Automated Classification of Indonesian Candi Structures Using Advanced Machine Learning Models: YOLOv7, YOLOv8, and DeepSORT

Indonesia is a culturally rich country, with converging tradition and technology mainly through integrating AI and computer vision. This research investigates advanced deep learning models for the detection and tracking of Indonesian temple structures using the YOLOv7, YOLOv8, and DeepSORT methods. The models are trained and evaluated for their effectiveness in accurately identifying and classifying those very cultural landmarks, done on a unique dataset of temple images. The results indicate that both the YOLOv7 and YOLOv8 models performed very similarly with regard to precision, recall, accuracy, and the F1 score, thus proving that improvements to the architecture in YOLOv8 did not greatly improve this scenario. However, this was very effective when integrated into DeepSORT for the robust maintenance of consistent obj ect identification and tracking across video frames under changes in different conditions. Future work may be focused on training models with a more balanced dataset to mitigate class-based limitations and further experiments with other pre-trained models to improve the model's accuracy and application potential in real-world cultural heritage preservation.