Developed a U-Net-based deep learning model for binary segmentation of porosity defects in steel microstructures using X-ray Computed Tomography (XCT) images. Trained on 2,000 images with OpenCV & TensorFlow, achieving 85% IoU on validation data. Currently optimizing the model to improve segmentation accuracy.

Key Features:

U-Net architecture with skip connections for precise defect segmentation Preprocessing with Otsu’s thresholding for binary mask generation IoU-based performance evaluation for model accuracy tracking Ongoing improvements to enhance defect detection