Location: Application Technology Research
Project Number: 5082-21620-001-024-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Mar 1, 2024
End Date: Feb 28, 2025
Objective:
Build a model with digital image data from a commercial grade stereo vision system in order to determine agronomic data for assisting decision making processes that optimize pesticide uses for targeted applications.
Approach:
Digital RGB and depth images of specialty crop canopies e.g., orange, blueberry and strawberry will be collected with a commercial grade stereo vision system under various field conditions. The system will be mounted on a tractor or a similar platform and will capture crop canopy development data. Different ground truth agronomic data e.g., growth stages, crop size and volume, blooming, and nutrition inputs will be acquired by manual assessment. Both data types, ground truth data and images, will be used to develop mathematical/machine learning models to predict crop data from digital images. Multiple approaches will be taken to develop the models. The accuracy of these models will be validated with the other half of the data and images. After the validation, an apparatus, including a computer program with the models and a stereo vision system, will be developed to predict the agronomic data from the digital images in real time, which will be tested under field conditions. The influences of the environmental conditions in the model performance will be evaluated to build robust models under a wide range of outdoor conditions.
