Date of Award
8-2022
Document Type
Thesis
Degree Name
Master of Science (MS)
Department
Plant and Environmental Science
Committee Chair/Advisor
Joe Mari Maja
Committee Member
Juan Carlos Melgar
Committee Member
Haibo Liu
Committee Member
Matthew Cutulle
Abstract
The use of hyperspectral imaging is a valuable piece of technology used in precision agriculture. Hyperspectral imaging system could help farmers and researchers alike in analyzing tissue samples quicker compared to the current way of running a plant tissue analysis on the different nutrient levels that often takes time in generating results. The objective of this research was to measure the potassium level of peach trees using its leaves with a hyperspectral camera. The Senop HSC-2 was used to capture images of peach tree leaves during the fall of 2020 and 2021. The collected data were used to create a model using Principal Component Analysis (PCA) to determine the hyperspectral camera’s suitability in detecting the three nutrient levels (high, medium, and low) of peach tree leaves and Partial Least Square (PLS) regression to predict the exact concentration of K nutrients. Four pre-treatment methods were applied to the raw data analysis to address the multiple shifts in the plot due to the varying schedules of data collection. Multiplicative Scatter Correction (MSC), Savitzky-Golay in the first derivative (SGolay-1), Savitzky-Golay in the second derivative (SGolay-2), and Standard Normal Variate (SNV) were applied to the raw data.
Results from this research show that the R2 values of the pre-treatment methods were 0.8099, 0.6723, 0.5586, and 0.8446 respectively. Results show that the SNV prediction model had the highest R2 and lowest RMSE. SNV was used to predict the K nutrient for validation of the data. The result shows a lower R2 of 0.8101 compared to the training samples. Samples were submitted to the Clemson Agricultural Service Laboratory for nutrient sampling immediately after images of the samples were captured using the hyperspectral camera. Based on the result of the PCA, the model yielded a result of 93% of the total variability adequately described in two Principal Components (PC). The number of components increased to 6 in PLS using the SNV method due to the prediction of the different nutrient amounts of the high, medium and low potassium levels.
Recommended Citation
Abenina, Megan Io Ariadne Salimbangon, "Potassium Level Prediction of Peach Tree Leaves Using a Hyperspectral Camera and Multivariate Analysis" (2022). All Theses. 3867.
https://open.clemson.edu/all_theses/3867