Challenge
Given a Bayer color filter’s black-and-white SEM image dataset, including 300 images, we need to detect and recognize the Red, Green, and Blue cells.
- Highly noisy
- The quality of the images input is variable
- In some samples, many mix colors between two cells adjacent
- Imbalance data (R, G, B) in training and testing data
Solution
We use a deep learning approach (YOLO5 model) which detects the color of cells and can overcome the difficulties that we have faced up with. There is only one problem with deep learning is the unbalance between the red and blue pattern datasets, In order to solve this problem, we use image augmentation to increase the samples of the red pattern by rotating, clipping, blur, sharpen the original images
Outcome
The testing result reach to 100% accuracy.
