Abstract: Lodging in agricultural crops is the permanent displacement of a plant from its upright position . It may be caused by several weather and environmental conditions. Harvesting severely lodged crops may take twice as much time and results in reduced yield. Plant breeders seek to identify and select for lodging-resistant varieties. Currently lodging in plant breeding trials is assessed by manual observation. This can be both time consuming and subjective. The objective of this work is to automate detection of lodging in breeding plots using drone imagery. Texture features were extracted from drone images of wheat and canola breeding plots using Gray Level Co-occurrence Matrix (GLCM), local binary patterns (LBP) and Gabor filters. For each type of feature, a Support Vector Machine (SVM) was trained and used to predict if plots were lodged. The GLCM features exhibited the best performance, achieving 96.0% prediction accuracy for canola and 92.6% accuracy for wheat. Qualitative visualizations were constructed to show spatial distribution across and within plots. This system will be of assistance to breeders for selecting lodgingresistant varieties.
Authors: Sajith Rajapaksa, Mark Eramian, Hema Duddu, Menglu Wang, Steve Shirtliffe, Seungbum Ryu, Anique Josuttes, Ti Zhang, Sally Vail Curtis Pozniak, Isobel Parkin
Associations: Dept. of Comput. Sci., Univ. of Saskatchewan, Saskatoon, SK, Canada