Developing a Labeled Dataset for Chili Plant Health Monitoring: A Multispectral Image Segmentation Approach with YOLOv8

This research explores a new method for assessing chili plant health using multispectral camera imagery and deep learning-based image segmentation. Data from Bale Tatanen Universitas Padjadjaran chili farms were used to create a labeled dataset with four health categories. Initial image processing involved FastSAM inference to generate binary masks, followed by training a YOLOv8 model for improved segmentation accuracy. This model enabled NDVI calculation and automatic health labeling in unseen images, contributing to automated chili plant health monitoring systems. Evaluation showed an average Dice Coefficient of 63.80%, indicating moderate overlap between predicted and true masks. High precision and robust segmentation across various conditions were observed, though improvements are needed in fine detail segmentation. This approach enhances understanding of chili plant health and supports further studies in the field.