Yield forecasting is a common technique utilized to predict the amount of fruit expected at harvest. Orchard managers forecast yield to predict future packaging requirements, labor requirements, and to make agricultural decisions to help improve future yields. In order to forecast yield, one must first count the number of fruit on a representative sample of trees. Next, one must use a model to predict the total yield expected given the number of fruit counted. However, as population and labor costs continue to increase, a need for automation grows. While research has explored automated yield forecasting for various fruits, there currently isn't any research on automated avocado detection/forecasting. This project explored various methods to automate avocado detection in an orchard setting using computer vision. Additionally, this project constructed a model to predict the yield of avocados at harvest when after counting the current number of avocados earlier in the year. The computer vision pipeline plans to utilizes both thermal images and visible RGB images to make an avocado classifier. However, this system has currently shows the potential of thermal-based avocado detection at various times of the day, segmenting all avocados from the background. Next, this project will continue to utilize visible RGB images to further eliminate the background.
This work is licensed under a Creative Commons Attribution 4.0 International License.