Defect Detection in Betaguard Yellow Sealing (Volvo Cars)
Master Thesis project focused on developing an automated defect detection system for Volvo Cars' battery lid sealing process. The system replaces manual inspection with a deep learning–based pipeline capable of identifying cracks, uneven seals, seal overflow, and other critical defects in Betaguard Yellow sealing.
- Developed a balanced dataset with 400 images per defect class
- Applied CNNs and VGG16 for high‑accuracy defect classification
- Implemented preprocessing: frame extraction, noise reduction, normalization
- Achieved significant improvements over traditional inspection methods
- Validated performance using accuracy, precision, recall, and F1‑score
Thesis Document: View Thesis
Brain Tumor Segmentation (Deep Learning)
A Streamlit-based web application that performs brain tumor segmentation on MRI images using a UNet deep learning model. The app displays the uploaded MRI, predicted tumor mask, and a large high-resolution visualization of the segmented tumor.
- Real-time tumor segmentation
- PyTorch-based UNet model
- Interactive Streamlit interface
- Overlay visualization for medical interpretation
Live Demo: Open App
GitHub Repository: View Code
Privacy-aware Object Detection (Decentralized AI)
Designed a privacy-focused object detection system for assistive applications.
Object Detection (YOLO vs Custom Model)
Compared YOLO performance on COCO vs custom datasets for industrial use cases.
GitHub Repository: View Code
Tourism Forecasting
Developed predictive models to forecast tourism demand using time-series techniques.
GitHub Repository: View Code
Sentiment Analysis (NLP)
Extracted insights from large-scale Amazon review datasets using NLP pipelines.