Explosion Detection using Ensemble Learning [Slides]
Abstract not provided.
Engineering topics
Publications and source records attributed to Takazawa, Samuel Kei.
Abstract not provided.
Seismoacoustic signatures from a chemical explosion was recorded on two seismometers and eight smartphone accelerometers at Idaho National Laboratory (INL). The captured waveform’s propagation and its time frequency representations are compared. The smartphone accelerometers were able to pick up a clear seismoacoustic explosion signal at within 1 km (150 m/kg 1/3 ) and a noisy signal up to 5.3 km (790 m/kg 1/3 ).
Poster on Transfer Learning for Explosion Detection on Smartphones
Improved Smartphone Network Deployment Methods for Explosion Collection Samuel Kei Takazawa1, Sarah Popenhagen1, Milton A. Garcés1, Luis Ocampo Giraldo2, Jay Hix2 1University of Hawai?i, 2Idaho National Laboratory Milton A. Garcés, milton@isla.hawaii.edu ETI – Consortium for Enabling Technologies and Innovation: Director, Anna Erickson, Ph.D.
University Program Review (UPR) 2023 Meeting Comparative Analysis of Explosion Signals on Smartphone and Legacy Infrasound Microphones MTV – Consortium for Monitoring, Technology, and Verification
Presentation on explosion source localization
These slides are for a presentation to the MTV consortium
The acoustic data of explosions and noise collected on smartphones were fed to the YAMNet model to obtain the scores of each class. The data was split 60/20/20 for training, validation, and testing. A single-layered neural network was trained using the computed scores to predict if the data was an explosion or noise. The trained model was then combined with the YAMNet model. The combined model was tested with the ESC-50 dataset to investigate overall accuracy and class based false positive rates.
The prompt detection of explosions is a key element of the nuclear non-proliferation mission. With traditional sensors being limited in number and scale, smartphones as compact and economical multi-modal sensors are gaining traction and are being deployed. To address the flood of heterogeneous smartphone data, our team proposes a feature extraction using standardized constant-Q frequency bands across acoustic, barometric, and accelerometer data. The work presented in this poster utilizes three explosions collected on smartphones with Idaho National Lab. More explosions are planned in the near future at INL and NNSS. These extracted features will be used in machine learning methods in future iterations.
The prompt detection of explosions is a key element of the nuclear non-proliferation mission. With traditional sensors being limited in number and scale, smartphones as compact and economical multi-modal sensors are gaining traction and are being deployed. To address the flood of heterogeneous smartphone data, our team proposes a feature extraction using standardized constant-Q frequency bands across acoustic, barometric, and accelerometer data. The work in this presentation contains data collected at Idaho National Laboratory during planned detonations.