Job Details – Good & Defective Items Detection Using YOLO
Project Overview:
Build an advanced computer vision system that automatically detects and classifies items into Good or Defective categories using YOLO (You Only Look Once). The project helps automate quality control processes for manufacturing, agriculture, or product inspection pipelines.
Responsibilities:
Prepare and clean a dataset containing images labeled as Good and Defect (around 60 defect classes and 70 good samples or more).
Annotate images using tools like Roboflow, LabelImg, or CVAT.
Train a custom YOLO model (YOLOv8/YOLO11s) with optimized hyperparameters for high accuracy and low latency.
Implement data augmentation to improve generalization in real-world environments.
Build a script for real-time detection using camera input or video files.
Generate predictions and automatically export results to CSV or Excel, including class, confidence score, and bounding box coordinates.
Optimize model performance for deployment on edge devices (optional).
Document the workflow and provide clear instructions for future updates.
Technical Skills Needed:
YOLO (YOLOv5, YOLOv8, or YOLO11s).
Python (NumPy, OpenCV, Torch).
Dataset annotation and preprocessing.
Model training, validation, and hyperparameter tuning.
Real-time computer vision pipelines.
Exporting structured detection results to CSV/Excel.
Understanding of defect inspection systems and manufacturing QC (optional).
Deliverables:
Trained YOLO model for Good vs Defect classification.
Confidence-based detection results for single images, video streams, or folders.
CSV/Excel output containing detection metadata.
Full training code, inference code, and documentation.
