Machine Learning Techniques for Early Disease Detection in Potato Plants
Open Access
- Author:
- Ryan, Philip John
- Area of Honors:
- Electrical Engineering
- Degree:
- Bachelor of Science
- Document Type:
- Thesis
- Thesis Supervisors:
- Zhiwen Liu, Thesis Supervisor
Dr. Julio V. Urbina, Thesis Honors Advisor - Keywords:
- machine learning
unsupervised learning
principal component analysis
PCA
optics - Abstract:
- The standard method of diagnosing early blight disease in potato plants is a visual inspection. This system is insufficient because potato plants can become contagious before they show visual symptoms of disease. Optical spectroscopy may prove to be a superior alternative; however, computer programming is needed to determine plant health from measurements. This thesis proposes a software solution to be used with mobile optical sensing equipment from ATOPTIX, Inc. The solution uses principal component analysis (PCA) to reduce the dimensions of the data. The two principal components that best model the data are plotted and the centroid of the resulting data cluster is computed. For this experiment, specific rows of potato plants were directly inoculated with early blight; these were the spreader rows. The disease then spread to the remaining rows, which were marked by orange flags. In just one day after inoculation, the percent difference between the healthy cluster of data and the diseased cluster of data exceeded 250% for the spreaders. For the remaining plants, the percent difference between the healthy and diseased clusters exceeded 100% three days after the spreaders were inoculated. Based on these results, PCA could be an effective way of determining plant health, especially if used in conjunction with other machine learning techniques.