Session: General Poster:

Edamame has been widely consumed in China and Japan for centuries as a snack or side dish. In the US, the consumption of edamame has been increasing, and edamame has become the second most highly consumed soy-food after soymilk because it is nutritious and has potential health benefits. The edamame harvest window is narrow but critical for producing high-quality beans. Current methods for determining the optimal harvest time of edamame rely on the ability of experienced edamame growers to detect these changes visually, by touch or taste. These determination methods can be quite subjective and pose a major obstacle for inexperienced growers in the US. Therefore, to foster growth of the domestic edamame industry, a rapid, consistent, and standardized method for determining optimal harvest time is desired. This study aims to understand the physical and chemical property changes of edamame over bean development (R5 to R7 stages) and develop a spectroscopy-based machine learning technique to rapidly predict optimal harvest time. The results showed that pod weight, bean weight, and pod thickness peaked at R6 and remained stable for the remainder of bean development. Sugar, starch, alanine, and glycine contents also peaked at R6 but proceeded to decline thereafter. Meanwhile, fat, fiber and ash contents remained low at the R6 stage. Based on the physical and chemical properties, the harvested edamame samples was labeled as ‘early’, ‘ready’ or ‘late’ to indicate early, ready or late to harvest. The machine learning method based on the spectral reflectance of pods had a high accuracy of 0.95 for classifying “early” and “late” harvested edamame and 0.87 for classifying “early” and “ready” harvested edamame. This work would provide a platform technology for developing rapid and accurate prediction of optimum harvest time of edamame, which is essential to ensure to produce consistent and high-quality edamame for the market.

Justin Polk

Appearances