Activities and societies
圍棋社社長、科普志工社
Description
1. Six Defect Classification of Printed Circuit Boards based on Machine Learning-(In progress)
• PCB defect detection using Python, combining Generative Adversarial Network (DCGAN) and Active
Convolutional Fused Bidirectional Feature Pyramid Network with Remote Sensing Attention Mechanism.
2. AI-driven Building Footprint and Height Estimation from Aerial Imagery
• Designed a method for automatically segmenting building footprints and extracting height information
from aerial images. By leveraging semantic segmentation technology, the system can accurately identify
building outlines, estimate building heights, and combine 2D and 3D data, providing a powerful framework
for building analysis and 3D scene reconstruction.
3. Integrated OCR & NLP for Product Appraisal and Battery Recycling Classification
• Implemented an automated pipeline using Pillow for image pre-processing and PaddleOCR for optical
character recognition. The OCR results are stored in CSV format and subsequently processed through an
NLP module, streamlining the tasks of product appraisal and recycling battery classification.