Estimating cotton yield in breeder plots using unmanned aerial vehical (UAV) imagery

Traditionally, cotton breeding programs have depended on mechanical harvest of small breeding plots to evaluate cultivars for lint yield. The primary limiting factor in testing of breeding lines is the time and equipment costs associated with harvest; large increases in a testing programs plot-load are not possible without additional investment. UAVs may provide a method to evaluate early-generation progeny rows and increase the overall volume of material evaluated without additional investment in harvest equipment. Unmanned Aerial Vehicles (UAVs) are commonly used for high-resolution imagery and have become popular for cotton (Gossypium hirsutum L.) phenotyping. Advancements in UAV imagery and image analysis may enable cotton researchers to advance cultivars based on UAV imagery yield estimates. This study was designed to evaluate methods of cotton lint yield estimation using UAV imagery collected before mechanical harvest. This study was conducted using data from three cotton breeding regions, Coastal Bend, High Plains, and Rolling Plains. Images were classified for pixel counts of lint, and boll counts were acquired by counting each contiguous group of lint pixels. Boll counts were found to be more closely correlated with harvested yield and visual ratings, suggesting that UAV boll counting methods may be appropriate for large-scale field breeding trials. UAV yield estimates were further enhanced when data was limited to analysis within pedigrees, suggesting that aerial imagery can be useful for advancement of early-generation cultivars as an alternative to traditional mechanical plot harvest.