Projects

Wearable Soft Magnetic Sensor Arrays for Robust FMG Recognition

2025 – Present

• Building upon a soft-magnetic FMG prosthetic sensing system for real-time muscle pattern recognition
• Developing algorithms to handle sensor displacement and subject variability due to different wearing positions
• Exploring OOD detection and cross-domain adaptation methods to improve robustness and classification accuracy

Intelligent Data Analysis and Decision System for Construction Sites

2022

• Implemented real-time safety compliance inspection using YOLO for unsafe behaviors and warning-sign detection
• Designed server-side multi-stream video scheduling and parallel detection interfaces for early-warning analytics

Anti-Forgery Code Identification System for Cigarette Packaging

2021

• Built a commercial-grade 32-bit CRNN-based recognition system for transparent laser anti-counterfeiting codes
• Migrated PyTorch models (CRNN, CTPN, YOLO, DBNet) to NCNN through ONNX
• Optimized C++ inference pipelines for cross-platform deployment
• Integrated C++ and Java via JNI for Android and WeChat mini-program applications

Recognition System for Traditional Chinese Medicine Pieces

2021

• Developed a ResNet-based classification and quality-grading model
• Improved accuracy from 79% → 96.9% via dataset curation, augmentation, and iterative refinement
• Deployed a lightweight network service for mobile access

Big Data Safety Monitoring and Management System for Tourism

2019

• Enhanced SlowFast-based violence recognition model (↑ 5% accuracy) for real-time deployment
• Improved crowd-counting accuracy by 30% through algorithmic optimization
• Developed abnormal-crowd detection and real-time alerting functions for high-risk scenes