Hyperspectral reflectance and machine learning to monitor legumes for feed quality

Authors: Flynn, Kyle; CHINMAYI, H - Oak Ridge Institute For Science And Education (ORISE); BAATH, GURJINDER - Texas Agrilife Research; LEE, TREY - University Of Oklahoma; Gowda, Prasanna; Northup, Brian; Ashworth, Amanda

Submitted to: American Geophysical Union Meeting Abstract
Publication Type: Abstract Only
Publication Acceptance Date: 10/4/2023
Publication Date: N/A
Citation: N/A

Interpretive Summary: Hyperspectral remote sensing is an innovative technology that is transforming precision agriculture by capturing detailed spectral data and offering valuable information to stakeholders in production. Here, we aimed to assess the quality of forage from three different legume varieties (soybean, tepary bean, and mothbean) using in-situ hyperspectral data. The data was analyzed using various machine learning techniques such as k-Nearest Neighbors (KNN), partial least squares regression (PLS), support vector machine (SVM), and random forest (RF) to predict nitrogen detergent fiber (NDF), acid detergent fiber (ADF), in vitro true digestibility (IVTD), and crude protein (CP). Additionally, the study explored the potential applications of a future hyperspectral satellite called CHIME by adapting the hyperspectral wavelengths to match the satellite's bands. Using these spectral datasets, we compared various models that included combinations of in-situ data, CHIME data, machine learning methods, legume type as an additional variable, and models that divided the data into separate legume type subsets. The results indicate that hyperspectral remote sensing is a valuable tool for both on-site measurements and potential satellite observations (e.g., CHIME). Among machine learning options, RF and SVM models consistently outperformed other models in predicting legume characteristics, with SVM being computationally more efficient than RF. Comparing different model approaches, building individual legume-specific models generally yielded better results than models incorporating legume type as a variable. These findings provide insights into the nutritional composition of legume forages, aiding in their selection and application in various agricultural contexts. Moreover, this research contributes to the advancement of precision agriculture and offers guidance for future developments in the field.

Technical Abstract: Hyperspectral remote sensing is an innovative technology that is transforming precision agriculture by capturing detailed spectral data and offering valuable information to stakeholders in production. Here, we aimed to assess the quality of forage from three different legume varieties (soybean, tepary bean, and mothbean) using in-situ hyperspectral data. The data was analyzed using various machine learning techniques such as k-Nearest Neighbors (KNN), partial least squares regression (PLS), support vector machine (SVM), and random forest (RF) to predict nitrogen detergent fiber (NDF), acid detergent fiber (ADF), in vitro true digestibility (IVTD), and crude protein (CP). Additionally, the study explored the potential applications of a future hyperspectral satellite called CHIME by adapting the hyperspectral wavelengths to match the satellite's bands. Using these spectral datasets, we compared various models that included combinations of in-situ data, CHIME data, machine learning methods, legume type as an additional variable, and models that divided the data into separate legume type subsets. The results indicate that hyperspectral remote sensing is a valuable tool for both on-site measurements and potential satellite observations (e.g., CHIME). Among machine learning options, RF and SVM models consistently outperformed other models in predicting legume characteristics, with SVM being computationally more efficient than RF. Comparing different model approaches, building individual legume-specific models generally yielded better results than models incorporating legume type as a variable. These findings provide insights into the nutritional composition of legume forages, aiding in their selection and application in various agricultural contexts. Moreover, this research contributes to the advancement of precision agriculture and offers guidance for future developments in the field.