The project aims to automate the manual dead/alive plant counting process in a greenhouse. The manual process can takes 1 day for each experiment.

  • Define data annotation protocol for scientists in order to obtain bounding box annotation for trays in a greenhouse bench, and dot annotation for dead/alive plants in a tray
  • Develop deep learning segmentation model (U-net) to segment out individual trays in high resolution bench images using Python and Tensorflow
  • Develop computer vision models (color threshold-based) to detect individual plants in a tray and classify them as dead/alive using Python, OpenCV
  • Deploy the models in production, reducing the counting time from 1 day (manual counting) to 1 hour