<![CDATA[Manual determination of oil palm fruit maturity remains a significant challenge for farmers, leading to inefficiencies and reduced harvest quality. This research aims to design and implement a computer vision application based on Convolutional Neural Network (CNN) to automatically, objectively, and efficiently detect oil palm fruit maturity. The methodology involves web-based system development using the Waterfall model, integrating a modified YOLOv5s architecture enhanced with a Convolutional Block Attention Module (CBAM), C3 Transformer (C3TR), and Bidirectional Feature Pyramid Network (BiFPN). A dataset of 4160 oil palm fruit images underwent auto-orient, resize, color normalization, and data augmentation, then split into 70% training, 20% validation, and 10% test sets. Model training was conducted on Google Colab using an Nvidia A100 GPU for 21.36 minutes. Evaluation results demonstrate very high performance with a Precision of 0.9858, Recall of 0.9910, mAP@0.5 of 0.9941, and mAP@0.5:0.95 of 0.8544. The Confusion Matrix and Precision-Recall curve indicate excellent classification accuracy and strong model generalization capabilities. The web-based system allows farmers to upload images or videos and use a webcam for real-time detection. This application is expected to assist farmers in identifying oil palm fruit maturity more quickly and objectively, minimizing errors, and improving harvest quality and efficiency.]]>
Copyrights © 2026
