Yuchen Xue

Technical assistant at Fraunhofer IFF

2023-05 - 2024-04, Magdeburg (Germany)

I developed a industrial-level defect detection solution based on an existing neural network, achieving an AUC score of 97.6% with limited amount of data. This solution features a technique for automatically producing point prompt, which improves the performance of image segmentation on low-contrast images of irregular machinery parts.

• Improved segmentation result through automatic point prompt generation
• A reliable solution unaffected by poor image quality
• High performance regardless of irregular shapes of different machine parts
• Zero-shot solution without the involvement of a large dataset
• Enhanced quality control in industrial processes

link to the project

Master Thesis: "Survey on Facial Recognition Models Using Deep Learning Technologies"

This is an academic survey of different deep-learning based face recognition models, aiming at bringing insights to future researches in the field of deep learning.

• Literal survey of different face recognition models
• Scientific analysis of the fundamental differences of selected solutions
• Use of experimental method for comparing the performance of selected models

Publication and Academic Speech with the Topic "Improved Part-aligned Deep Features Learning for Person Re-Identification"

This is a solution to the challenge of person re-identification (Re-ID) problem. The Re-ID problem is about recognizing and matching pedestrians with different appearances across different cameras.

• Development of a solution to Re-ID problem with the help of computer vision and deep learning methods
• Contributed to a scientific paper, which was accepted by the 16th IEEE International Conference on Advanced Video and Signal-based Surveillance (AVSS)
• Delivered a workshop speech on this conference

Real-Time Detection of Pedestrians and Vehicles in Video Streams of Surveillance Cameras

A deep learning based object detection solution for traffic monitoring inside the campus. Image data was manually collected, labelled and used for training a detector based on a pre-trained YOLOv3 model. The newly trained model shows significant improvement in our use case

• Improvement of 20% in precision compared to the base model
• Improvement on small, obscured, and overlapping objects
• Training of a reliable model with limited data using transfer learning technique

Energy Consumption Analysis of a Commercial Building Using a Generalized Additive Model

This project, which serves as the final project of the course "Regression Analysis", is about the analysis and prediction of the energy consumption in an office building using a Generalized Additive Model (GAM). Data used in this project is originated from an open source dataset, namely "EnerNOC GreenButton Data".

• Implementation of a GAM model with high capability of generalization using R language
• The model exhibits a MAPE (mean absolute percentage error) value of 14.89 when predicting the energy consumption of the next two weeks
• Composing and hosting of a HTML report using R language, Quarto and Github Action

Online R Tutorial Website Based on Quarto

This is a website that contains my study note and projects of R language.

• Developed of a website using Quarto
• Automated deployment using Github Action

Web Application for Image Object Detection Based on Spring Framework

A web application for hosting a deep learning based image object detection with the following function: a user can upload an image and download a new image with bounding boxes, with the detection result is stored in a database.

• Implementation of the Spring Boot backend;
• Deployment of an object detection model using TensorFlow for Java;
• Implementation of MySQL database and Spring Database Connector for storing detection result.

"You Smile I Serve" – A Concept of Smart Restaurant Based on Raspberry-Pi and Python

This project is a prototype of an intelligent restaurant that provides a better customer experience when ordering. A customer is encouraged to smile to a camera to trigger a chatbot and can order their food by talking to this chatbot. A GUI was also developed for the restaurant owner to keep track of customer orders. The main components include a Raspberry Pi single board computer, a microphone, a webcam and a Pi-camera.

• The design of the whole software and hardware structure
• Software implementation of smile detection and QR code reading
• Use of Google Cloud Platform API for implementing the voice recognition function
• Development of a user interface for managing customer's order

Bachelor Project: "Self-Navigable Quadcopter based on Arduino and MultiWii"

A collaborative project aims to build a quadcopter capable of navigating along a pre-defined route. This project won the best bachelor project of the Department of Electrical Engineering of Thunhai University during the year 2016.

• Assemble of a quadcopter from hardware components, including a control board, motors, propeller, a frame and a battery
• Programming and testing of the control system
• Attempt of implementing a software program for the auto-piloting function

Improved Part-aligned Deep Features Learning for Person Re-Identification

2019-11-25

16th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS)

Authors

• Sheng-Luen Chung
• Yuchen Xue
• Shih-Chao Chien
• Ruei-Shan Chan

Cloud Encryption Using Distributed Environmental Keys

2016-12-22

10th International Conference on Innovative Mobile and Internet Services in Ubiquitous Computing (IMIS)

Authors

• Kun-Lin Tsai
• Fang-Yie Leu
• Yi-Fung Huang
• Chi Yang
• Cheng-Hsin Chang
• King-Shing Yip
• Yuchen Xue
• Guan-Chi Lai

Qualification Program -- Statistics Software R (advanced)

2024-09 - Rostock (Germany)

A 4-day workshop that features multiple theoretical topics and practical sessions related to multiple statistics topics, including linear model, anova, mean comparison with post-hoc tests, experimental designs and professional illustration

The best annual final year project of the Department of Electrical Engineering

2016-08 - Taichung (Taiwan)

"Self-Navigable Quadcopter based on Arduino and MultiWii" was granted as the best annual final year project of the Department of Electrical Engineering at Thunhai University

Calculus Competition at Thunghai University

2014-09 - Taichung (Taiwan)

Won a prize in the Calculus Competition held by the Department of Applied Mathematics at Thunhai University

Teaching assistant of practice sessions

2015-09 - 2017-06, Taipei and Taichung (Taiwan)

I volunteered as a teaching assistant for three courses during my study at Thunghai University (Taichung, Taiwan) and National Taiwan University of Science and Technology (Taipei, Taiwan). My role was to introduce each assignment to the students, as well as to assist them during the practice session. Summary of these course:

• "Logic Design Laboratory" -- Verilog HDL language programming using FPGA
• "Introduction to Cyber-Physical System" -- Python programming on Raspberry Pi
• "Regression Analysis" -- Theory of linear regression model and practices of R language