Background: Unmanned aerial vehicles (UAVs) equipped with lightweight sensors are making a significant impact in field-based crop phenotyping. UAV platforms have been successfully deployed to acquire phenotypic data in a precise and efficient manner that would otherwise be time-consuming and costly to acquire when undertaken through manual assessment. One example is the estimation of plant density (or counts) in field experiments. Challenges posed to digital plant counting models are heterogenous germination and mixed growth stages that are present in field experiments with diverse genotypes. Here we describe, using safflower as an example, a method based on template matching for seedling count estimation at early mixed growth stages using UAV imagery.
Results: An object-based image analysis algorithm based on template matching was developed for safflower seedling detection at early mixed growth stages in field experiments conducted in 2017 and 2018. Seedling detection was successful when tested using a grouped template type with 10 subgroups representing safflower at 2-4 leaves growth stage in 100 selected plots from the 2017 field experiment. The algorithm was validated for 300 plots each from the 2017 and 2018 field experiments, where estimated seedling counts correlated closely with manual counting; R2 = 0.87, MAE = 8.18, RSME = 9.38 for 2017 field experiment and R2 = 0.86, MAE = 9.16, RSME = 10.51 for 2018.
Conclusion: A method for safflower seedling count at early mixed growth stages using UAV imagery was developed and validated. The model performed well across heterogenous growth stages and has the potential to be used for plant density estimation across various crop species.
Keywords: Object-based image analysis; Plant phenotyping; Safflower; Seedling count; Unmanned aerial vehicle.