An Approach Using Emerging Optical Technologies and Artificial Intelligence Brings New Markers to Evaluate Peanut Seed Quality

Abstract

Seeds of high physiological quality are defined by their superior germination capacity and uniform seedling establishment. Here, it was investigated whether multispectral images combined with machine learning models can efficiently categorize the quality of peanut seedlots. The seed quality from seven lots was assessed traditionally (seed weight, water content, germination, and vigor) and by multispectral images (area, length, width, brightness, chlorophyll fluorescence, anthocyanin, and reflectance: 365 to 970 nm). Seedlings from the seeds of each lot were evaluated for their photosynthetic capacity (fluorescence and chlorophyll index, F-0, F-m, and F-v/F-m) and stress indices (anthocyanin and NDVI). Artificial intelligence features (QDA method) applied to the data extracted from the seed images categorized lots with high and low quality. Higher levels of anthocyanin were found in the leaves of seedlings from low quality seeds. Therefore, this information is promising since the initial behavior of the seedlings reflected the quality of the seeds. The existence of new markers that effectively screen peanut seed quality was confirmed. The combination of physical properties (area, length, width, and coat brightness), pigments (chlorophyll fluorescence and anthocyanin), and light reflectance (660, 690, and 780 nm), is highly efficient to identify peanut seedlots with superior quality (98% accuracy).