F.P. Rochim, A. Nugroho, M.I. Ardiansyah
Being in the Ring of Fire area makes Indonesia prone to volcanic eruptions. In 2022, a total of 253 volcanic eruptions were recorded in Indonesia. However, out of the 127 active volcanoes in Indonesia, only 69 are monitored by the Centre for Volcanology and Geological Disaster Mitigation (PVMBG). Innovation is needed to facilitate the monitoring process of active volcanoes that are not yet observed. This research aims to develop a volcanic eruption detection model using surveillance camera images that can be placed around volcano monitoring posts. A 233 volcanic eruption images dataset was trained to detect and segment volcanic eruption objects using the You Only Look Once (YOLOv8) algorithm. Model evaluation was conducted using the Confusion Matrix method, resulting in % an accuracy rate of 94.3%. The trained model achieved precision levels of 0.96 and 0.95 for segmenting volcano and volcanic cloud eruption objects, respectively. The recall values for the segmented objects were 0.93 and 0.904, respectively. Based on the model's test results, it has generally been able to detect volcanic eruptions and segment volcano and volcanic cloud eruption objects well, with a high prediction box value. However, in segmenting moving volcanic cloud eruption objects, the model has not maintained this prediction box value. © Published under licence by IOP Publishing Ltd.
Faculty of Engineering, Universitas Negeri Semarang, Indonesia
