Active Robotic Sensing Lab - Research Assistant (Computer Vision)
Location: Raleigh, NC
Duration: Jan 2023 - Sept 2023
Semantic Segmentation & Object Detection on Turfgrass Imagery
- Conducted applied computer vision research to segment damaged vs. healthy St. Augustine turfgrass under drought stress, supporting high-throughput phenotyping in agricultural research.
- Implemented HSV color space thresholding and mean clustering to segment drought-affected regions in aerial images, achieving an average Structural Similarity Index (SSIM) of 72%.
- Applied YOLOv8 for object detection to localize turf damage areas, achieving a mean Average Precision (mAP) of 89.8% on a custom-labeled dataset.
- Explored a range of data augmentation strategies (flip, scale, color jitter, etc.) to improve model generalization under diverse field conditions.
- Utilized transfer learning with the ViT-16 (Vision Transformer) model for semantic segmentation, enabling effective feature extraction from limited labeled data.
- Contributed to improving drought resistance screening workflows for plant scientists by providing accurate segmentation tools for field-level analysis.