A wildfire is an uncontrolled fire. Every year, wildfire causes significant destruction of huge forest land, loss of animal and human lives, and wildlife food. Eearly detection of fire can significantly shorten the reaction time. The longer it takes to locate a fire, the harder it is to contain for fire staff.
This is the submission for lets-stop-wildfires-hackathon-2.0 to early detect WildFire smoke conducted by AI For Mankind - a nonprofit organization.
Thanks to AIForManKind for providing Quick Start Demo and providing labeled smoke Image Data Set.
Also special thanks to HPWREN for providing access to HPWREN camera images.
Submitted fined tune model is trained with EfficientDet-d3 using TensorFlow.
Data Set - 737 images. After augmenting (Horizontal Flip and brightness), dataset used was :-
Training Images : 1739
Validation Images : 111
Total training steps : 107000
Saved model can be downloaded from https://drive.google.com/drive/folders/1R54ZCvD9-aNc-q59ZxUK_go9wO5qJKku?usp=sharingv
See Model Training notebook to do train model on smoke images.
For inference from saved model, refer to inference notebook
WildFire Resources
- FUEGO Wildfire Detection Slides by Kinshuk Govil
- Wildland Fire Assessment System
- How Wildfire Works
- Wildland Fire: What is Hazard Fuel Reduction?
Tensorflow Resources
Other Resources
- Faster RCNN ResNnet
- Train EfficientDet in TensorFlow
- Data Augmentation using roboflow
- Train object detection with Keras
- Google Colab for training
-
Explored SSD Mobile to solve this problem, but in results we found some limitations with some pattern of images. Training was very slow.
-
FatserRCNN ResNet101 - Got best accuracy and lowest loss with this. But it was giving many False Positive for Fog images test.
-
Faster_rcnn_inception_resnet_v2_atrous_coco also gave good results for true positives but the prediction time is very high and do not solve False Postives problem(predicting fog as smoke)
Anil, Khyati, Krishna and Rama Revuri
