Spinel: a framework for counting wheat spikes and kernels using UAV and ground-based imaging in breeding fields
| dc.contributor.advisor | Bais, Abdul | |
| dc.contributor.author | Mohammed, Ahmed | |
| dc.contributor.committeemember | Paranjape, Raman | |
| dc.date.accessioned | 2024-11-08T20:32:47Z | |
| dc.date.available | 2024-11-08T20:32:47Z | |
| dc.date.issued | 2024-08 | |
| dc.description | A Thesis Submitted to the Faculty of Graduate Studies and Research In Partial Fulfillment of the Requirements for the Degree of Master of Applied Science in Electronic Systems Engineering, University of Regina. xiii, 112 p. | |
| dc.description.abstract | Quantifying wheat spikes and kernels is crucial for evaluating the grain yield potential of diverse breeding lines. It enables breeders to measure grain yield and select superior breeding lines to improve crop productivity. Traditional methods for assessing these traits are labour-intensive, error-prone, and manual. Despite the widespread use of deep learning (DL) techniques in recent studies, there remains a gap in applying these methods to provide practical, quantified analyses for breeding field applications. This thesis introduces SPINEL (SPike and kerNEL), a comprehensive framework that combines unmanned aerial vehicle (UAV)-captured multispectral imaging and fieldcaptured RGB camera imaging for advanced phenotyping. The framework employs three distinct YOLOv8 models tailored to specific detection tasks. UAV-based imaging and infield imaging through mobile cameras are used to address the different scales of imaging required: UAV imaging for whole-field analysis and infield imaging for detailed kernel detection. Multispectral imaging offers more bands than traditional RGB and has proven effective in estimating plant stress and nitrogen content, though its use in spike counting is novel. The first YOLOv8 model focuses on plot detection, the second on spike detection using UAV-captured multispectral images, and the third on detecting spikes and kernels in field-captured RGB images. These models demonstrated high accuracy with mean average precision (mAP) scores of 95%, 86%, and 85%, respectively, indicating robustness in images with high spike density and diverse backgrounds. By integrating data with geolocation information from the multispectral images, SPINEL provides a comprehensive visual representation of spike count and average kernel per spike for each field plot, enabling breeders to assess spike-per-plot and kernel-per-spike traits efficiently. The SPINEL framework addresses the limitations of current methods by offering a precise, automated solution for phenotyping in wheat breeding, facilitating better decisionmaking for crop improvement. | |
| dc.description.authorstatus | Student | en |
| dc.description.peerreview | yes | en |
| dc.identifier.uri | https://hdl.handle.net/10294/16517 | |
| dc.language.iso | en | en |
| dc.publisher | Faculty of Graduate Studies and Research, University of Regina | en |
| dc.title | Spinel: a framework for counting wheat spikes and kernels using UAV and ground-based imaging in breeding fields | |
| dc.type | Thesis | en |
| thesis.degree.department | Faculty of Engineering and Applied Science | |
| thesis.degree.discipline | Engineering - Electronic Systems | |
| thesis.degree.grantor | University of Regina | en |
| thesis.degree.level | Master's | en |
| thesis.degree.name | Master of Applied Science (MASc) |