SBNPD
This project is a smoke detector that uses a very deep convolutional neural network architecture (VGG16) along with transfer learning and fully connected neural network for bounding box regression.
Created on 20th May 2023
•
SBNPD
This project is a smoke detector that uses a very deep convolutional neural network architecture (VGG16) along with transfer learning and fully connected neural network for bounding box regression.
The problem SBNPD solves
This smoke detector was created with two potential solutions in mind. The first solution was to detect wild fires that destroy and burn the forests every year through the footage of forest in which smoke is visible but not that easy to distinguish. The National Interagency Fire Center has documented an average of approximately 70,000 wildfires per year. If employed appropriately this model might be able to prevent wildfires.
The second potential solution is to detect warfare smokes while terrorists ambush a land. This can help the military to utilise AI while preparing their anti-terrorist tactics.
Challenges we ran into
The model was not training/learning anything until we performed tuning for several hours. Due to the depth of the VGG16 architecture we ran into "Resource Exhaustion Error" quite often and had to restart the whole training process. Initially we even faced difficulty in finding the dataset.