Exploring a cost-effective way for nutrient management with machine learning for container plants

Providing timely references of plant nutrient conditions is essential for container plant production with improved fertilizer use efficiency and reduced environmental waste. Imaging systems, as an effective real-time plant monitoring (e.g., canopy area, leaf color change) approach, are less tested in container crops. We imposed four types of greenhouse container plants (basil, marigold, pepper, and sage) with five fertilization input rates based on nitrogen, and found that plant chlorophyll content, growth index, canopy area, and biomass were gradually increased along with increased fertilizer rates. Due to the close relationships between plant canopy color values and growth parameters, an RGB imaging system using ImageJ to process the image analysis with a supervised machine learning approach was used to classify fertilization input status with four levels: extreme-underuse, sub-underuse, sufficient, and overuse. By training the data using K-Nearest Neighbor (KNN), Support Vector Machine (SVM), and Naïve Bayes algorithms, this system could reach up to 0.7–1.0 accuracy in guiding fertilization schedules during the plant vegetative growth stage. This imaging system with machine learning models offered a cost-effective, real-time monitoring, and decision-making control platform for precision nutrient management for greenhouse plants.