Smart Camera for Volcano Eruption Early Warning System Based on Faster R-CNN and YOLO

This research uses two object detection algorithms, Faster R-CNN with ResNet50 backbone and YOLOv5, to develop an intelligent camera system for monitoring volcanic activities. The models were trained and evaluated using CCTV footage from Mount Semeru, a region prone to volcanic eruptions. Key performance metrics such as Precision, Recall, and mean Average Precision (mAP) were used to evaluate the performance of both models. The high precision numbers for YOLOv5 and Faster R-CNN show they are good at avoiding false positives, which is essential for volcanic monitoring. YOLOv5 has a precision of 83.2%, while Faster R-CNN is 84%. However, recall shows a more significant difference between the two models. Faster R-CNN has a recall of 82%, meaning it is better at detecting all relevant volcanic activities, even if that means catching a few false positives. The variations in performance can be attributed to their respective designs. YOLOv5 is designed to achieve rapid, real-time detection by simultaneously predicting bounding boxes and class probabilities. This approach enhances speed but may slightly reduce recall. Faster R-CNN uses a two-stage process, tending to be more accurate but can be slower and less flexible across different IoU thresholds. Its higher recall means it catches more objects, contributing to its lower mAP@50-95 since it could struggle with overlapping or varying-sized objects.