, necessitating rapid and accurate diagnostic tools. Conventional Chest X-ray (CXR) interpretation is often hampered by resource constraints and human variability. This paper presents DPVM-MS (Dual-Path Vision Model with Mirage Synthesizer), a novel hybrid architecture for pulmonary tuberculosis (PB) detection from chest radiography (CXR) images, that integrates both ResNet50 for local image features and ViT-Base Transformer for the global context in the image via Multihead Attention mechanism. The architecture was deliberately designed to enhance generalization and regularization by introducing the novel Mirage Synthesizer (MS), a learnable and perturbatively designed feature-space augmentation module, along with a stabilizing Progressive Fine-Tuning (PFT) regimen. The evaluation included strict patient-wise stratification and cross-dataset validation across four cohorts: Shenzhen (N=662) and Montgomery (N=138) and DATASET A (N=156) and DATASET B (N=122) (from the National Institute of Tuberculosis and Respiratory Diseases, New Delhi). DPVM-MS achieved high validation accuracies of 90.62% to 96.00% and significantly outperformed the single-path baselines; and for balanced generalization, it achieved perfect Specificity (1.0000) in DATASET B and perfect Sensitivity (1.0000) in the Montgomery dataset, which demonstrated it was a highly accurate and robust positioning it as a resilient diagnostic tool suitable for WHO screening and triage recommendations in low-resource settings.
- 20 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1279-1283
An Enhanced Skin Cancer Detection and Classification Using Deep Convolutional Neural Network
Skin cancer is an important health concern, and its early detection and prompt diagnosis are key to improving outcomes for patients. This research undertakes a comparison of three deep learning models on the subject of multi-class classification of dermoscopic skin lesions using a publicly available HAM10000 dataset. A custom-designed CNN, a pretrained ResNet50, and a Swin transformer were the models evaluated in this study. The proposed CNN achieved 95.0% overall accuracy with closely matched macro averaged precision and recall across all classes, indicating stable performance despite class imbalance. In comparison, the Swin Transformer and ResNet50 reached 88.0% and 85.4% accuracy, respectively. these results suggest that a domain tailored CNN with carefully chosen capacity and inductive biases can outperform more complex, general-purpose architectures on dermoscopic images. This study concludes that specialized CNNs are a reliable, effective option for computer-aided skin lesion diagnosis and can support safer, earlier detection in practice.
- 20 December 2025
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 1279-1283
An Enhanced Skin Cancer Detection and Classification Using Deep Convolutional Neural Network
Skin cancer is an important health concern, and its early detection and prompt diagnosis are key to improving outcomes for patients. This research undertakes a comparison of three deep learning models on the subject of multi-class classification of dermoscopic skin lesions using a publicly available HAM10000 dataset. A custom-designed CNN, a pretrained ResNet50, and a Swin transformer were the models evaluated in this study. The proposed CNN achieved 95.0% overall accuracy with closely matched macro averaged precision and recall across all classes, indicating stable performance despite class imbalance. In comparison, the Swin Transformer and ResNet50 reached 88.0% and 85.4% accuracy, respectively. these results suggest that a domain tailored CNN with carefully chosen capacity and inductive biases can outperform more complex, general-purpose architectures on dermoscopic images. This study concludes that specialized CNNs are a reliable, effective option for computer-aided skin lesion diagnosis and can support safer, earlier detection in practice.
- 14 February 2025
- Published by Springer Nature
- p. 165-183
Lightweight Unmanned Surface Vehicles Detection in Maritime Regions through Yolov8x with Focus and Attention Mechanisms
- 14 February 2025
- Published by Springer Nature
- p. 165-183
Lightweight Unmanned Surface Vehicles Detection in Maritime Regions through Yolov8x with Focus and Attention Mechanisms
- 2 December 2024
- Published by Taylor & Francis
- p. 187-208
Tutor Ward Mentoring Chatbot
In today’s ever-changing education landscape, the demand for personalized mentorship is more crucial than ever. Educators play a vital role in shaping students not just academically, but also in developing their character, nurturing critical thinking, instilling values, and fostering a love for learning. The enduring teacher–student relationship forms a foundation for creating well-rounded individuals equipped with life skills, empathy, and a basis for future success as education continues to evolve. Therefore, educational institutions need to support and mentor students through a tutor ward system. In this chapter, we propose an artificial intelligence (AI)-driven chatbot designed to enhance mentoring in educational institutions. Tailored to influence students’ personal development, it guides ethical decision-making, academic progress, and navigating financial challenges. This innovative AI-guided mentoring platform serves as a contemporary solution to address students’ challenges effectively, ensuring the seamless allocation of tutors based on individual needs.
- 2 December 2024
- Published by Taylor & Francis
- p. 187-208
Tutor Ward Mentoring Chatbot
In today’s ever-changing education landscape, the demand for personalized mentorship is more crucial than ever. Educators play a vital role in shaping students not just academically, but also in developing their character, nurturing critical thinking, instilling values, and fostering a love for learning. The enduring teacher–student relationship forms a foundation for creating well-rounded individuals equipped with life skills, empathy, and a basis for future success as education continues to evolve. Therefore, educational institutions need to support and mentor students through a tutor ward system. In this chapter, we propose an artificial intelligence (AI)-driven chatbot designed to enhance mentoring in educational institutions. Tailored to influence students’ personal development, it guides ethical decision-making, academic progress, and navigating financial challenges. This innovative AI-guided mentoring platform serves as a contemporary solution to address students’ challenges effectively, ensuring the seamless allocation of tutors based on individual needs.
- 20 September 2024
- Published by Springer Nature
- p. 469-484
Severity Levels Categorization for Detecting Diabetic Retinopathy Using YOLOv7
- 20 September 2024
- Published by Springer Nature
- p. 469-484
Severity Levels Categorization for Detecting Diabetic Retinopathy Using YOLOv7
- 3 February 2024
- Published by Springer Nature
- p. 310-321
A Deep Learning Based Bio Fertilizer Recommendation Model Based on Chlorophyll Content for Paddy Leaves
- 3 February 2024
- Published by Springer Nature
- p. 310-321
A Deep Learning Based Bio Fertilizer Recommendation Model Based on Chlorophyll Content for Paddy Leaves