<![CDATA[This study investigates the performance of two state-of-the-art object detection models, YOLOv8 and RT-DETR, in identifying diseases in chili plants, which represent a major challenge affecting horticultural productivity. Diseases such as anthracnose and Cercospora leaf spot often cause significant yield losses, and traditional manual identification tends to be inefficient, subjective, and error-prone due to the visual similarities found among disease symptoms. The objective of this research is to evaluate and compare the capabilities of both models using the Chili dataset from Roboflow Universe consisting of four classes: Anthracnose, Cercospora Leaf Spot, Healthy Fruit, and Healthy Leaf. The methodology includes data preprocessing, training using identical hyperparameters, and performance evaluation through accuracy and model behavior analysis during real-world testing. The findings indicate that RT-DETR achieves higher accuracy in controlled testing, reaching 90% for Anthracnose, 95% for Healthy Leaf, 100% for Healthy Fruit, and 85% for Cercospora Leaf Spot, supported by its transformer-based architecture that enhances spatial understanding. However, YOLOv8 demonstrates superior stability and consistency in real-world scenarios involving varying lighting, leaf orientations, and natural texture variations. The model also produces fewer misclassification errors, making it more reliable for practical field deployment. The implications of these results show that YOLOv8 is the most suitable model for integration into a Streamlit-based application due to its fast, responsive, and accurate inference, supporting early disease detection for chili farmers.]]>
Copyrights © 2025
