Mingyuan Yang

Mingyuan Yang

Graduate researcher specializing in Computer Vision and SLAM, with full-pipeline experience from algorithm design to edge deployment on NVIDIA Jetson. Passionate about visualizing academic work through modern web technology.

Guangzhou, China

Education

South China University of Technology

Master of Engineering · Computer Science and TechnologyGPA 3.8 / 4.0

2023-092026-06
  • Research focus: Underwater object detection & multi-object tracking
  • Lab: Computer Vision & Intelligent Systems Lab

South China University of Technology

Bachelor of Engineering · Software EngineeringGPA 3.7 / 4.0

2019-092023-06
  • Outstanding Graduation Thesis
  • National Encouragement Scholarship

Publications & Patents

Lightweight Fish Re-Identification via Part-Aware Feature Fusion for Aquaculture IoT Systems

Mingyuan Yang, Wei Chen, Haoran Liu, Jianfeng Zhang

IEEE Internet of Things Journal, 2025

We propose a lightweight fish re-identification method combining part-aware feature fusion, achieving 20+ FPS real-time inference on Jetson Nano with 5.2% mAP improvement over baselines.

ReIDEdge AIIoTAquacultureTensorRTDOI arXiv

一种面向嵌入式端的 YOLOv11 通道剪枝与量化压缩方法

杨明远, 陈伟, 刘浩然

China National Intellectual Property Administration, 2024(CN202410000001.X)

A systematic compression scheme for YOLOv11 via structured channel pruning and INT8 quantization, reducing model size by 60% and achieving 3.7× inference speedup for embedded real-time deployment.

YOLOv11PruningQuantizationEmbeddedPatent

Featured Projects

View all

Underwater VIO-SLAM Localization System

Tightly-coupled Visual-Inertial Odometry prototype fusing IMU and monocular camera

ATE RMSE 0.043mIMU-Camera Latency < 2ms
ORB-SLAM3ROS2IMU 预积分g2oC++17+3

Underwater Fish Re-ID & Tracking System

Edge real-time MOT pipeline based on YOLOv11 + BoT-SORT

+5.2% mAP@503.7× Inference Speedup3.1% ID Switch Rate
PyTorchTensorRTYOLOv11BoT-SORTCUDA+3