Thesis
Automated Pruning Decisions in Dormant Sweet Cherry Canopies using Instance Segmentation
Washington State University
Master of Science (MS), Washington State University
01/2022
DOI:
https://doi.org/10.7273/000004557
Handle:
https://hdl.handle.net/2376/125378
Abstract
Pruning is a perennial orchard operation vital to orchard health, fruit yield, and fruit quality. However, pruning is laborious, requiring substantial human resources; with more than 40 hours required per acre of cherries. As a result, there is great interest in automated pruning for modern tree fruit orchards. Any automated pruning system must possess robust machine vision capable of making accurate pruning decisions in the complex orchard environment. Deep neural networks are powerful tools in developing robust machine vision systems for orchard environments, and herein I demonstrate how deep convolutional neural networks can be used in an automated pruning system. One of the fundamental pruning rules in the Upright Fruiting Offshoots (UFO) architecture is to remove vigorous (i.e., large diameter) leaders. My objectives were to: 1) develop an instance segmentation approach to detect leaders and estimate leader diameter in the UFO architecture, and 2) compare the performance of instance segmentation networks trained with active and natural lighting images. Stereo images of dormant sweet cherry (Prunus avium L.) trees trained to the UFO architecture were collected using active and natural lighting. Images were annotated for two classes of objects—trunks (horizontal base) and leaders. Two instance segmentation networks (Mask R-CNN) were trained to detect leaders—one using active lighting images and one using natural lighting images. Deep stereo matching enabled generation of synthetic images to increase the size of the training dataset, and large learning rates were employed to accelerate learning (called super-convergence training). Predictions from the active and natural lighting Mask R-CNNs were compared to ground truth annotations for mask IoU, precision, recall, and accuracy of correctly identifying the largest diameter leader in an image. The active lighting Mask R-CNN demonstrated 11% greater mask IoU, 8% greater precision, 15% greater recall, and 22% greater accuracy of selecting the largest diameter leader than the natural lighting Mask R-CNN. Overall, the active lighting Mask R-CNN correctly identified the largest diameter leader in 94% of test images. My results indicate that instance segmentation is a robust approach to make automated pruning decisions in the UFO architecture.
Metrics
7 File views/ downloads
26 Record Views
Details
- Title
- Automated Pruning Decisions in Dormant Sweet Cherry Canopies using Instance Segmentation
- Creators
- Daniel Borrenpohl
- Contributors
- Manoj Karkee (Advisor)Qin Zhang (Committee Member)Matthew Whiting (Committee Member)Abhisesh Silwal (Committee Member)
- Awarding Institution
- Washington State University
- Academic Unit
- Biological Systems Engineering, Department of
- Theses and Dissertations
- Master of Science (MS), Washington State University
- Publisher
- Washington State University
- Number of pages
- 83
- Identifiers
- OCLC#: 1371068597; 99900898639701842
- Language
- English
- Resource Type
- Thesis