Research Assistant Professor Texas A&M University-Corpus Christi
Disclosure: Disclosure information not submitted.
Unmanned Aircraft Systems (UAS) and sensor technologies have made it possible to collect fine spatial and high temporal resolution data throughout the cropping season and provide accurate phenotypic data to assess overall crop growth, health status, and yield. The use of phenotypic data acquired by UAS combined with Artificial Intelligence (AI) algorithms in agriculture can provide useful data on disease detection and monitoring, variety selection, and yield prediction allowing in-season management. Although there are many general modelling strategies available for hyperparameter or model optimization, less studies were conducted regarding data preparation and variable selection particularly in agricultural context. In this study, UAS-based phenotypic data over tomato open field was used to validate the effectiveness of UAS data using machine learning approach. USA. Multi-temporal UAS data (Level 0 data) was collected using RGB and multispectral cameras with significant overlap at 30-50m flying altitude above ground. Level 1 data, Orthomosaic images and Digital Surface Models (DSM), was generated from raw images (Level 0 data by using a Structure from Motion (SfM) algorithm with Ground Control Points (GCPs). Level 2 phenotypic data such as canopy cover, canopy height, and vegetation indices was extracted from the processed Level 1 data. Multi-temporal Level 2 data were used to generate crop growth curve and its first derivative (growth rate curve) followed by calculation of Level 3 phenotypic data, i.e., maximum growth rate, date at maximum growth rate, duration of growth, etc. The phenotypic data acquired from Level 2 to 3 will be selectively input into an AI model to verify which subset of phenotypic data are more useful to estimate tomato yield. It is expected that this presentation will demonstrate the usefulness of UAS-based phenotypic data and AI algorithms for tomato yield prediction.