Potential applications of Unmanned Aerial Vehicle multispectral imagery in vegetables

Abstract

A proof of concept project using UAV technology was conducted in conjunction with Rugby Farms, Gatton to assess the potential application of UAV crop sensing imagery in vegetable systems. The work focused on 2 key areas: detection of Sclerotinia in green beans using NDVI as an indicator of crop stress; and the development of predictive capacity for yield and final plant parameters. Green beans, lettuce, sweet corn and broccoli were the key crops studied.
Sclerotinia was evident in field sampling of green bean crops but not in the initial crop sensing data. Higher resolution flights at key crop stages might have been more successful. UAV NDVI imagery successfully identified spatial variability in all crops. Small plot yield assessments in green beans identified a 20% reduction in yield in lower NDVI areas (approximately 27% of the field).
In additions to identifying spatial variability, UAV NDVI imagery was also utilised to determine what predictive capacity could be developed for yield and final plant parameters across the experimental sites. The small scale assessments reported here confirm that early season variability is maintained through to maturity suggesting that early season data measurements could be used to predict final yield and plant characteristics.
Classification and automated counting process were applied to corn data as a mechanism for potentially predicting final yield. The automated corn counts were 10% less than manual counts. Automated counts of lettuce were 98% accurate relative to manual counts. Field measurements of plant parameters (plant and head diameter and final weights) were highly correlated with predicted harvest data (R2 0.608-0.915).
Given the project was a small scoping study, significant development would be required to further assess this application of technology in vegetable systems. This would include refinement of the algorithms and classification counting processes with significant larger datasets and rigorous validation and testing