Publications
Peer-reviewed research published at IEEE and Springer venues.
Brain Tumor Detection and Segmentation using SAM Integrated YOLOv9 Scheme
Mridul Mayankeyshwar, Lookinder Kumar, Dev Yadav, Mamata P. Wagh
2024 IEEE 1st International Conference on Advances in Signal Processing, Power, Communication, and Computing (ASPCC) · Bhubaneswar, India
Proposes SIYO (SAM Integrated YOLOv9), integrating Meta's Segment Anything Model (SAM) with YOLOv9 for brain tumour detection and segmentation on MRI images. Evaluated on Br35H dataset (800 MRI images across training, validation, and test sets). Achieves mAP@0.5 = 0.947, overall accuracy = 0.94. YOLOv9 localises and detects tumours; SAM segments precisely using zero-shot generalisation. Data augmentation (rotation, scaling, flipping, noise) applied exclusively to training data.
Deep Learning-Based Tomato Plant Disease Detection using TomatoDoc Dataset: Agricultural Applications
Ahmad Ashraf Zargar, Mridul Mayankeyshwar, Lookinder Kumar, Debendra Muduli
Computational Intelligence in Pattern Recognition (CIPR 2024), Lecture Notes in Networks and Systems, vol. 1153 · Singapore
Introduces the TomatoDoc segmented dataset of 10,000 tomato leaf images across 10 disease classes (1,000 images each, 256×256px). K-means clustering isolates infected regions; MobileNet-V3 with a custom 3-layer dense head classifies disease. Advances agricultural diagnostics for smallholder farmers through smartphone-deployable disease detection. Bridges precision agriculture and accessible deep learning tools.
My Research Focus
My published work spans computer vision and agricultural AI — demonstrating an ability to apply deep learning to high-stakes, domain-specific problems. The SIYO paper (IEEE ASPCC 2024) tackles medical imaging, where precision and reliability are non-negotiable. The TomatoDoc paper (Springer CIPR 2024) addresses food security challenges through accessible, smartphone-deployable AI — published alongside an open-source dataset designed to advance agricultural diagnostics globally.
My current MSc thesis extends this research trajectory into financial AI and explainability — investigating adversarial robustness and SHAP explanation stability in fraud detection models, mapped against EU AI Act 2024 compliance requirements. The thread connecting all my research is the same: building AI systems that are rigorous enough to publish, reliable enough to deploy, and honest enough to explain.