Rust Detection in Coffee Leaves Using Otsu's Method and Mathematical Morphology
Abstract
Coffee leaf rust is a major threat to global coffee production, causing significant yield and quality losses. Traditional detection methods rely on manual inspection, which is labor-intensive, time-consuming, and susceptible to human error. This study presents an automated system for detecting coffee leaf rust using image processing techniques, combining Otsu's thresholding method with mathematical morphology to accurately identify rust-affected areas in digital images. The methodology addresses challenges such as variability in image quality, noise, and segmentation accuracy. Using a dataset of manually annotated images, the system's performance was evaluated with the Dice coefficient, demonstrating its effectiveness. Finally, we compare the results with a method that uses genetic algorithms to achieve the same goal, obtaining good results despite using a much simpler process.
