Yisti Vita Via, Eva Yulia Puspaningrum, Zain Muzadid Zamzani, Salamun Rohman Nudin, Siti Mukaromah, Afina Lina Nurlaili
Chili pepper (Capsicum frutescens L.) is a highvalue horticultural commodity whose quality determines its market price and consumer acceptance. Manual quality assessment remains subjective and inefficient, highlighting the need for an automated system. This study proposes a digital image processing approach for classifying chili pepper quality using the eXtreme Gradient Boosting (XGBoost) algorithm. Three feature types were extracted from chili pepper images: texture features using Local Binary Pattern (LBP) and Gray Level Co-occurrence Matrix (GLCM), and color features using Hue Saturation Value (HSV). A dataset of 1,200 primary images representing six quality categories (raw, half-ripe, ripe, dry, rotten, and anthracnose) was used. The images underwent preprocessing including background removal, resizing, and augmentation. Experimental results show that the combination of LBP, GLCM, and HSV features achieved the best performance, with an accuracy of 95.83 %, precision of 0.9586, recall of 0.9583, and F 1-score of 0.9581. The optimal XGBoost parameters were a learning rate of 0.1,100 estimators, and a maximum depth of 12. These findings demonstrate that integrating color and texture features effectively enhances chili pepper quality classification accuracy. © 2025 IEEE.
Universitas Pembangunan Nasional Veteran Jawa Timur, Informatics, Faculty of Computer Science, Surabaya, Indonesia; Universitas Negeri Surabaya, Informatics Management, Vocational Program, Surabaya, Indonesia
