Autonomous Controlled Pollination of Kiwifruit
Format
SOECS Senior Project Demonstration
Faculty Mentor Name
Cherian Mathews
Faculty Mentor Department
Electrical and Computer Engineering
Additional Faculty Mentor Name
Vivek K. Pallipuram
Additional Faculty Mentor Department
Electrical and Computer Engineering
Abstract/Artist Statement
Physiological characteristics of the Kiwifruit has led to a widespread adoption of controlled pollination, a process that involves humans manually applying pollen to flowers. There exist several controlled pollination methods, however, all of them are either extremely labor intensive or inefficiently utilize the expensive pollen mixture. Thus, we propose an autonomous robotic system which can detect flowers needing pollination in real-time and then deliver the pollen to the flowers in a controllable and repeatable manner. Additionally, the system produces valuable data about the orchard which we capture and store for external processing.
To achieve flower recognition, we make use of state-of-the-art computer vision techniques. Specifically, we re-train the Inception-v3 Convolutional Neural Network on more than 1000 images from Kiwifruit orchards to achieve a validation accuracy of 76.2% and a testing accuracy of 92.5%. The flower detection system triggers eight different solenoids which control pollen emission through eight diffusers. Using computational fluid dynamics, we design diffusers which accurately project pollen onto the orchard canopy, and thus minimize waste. In an effort to make our system user-friendly, all of the electronics onboard the system are connected to a local area network, allowing them to be configurable from an iOS application running on an Apple iPad.
Our preliminary testing in an artificial environment provides reasons for optimism, as our system successfully identifies flowers in printed images of a real orchard canopy. With help from our industry partner, Antles Pollen, the system will be tested in real kiwifruit orchard soon.
Location
School of Engineering & Computer Science
Start Date
5-5-2018 3:30 PM
End Date
5-5-2018 4:30 PM
Autonomous Controlled Pollination of Kiwifruit
School of Engineering & Computer Science
Physiological characteristics of the Kiwifruit has led to a widespread adoption of controlled pollination, a process that involves humans manually applying pollen to flowers. There exist several controlled pollination methods, however, all of them are either extremely labor intensive or inefficiently utilize the expensive pollen mixture. Thus, we propose an autonomous robotic system which can detect flowers needing pollination in real-time and then deliver the pollen to the flowers in a controllable and repeatable manner. Additionally, the system produces valuable data about the orchard which we capture and store for external processing.
To achieve flower recognition, we make use of state-of-the-art computer vision techniques. Specifically, we re-train the Inception-v3 Convolutional Neural Network on more than 1000 images from Kiwifruit orchards to achieve a validation accuracy of 76.2% and a testing accuracy of 92.5%. The flower detection system triggers eight different solenoids which control pollen emission through eight diffusers. Using computational fluid dynamics, we design diffusers which accurately project pollen onto the orchard canopy, and thus minimize waste. In an effort to make our system user-friendly, all of the electronics onboard the system are connected to a local area network, allowing them to be configurable from an iOS application running on an Apple iPad.
Our preliminary testing in an artificial environment provides reasons for optimism, as our system successfully identifies flowers in printed images of a real orchard canopy. With help from our industry partner, Antles Pollen, the system will be tested in real kiwifruit orchard soon.