Publication: Using deep learning to detect citrus trees

A new publication is now open access in Drones, a new MDPI journal. This is the first open peer-review publication for this journal, meaning that you can access the review reports and the responses to reviewers.

This paper is a result of a great collaboration that was a result of my research stay at UC Berkeley in 2017, where this idea and work started.

Csillik, O., Cherbini, J., Johnson, R., Lyons, A., Kelly, M., 2018Identification of Citrus Trees from Unmanned Aerial Vehicle Imagery Using Convolutional Neural Networks, Drones, 2(4), 39.

The paper uses a combination of a convolutional neural network with object-based post-processing using superpixels to identify citrus trees from images collected by drones. The workflow was developed in Trimble’s eCognition Developer software by using the newly available CNN features and SLIC superpixel segmentation. Our approach reached an accuracy of over 96% in identification of citrus trees in a study area in San Joaquin Valley, California, and proved to be a viable approach in automating the creation of trees inventory. Check the paper for more details!

2 thoughts on “Publication: Using deep learning to detect citrus trees

  1. Dear Sir,

    This is with reference to recent Trimble Webinar “Deep Learning, UAVs & Precision Agriculture”. I am much interested to apply this approach in my area of interest to ascertain its results. I will be grateful if you could please share ruleset for further insight in my self learning initiative.

    Looking forward,

    Nasir Hameed

    1. Hello Nasir,
      The ruleset shown during the webinar was similar to what you can find on the free eCognition community website. You will need an account to access many rulesets and data sets. (
      Let me know if you need more info on that.

Leave a Comment

Your email address will not be published.