Back
Deep Learning AOI Model
Project Description:
For this project, I was tasked to create a Deep Learning Model to detect automatically inspect electronic components on circuit board. This project is to be displayed and represent Fraunhofer Singapore in the coming trade show.
I have chosen to use CVAT to annotate and preprocess the image data in COCO json format. Additionally, using Detectron2, Facebook AI Research's next generation library that provides state-of-the-art detection and segmentation algorithms to create the Deep Learning Model.
This work includes:
- Learning about the theory of CNNs for object detection
- Data collection & extension of an existing data set with a semi-automatic labelling tool
- Training of a CNN model with the extended data
- Evaluation of model performance (Accuracy, Precision, Recall etc. ) and comparing it to previous models
- Development of an additional classification module to analyse the connectivity of the electric components
My Contributions:
- Collect Data
- Annotate & Label data
- Extract & Combine data with colleagues
- Split data into train & test dataset with [8:2]
- Train the model using extracted data
Github Repositories:
Created Using:
- CV2
- Cuda
- Docker
- Detectron2
- Iriun
- Python | PyTorch
Professor Marius Erdt, Deputy Director of Fraunhofer
TestimonialPresentation Slides
Goh Ee Sheng © 2023