search

akshathmangudi / rgb-to-hyper

end-to-end implementation of microplastics detection in water using hyperspectral imaging.
MIT License
0 stars 2 forks source link

Complete Stage 2 #4

Open akshathmangudi opened 4 weeks ago

akshathmangudi commented 4 weeks ago

TASK: Microplastic Detection using the GENERATED HSI image. Should be complete by November 4th.

harshithasudhakar commented 2 weeks ago

We will be implementing UNet for this stage. This is how we will be doing it: The microplastics dataset has 368 RGB images and the corresponding masks. (Only focusing on fibres for now, future works: train model for all kind of microplastics)

  1. Extract imgs, organize the images by segregating into different dirs.(implemented in unet_main.py)
  2. Preprocess the RGBs (done by the load_rgb() in utils.py), load_model_and_predict() [see unet_main.py and utils for implementation]
  3. Customize UNet to take in 31 channels, because it's generally 2-3 channels (gotta add more layers)
  4. Overlay the segmentation mask on the original RGB image. This method gives a clear, direct view of the exact regions where microplastics are detected.
  5. Display the original image - segmentation model with a side-by-side comparison of the predicted segmentation masks and the ground truth masks
harshithasudhakar commented 2 weeks ago

UNet Architecture

image