Performing Department
(N/A)
Non Technical Summary
The Central Coast region of California faces significant challenges in weed management within specialty crop systems like strawberries, broccoli, and leafy green vegetables. Traditional methods, including manual weeding and robotic weeders based on shape differences between crops and weeds, are increasingly uneconomical due to labor shortages and high costs. This project proposes a novel approach: leveraging Hyperspectral Imaging (HSI) in combination with machine learning to revolutionize weed control in agricultural settings. This technique uses specialized cameras to measure light reflection amongplant species as a way to differentiate weeds from crops.HSI offers a significant advantage over conventional methods of weed management. It is effective even in complex scenarios like partial visibility of weedsor size variations, and is adaptable to different planting methods. Our objective is to harness HSI's potential, creating an extensive dataset of hyperspectral and digital images. This dataset will focus on weeds coexisting with leafy green specialty crops, particularly those that host economically damaging diseases.
Animal Health Component
(N/A)
Research Effort Categories
Basic
30%
Applied
(N/A)
Developmental
70%
Goals / Objectives
Goal 1. Development and Validation of Machine Learning Models: Our first goal is to develop robust machine learning models that utilize HSI data for accurate and efficient weed identification. These models will be trained and validated using the collected dataset, ensuringtheir effectiveness across various growth stages of weeds. This approach is expected to markedly improve weed control strategies, thereby enhancing crop health and productivity.Goal 2. Comprehensive Data Collection and Analysis: The second goalinvolves the systematic collection of hyperspectral images of weeds, capturing their growth stages and conditions. This data will be instrumental in constructing machine learning models. The analysis will include identifying key spectral signatures unique to different weed species, especially those harboring INSV. This extensive collection will form a valuable resource for ongoing and future research in agricultural technology.Goal 3. Student Training and Collaborative Engagement: The final goal focuses on the educational and collaborative aspects of the project. Students, particularly from the Agricultural Plant and Soil Science B.S. program, will be actively involved in data collection, model development, and AI applications. This hands-on experience will prepare them for future roles in the agricultural sector, such as Pest Control Advisors and research specialists. Additionally, the project will foster collaboration between academia and industry professionals, promoting knowledge exchange and innovation in agricultural practices.
Project Methods
We propose the curation of a dataset of images chronicling the growth stages from seedling to fully grown plants of three weed species that are major hosts of INSV throughout the region. Specifically, we will focus our efforts on the following species: little mallow (Malvaparviflora), annual sowthistle (Sonchus oleraceus), and common purslane (Portulaca oleracea). These species were selected because they have been shown by our USDA ARS and Cooperative Extension counterparts to be major hosts of INSV. These species are also highly competitive with crops and have been shown to depress yields if not adequately controlled. An inventory of tagged images for the entire life cycle of these weeds will offer immense opportunities for agricultural weed management in the immediate timeframe for:(1)development of automated weed termination, (2)targeted spraying of weeds by automated sprayers, and (3) reduction of the chemical waste due to excess spraying of herbicides. We propose to acquire images using multiple devices and conditions. The techniques for imagecapture are: (1) Pictures in the field using cell phone cameras. (2) Pictures under laboratory conditions with off-the-shelf consumer cameras (3) Hyperspectral images using a hyperspectral camera.This project would occur in the major leafy greens production region of the Central Coast, including Monterey, Santa Clara, Santa Cruz, and San Benito Counties. The project PI and Co-PI are based 15 minutes from Salinas, CA, also known as the "Salad bowl of the world." TheSalinas Valley is a major leafy greens production region that provides ample opportunities for weed collection. The Co-PI has strong work relationships with professionals in the local agricultural private sector, University of California Cooperative Extension, and USDAAgricultural Research Service, all of which have committed to providing access to fields for the project. The Co-PI additionally has the availability of on-campus greenhouse space forconducting research. Weeds will also be grown from seed under these conditions and theninoculated with INSV so that pictures can be taken of infected plants at different life stages. This step is important because it can be difficult to find infected weed species in the field at all their different growth stages. Weeds grown in the greenhouse will supplement weeds collected from the field. There also will be cases in which field collected weeds will be transplanted to pots to place in the greenhouse for photographing using off the shelf and hyperspectral cameras. The hyperspectral camera instrumentation, image capture facilities, and a dedicated workstation to store and analyze these images will be set up in the PI's laboratory space.In our project, we are committed to incorporating best practices in HSI. We will utilize standard types of HSI cameras and techniques to ensure high-quality, reliable data collection. Our approach includes using spatial scanning for detailed line-by-line analysis, spectral scanning for precise wavelength-specific imaging, and non-scanning methods for efficient, comprehensive scene capture. Byadhering to these established methods, we aim to optimize the accuracy and effectiveness of our HSI picture taking, ensuring our data is of the highest standard for our machine learning applications in weed identification and agricultural research. The PI and Co-PI will engage students from their programs in this research, offering practical training in HSI instrument operation and fieldwork for weed identification and collection in the Central Coast region. Participants will gain skills in data management, organization, and machinelearning, with a focus on AI. They will prepare and share data on public platforms like Zenodo and GitHub. Many involved students are from the Agricultural Plant and Soil Science B.S. program, aspiring to roles in the local agricultural sector as Pest Control Advisors, researchspecialists, or cooperative extension agents. This project enables collaboration with a wide array of professionals in pest and weed scouting, mirroring post-graduation employment.