Deep Learning Approach for Identifying Fresh and Rotten Chili Peppers using YOLO

Abstract

Chili peppers are a vital agricultural commodity, yet post-harvest quality assessment primarily relies on manual inspection, which is subjective, labor-intensive, and prone to inconsistency. This study proposes a deep learning-based computer vision system using the You Only Look Once (YOLO) framework to automate the identification and classification of fresh and rotten chili peppers. The model was trained using a dataset of 400 web-crawled images, annotated and augmented to handle visual diversity. A novel evaluation strategy was employed using synthetic image generation to simulate real-world scenarios, including neatly arranged grids and randomly distributed objects with varying orientations. Experimental results demonstrate that the proposed model effectively localizes and classifies chili peppers with confidence scores ranging from 0.55 to 0.90. The system successfully distinguishes between fresh and rotten categories even under conditions of intra-class variation, such as discoloration and shriveling. These findings validate the robustness of the YOLO-based approach, offering a promising, efficient solution for automated post-harvest quality control and smart agriculture applications