Yield Modelling and Harvest Scheduling of Snap-bean Using Remote Sensing: A Greenhouse Study

It is estimated that the global population will grow by 25% by 2050, which will result in an increased demand for food production. Farmers therefore are required to efficiently make use of crop lands, while maximizing yields. Precision agriculture mitigates the expense of resource utilization toward optimized productivity. This study aims to model yield and assess crop maturity of snap-bean crops via remote sensing and precision agriculture. A greenhouse study was conducted to investigate the proposed objectives using an in situ spectroradiometer (400-2500 nm), along with coincident plant physical attributes (e.g., width, height, etc.) at a high temporal resolution of 25 measurements over the growing season, for 48 plants. A best-case scenario, enclosed setup with two artificial lights was used. Harvest maturity was assessed using machine learning approaches to classify plant growth at different stages, e.g., vegetative growth, budding, flowering, and pod formation, along with pod maturity classification (six pod maturity classes). Yield was investigated utilizing a PLSR-based method to identify the period in which yield prediction is most accurate. Our results show that one can classify between different stages of plant growth using only spectral information with an accuracy ranging from 83-94%, while using both spectral and physical attributes results in an improvement of the accuracy range to 85-98%. Future research will emphasize the mentioned yield modelling and pod maturity classification objectives. Finally, this study mandates further work to assess the applicability of the mentioned concepts to a scaled-up, field-level experiment using unmanned aerial systems (UAS) or airborne data.