Published on

UBC Agrobot

Authors
Wheat Head Detector

Summary

Agrobot Blogs and Agrobot Website

Image recognition work for the UBC Agrobot Design Team

  • Prepared and trained a plant health classifier with data augmentation in the background, with another team member — trained on web scraped leaf images
  • Completed a wheat head detector utilizing TensorFlow Object Detection API and transfer learning for Edge TPU deployment–with mAP of .84 on the validation set
  • Implemented a leaf health classifier with hough line and circle transform
  • Working with Intel Realsense to get leaf surface area computations for biomass estimation

Results

Wheat Head Detector

Trained on google colab, using tensorflow 2 for edge devices. Uses transfer learning based on SSD Mobilenet.

Wheat Head Detector

Frame Extractor

Python script extracts frames from our robot's video feed based on the speed of the robot.