Deep Learning Approaches for Image Segmentation and Object Counting in Plant Phenotyping
by Ian Stavness
Image-based plant phenotyping is being rapidly adopted in plant physiology and plant breeding research. Extracting phenotypic information from images of plants and crops remains a core challenge for the field. Deep learning approaches have shown promising initial results for meeting the challenge, particularly for outdoor images of plant and crops that are captured under highly variable conditions in terms of lighting, wind, and ground background. In this talk, I will present our submission to the 2017 CVPPP Leaf Counting Competition that involved convolutional neural networks for segmenting plant from background and for counting the number of leaves on rosettes. I will also discuss recent work applying similar techniques to outdoor images of crops.
Dr. Ian Stavness is an Associate Professor in Computer Science at the University of Saskatchewan. He leads the image and data analysis group for the Plant Phenotyping and Imaging Research Centre (P2IRC) at the University of Saskatchewan. He also directs the Biomedical Imaging & Graphics (BIG) lab focused on 3D modeling, image analysis and deep learning for biological and biomedical applications. Ian is an OpenSim Fellow in neuro-musculoskeletal modeling and simulation and its translation to rehabilitation medicine. He completed a post-doc at Stanford University with Scott Delp at the NIH Center for Biomedical Computation and his PhD on 3D biomechanical simulation at the University of British Columbia in 2011.