<![CDATA[Indonesia is home to over 120 active volcanoes, many of which pose significant risks to surrounding communities and global atmospheric conditions. The rapid and accurate identification of volcanic objects remains a challenging task for effective volcano monitoring and disaster mitigation. To enable rapid and accurate automatic identification of volcanic objects, this study leverages two Deep Learning models focused on instance segmentation. The two instance segmentation models, You Only Look Once version 7 (YOLOv7) and Mask Region-Based Convolutional Neural Network (Mask R-CNN), were evaluated for their effectiveness in segmenting volcanic objects, namely lava flows, volcanic ash, mountains, and vegetation, from volcanic imagery and videos. A dataset of 140 labeled volcanic images was used for training and validation, while video data were employed to assess model performance under real-time conditions. Both models were trained using transfer learning with pre-trained MS COCO weights and their performance was measured using mean Average Precision (mAP). The results indicate that YOLOv7 achieves higher accuracy (mAP50 = 0.872) along with faster training and inference times, while Mask R-CNN obtains lower accuracy (mAP50 = 0.535) but consistently produces higher confidence scores (>0.90). These findings suggest that YOLOv7 is more suitable for real-time volcano monitoring, while Mask R-CNN is preferable in applications where high detection confidence is prioritized. The results imply that deep learning–based instance segmentation can support automated volcano monitoring systems and enhance disaster mitigation efforts through rapid identification of critical volcanic features.]]>
Copyrights © 2026
