roshankoirala / fire_detection_model

A CNN based fire detection model using TensorFlow (Keras) and transfer learning. Includes a Python script to scrap image data from the web.

Introduction

Fire is very useful discovery of the humanity. Like any other human invention, mis-handeling of fire can cause huge damage to the humanity and nature. Every year, urban fire results huge life and property damage. Similary the tragic loss of natural resources in uncontroleld wildfire is well known. The control of wildfire has becoming a huge challenge by using the treditional technologies. On the other hand, recently, the advancement in technology and especially the machine learning have benifited the society a lot in diverse aspects, ranging from self driving car to cancer research. Hence, it can surely contribute to the technology to early detect the fire so that we can react it earlier before getting it worse and making a lots of damage or being it out of control. We can train a deep neural network to distinguish fire and non-fire situation with very good accuracy. This technology aided with suitable hardware design can be vary useful to minimize the fire hazard. In this work we train a convolutional neural network which can achieve out of sample accuracy of 97% classifying the fire and non-fire images.

List of content

This repository contains the following files

• A notebook about data preperation
• A notebook about VGG16 training
• Report writing