Faculty Mentor Information
Dr. John Shovic (Mentor), University of Idaho
Additional Funding Sources
This award was funded by an OUR Semester (or SURF) award 2024.
Abstract
This research, conducted under Project Evergreen, aims to develop a computer vision-based solution to detect weeds in conifer sapling rows within a United States Forest Service (USFS) nursery. The presence of weeds in these rows significantly impacts seedling viability, posing a challenge to reforestation efforts. Leveraging advanced machine vision techniques, this study aims to enhance the precision and efficiency of weed management practices by identifying weeds in the field, thereby improving seedling survival rates. The YOLO v8 model was selected for its superior real-time capabilities and high accuracy. YOLO (You Only Look Once) is a single-pass convolutional neural network known for its speed and precision in object detection. The research process included collecting and annotating 729 images, model training, and validation. Though the size of the dataset is small, initial results demonstrate that the YOLO v8 model can effectively detect weeds among crop rows, with an F1-score of 0.36 showing moderate performance. Though not ready to implement in a weed detection system, the model demonstrates promising learning trends for improvement through further dataset augmentation, hyperparameter tuning, and an expanded dataset.
Project Evergreen: AI Weed Detection in a USFS Conifer Nursery
This research, conducted under Project Evergreen, aims to develop a computer vision-based solution to detect weeds in conifer sapling rows within a United States Forest Service (USFS) nursery. The presence of weeds in these rows significantly impacts seedling viability, posing a challenge to reforestation efforts. Leveraging advanced machine vision techniques, this study aims to enhance the precision and efficiency of weed management practices by identifying weeds in the field, thereby improving seedling survival rates. The YOLO v8 model was selected for its superior real-time capabilities and high accuracy. YOLO (You Only Look Once) is a single-pass convolutional neural network known for its speed and precision in object detection. The research process included collecting and annotating 729 images, model training, and validation. Though the size of the dataset is small, initial results demonstrate that the YOLO v8 model can effectively detect weeds among crop rows, with an F1-score of 0.36 showing moderate performance. Though not ready to implement in a weed detection system, the model demonstrates promising learning trends for improvement through further dataset augmentation, hyperparameter tuning, and an expanded dataset.