Fire-Smoke-Detection

Detecting Fire, Smoke using Computer Vision, Open CV and PyTorch

Early fire/smoke detection plays a very important role in protecting many lives also property loss can be reduced and downtime for the operation minimized through early detection. Therefore in this project I have developed an Computer Vision & Deep Learning pipeline for fire and smoke detection.

Demo Output -

Download the Dataset - download

Dataset Folder -

Train
    - Fire
    - Neutral
    - Smoke       
Test
    - Fire
    - Neutral
    - Smoke
    
Dataset contains 1000 images of each class

Model Structure -

For traing the model I have used transfer learning technique. Architecture used here is ResNet50 which is pretrained on ImageNet dataset. I have achieved validation accuracy of 93% using ResNet. For more info about training and graphs - open Training.ipynb

Sample Results

Steps -

1. Clone/Download the repo
2. Download the dataset
3. For training - open Training.ipnb
4. For inference - open Inference.ipynb

Requirements -

Python3

PyTorch

OpenCV

Matplotlib

Numpy

Upcoming Work -

RestAPI (rest api using flask)

References

1. PyImageSearch - https://www.pyimagesearch.com/2019/11/18/fire-and-smoke-detection-with-keras-and-deep-learning/
2. DeepQuestAI/Fire-Smoke-Dataset - https://github.com/DeepQuestAI/Fire-Smoke-Dataset