Development of phenotyping system using low altitude UAV imagery and deep learning

In this study, a lightweight phenotyping system that combined the advantages of both deep learning-based panicle detection and photogrammetry based on light consumer-level UAVs was proposed. A two-year experiment conducted to perform data collection accuracy validation. learning model, named Mask Region-based Convolutional Neural Network (Mask R-CNN), trained detect panicles in complex scenes paddy fields. total 13 857 images were fed into R-CNN, with 80% used for training 20% Scores, precision, recall, Average Precision (AP), F1-score 82.46%, 80.60%, 79.46%, 79.66%, respectively. complete workflow proposed preprocess flight trajectories remove repeated noises. Eventually, evident changed rice growth during heading stage visualized geographic distributions, number predicted before harvest. The average error amounts 33.98%. Experimental results showed feasibility using developed as high-throughput approach. Keywords: detection, vision-based phenotyping, learning, unmanned aerial vehicle (UAV) DOI: 10.25165/j.ijabe.20211401.6025 Citation: Lyu S X, Noguchi N, Ospina R, Kishima Y. Development low altitude UAV imagery learning. Int J Agric & Biol Eng, 2021; 14(1): 207–215.