Amy Li

• 1. Trained and evaluated multiple semantic segmentation models (U-Net, DeepLabV3+) using PyTorch and OpenCV to detect dead trees from UAV imagery, processing 2,000+ aerial images.
• 2. Improved model mIoU (mean Intersection over Union) by 11% through implementing data augmentation (rotation, flipping, brightness adjustment) and hyperparameter tuning (learning rate, batch size).
• 3. Built a comprehensive evaluation pipeline integrating IoU/F1 score calculation and real-time performance visualization via TensorBoard, enabling team to track model progress and iterate efficiently.

1. Developed a real-time vehicle detection system by deploying YOLOv5 on Raspberry Pi (with Linux environment) and OpenCV for image preprocessing, enabling accurate identification of parked vehicles with 92% precision.

2. Integrated REST API and a mobile app interface (using Flutter, designed with Figma prototypes) to display live parking space availability, allowing users to check spot occupancy in real time.

3. Reduced system false detection rate by 18% through designing improved filtering logic (background subtraction, motion tracking), enhancing overall reliability for outdoor parking lots.