Multispectral Image Analysis for Crop Health Monitoring System

Abstract

The goal of this research is to apply machine learning to classify healthy and unhealthy potato crops collected from UAV-based multispectral images, and to establish which spectral band provides the best separation for classification. Traditional detection and mapping approaches take time, involve a lot of human work, and are often subjective. The classification will use the Random Forest Classifier as the machine learning technique to classify based on two vegetation indices: the Normalized Difference Vegetation Index (NDVI) and the Red Edge Normalized Difference Vegetation Index (NDRE). The proposed method includes three primary components: (1) raw picture radiometric correction and orthomosaic combination; (2) dirt and weed removal using a thresholding method; and (3) classification and model training using Random Forest Classifier. The method’s performance is assessed using data from an experimental potato field published by the University of Idaho.