Kc C.

Key Skills

Software

LabelImg

Labelbox

Top Subject Matter

Manufacturing – Industrial Defect Detection & Quality Control

Railway & Logistics – Freight Inspection & Safety Monitoring

Computer Vision & AI – Custom Dataset Development & Annotation

Top Data Types

Image

Top Task Types

Bounding Box

Segmentation

Classification

Object Detection

Freelancer Overview

With hands-on experience spanning three full-cycle computer vision projects across electronics manufacturing and railway freight inspection, I bring deep expertise in data labeling, custom dataset development, and AI model optimization for real-world industrial scenarios. Across projects including PCB surface defect detection, open wagon side door lock status monitoring, and residual coal inspection, I designed standardized bounding box annotation workflows, built tailored datasets, and implemented rigorous quality control measures—such as cross-validation and guideline-driven reviews—to ensure high-quality, consistent training data, even under challenging field conditions like variable lighting, dust interference, and complex backgrounds. I further translated these datasets into production-ready solutions by refining state-of-the-art models (YOLOv7, RT-DETR) into lightweight, fast-inference variants optimized for on-site deployment, achieving 90–91% detection accuracy. Critically, two of my models have been successfully deployed in operational settings at Liangshan Station on the Haoji Railway, directly replacing manual inspection to reduce labor intensity and boost efficiency, while the PCB defect detection project delivered a scalable framework for intelligent quality control in electronics manufacturing. This end-to-end experience—from defining annotation requirements and building robust datasets to optimizing and deploying AI models—equips me with a rare combination of data labeling rigor, technical model-tuning skills, and a proven track record of solving practical industrial inspection challenges.

Labeling Experience

Railway Coal Wagon Residual Coal Visual Inspection: Data Annotation & Lightweight RT-DETR Model Development

ImageClassification

Developed an intelligent visual inspection system to meet the practical operational needs of railway freight, enabling automated residual coal detection in open-top coal wagons after unloading, replacing manual visual checks to improve inspection efficiency. Completed full-cycle data preparation and annotation: collected on-site images of coal wagons post-unloading, built a custom residual coal detection dataset using bounding box annotation to label residual coal regions, and developed standardized annotation guidelines to adapt to complex field conditions (e.g., varying lighting and coal dust interference). Implemented quality control measures including inter-annotator cross-validation and periodic accuracy audits to ensure annotation consistency and dataset reliability. Improved the RT-DETR detection model by designing a lightweight, fast-inference variant optimized for field deployment. The model achieved a 90% detection accuracy rate for residual coal, and has been successfully applied in practical operations at Liangshan Station on the Haoji Railway. The solution significantly enhanced the efficiency of residual coal inspection and reduced reliance on manual labor.

2024 - 2024

Railway Open Wagon Side Door Lock Status Visual Inspection: Data Annotation & AI Model Development

ImageObject Detection

Designed an AI visual inspection system to replace manual visual checks for railway open wagon side door lock status, fully compliant with railway freight inspection regulations and on-site operational requirements. Completed full-cycle data preparation and annotation: collected on-site images of open wagon side doors, built a custom labeled dataset using bounding box annotation to mark lock status categories (locked/unlocked), and developed standardized annotation guidelines tailored to complex railway yard conditions (e.g., varying lighting and weather). Implemented quality control measures including inter-annotator cross-validation and periodic accuracy audits to ensure annotation consistency and reliability. Improved the YOLOv7 detection model by integrating linear feature extraction and dynamic perception modules to enhance adaptability to field scenarios. The optimized model achieved 91% detection accuracy and has been successfully deployed in practical freight inspection operations at Liangshan Station on the Haoji Railway, effectively reducing manual labor intensity and improving railway freight inspection efficiency.

2023 - 2024

Printed Circuit Board (PCB) Defect Detection Data Annotation & AI Model Development

ImageObject Detection

Conducted on-site research at a PCB manufacturing plant to analyze production workflows and define requirements for intelligent transformation of manual visual inspection processes. Collected PCB surface images and built a custom defect dataset by annotating defects by category using bounding boxes, establishing standardized annotation guidelines to ensure data consistency. Implemented quality control measures including cross-checks and periodic reviews to maintain high annotation accuracy. Trained a computer vision detection model on the custom dataset, and optimized it into a lightweight network to meet real-time inference requirements in industrial scenarios. The solution effectively replaced manual visual inspection, improved production line efficiency, and delivered a practical AI-powered quality control solution for PCB manufacturing.

2022 - 2022

Education

Work History