Application of Deep Learning for Fig Fruit Detection in The Wild

This paper proposes an appropriate detection approach based on a deep learning methodology to accomplish fast and effective detection and localization of fig fruits in complicated environmental pictures, e.g., in the wild. A dataset is manually created by first taking pictures of fig fruits in their natural environment and then annotating them with the label "buah tin." The augmentation technique was then employed to prevent overfitting during the training phase. Thus, the dataset can be split into 30% for testing and 70% for training. The YOLOv3 and YOLOv4 one-stage detection deep learning models will be used in the Phyton environment with Google Colaboratory for this research. The experimental outcomes indicated that the YOLOv4 model-based recognition algorithm for detecting fig fruit had outperformed YOLOv3 in terms of average precision and other metrics. Compared to YOLOv3’s mAP value of 81.40%, YOLOv4’s is the highest at 90.02%. Thus, the YOLOv4 can fulfil essential practical criteria and has a good detecting impact on figs in a challenging environment.