Yield estimation in cotton using UAV-based multi-sensor imagery

Monitoring crop development and accurately estimating crop yield are important to improve field management and crop production. This study aimed to evaluate the performance of an unmanned aerial vehicle (UAV)-based remote sensing system in cotton yield estimation. A UAV system, equipped with an RGB camera, a multispectral camera, and an infrared thermal camera, was used to acquire images of a cotton field at two growth stages (flowering growth stage and shortly before harvest). Sequential images from the three cameras were processed to generate orthomosaic images and a digital surface model (DSM), which were registered to the georeferenced yield data acquired by a yield monitor mounted on a harvester. Eight image features were extracted, including normalised difference vegetation index (NDVI), green normalised difference vegetation index (GNDVI), triangular greenness index (TGI), a channel in CIE-LAB colour space (a∗), canopy cover, plant height (PH), canopy temperature, and cotton fibre index (CFI). Models were developed to evaluate the accuracy of each image feature for yield estimation. Results show that PH and CFI were the best single features for cotton yield estimation, both with R2 = 0.90. The combination of PH and CFI, PH and a∗, or PH and temperature were the best two-feature models with R2 from 0.92 to 0.94. The best three-feature models were among the combinations of PH, CFI, temperature and a∗. This study found that UAV-based images collected during the flowering growth stage and/or shortly before harvest were able to estimate cotton yield accurately.