Abstract
Remote sensing-based onion yield prediction is an effective method for stabilizing yield and streamlining work plans.
Onion leaves are elongated and erect, resulting in low crop coverage in aerial images. Conventional methods make it difficult
to obtain accurate growth information. To address this, we devised a method to reduce the effects of weeds and soil by
implementing mask processing to exclude areas with Normalized Difference Vegetation Index (NDVI) values below 0.2 in
NDVI images. Hence, high correlations were obtained between NDVI and biomass index, which was calculated by
multiplying grass height and leaf sheath diameter. Furthermore, using high correlations between the biomass index and
NDVI, we successfully developed a prediction model with an average RMS error of 816 kg/(10 a).
