June 1, 2016Open Access
Deep Residual Learning for Image Recognition
Key Points
- To develop an autonomous triggering system for high-speed cameras to record atmospheric electrical discharges without requiring manual human operation.
- Engineered an image processing pipeline incorporating segmentation, representation, description, and classification methods.
- Trained and evaluated a computer vision model using the EfficientNetB4 convolutional neural network architecture.
- Embedded the image processing algorithms into a Raspberry Pi 3 Model B+ single-board computer for real-time testing.
- Offline image processing achieved an accuracy of 81.81%, compared to 71.63% obtained by the EfficientNetB4 model.
- Real-time implementation on the Raspberry Pi 3 Model B+ platform attained an accuracy of 86.95%.
- Assessment of discharge multiplicity (counting subsequent return strokes in the flash channel) reached an accuracy of 66.66%.
Authors
Discussion
No takes yet. Share an insight, caveat, or question.
Implication
Validation study demonstrates automated lightning detection using computer vision on embedded hardware, highlighting viable autonomous triggers for high-speed cameras.
Cite This Study
He et al. (2016) studied this question.