Skip to content

Codes Used on My Thesis Work - Fire and Smoke Detection using Fully Supervised Training Methods and Search by QuadTree ( FireFront Project )

Notifications You must be signed in to change notification settings

g0nzal0rd/FireFront-Detect

Repository files navigation

FireFront-Detect

Codes Used on My Thesis Work - Supervised Methods on Fire and Smoke detection ( FireFront Project ). All the codes produced are ment to be executed on Google Colab environment using Python3.

On first part of the project I worked on a global image classifier using different networks. The objective was to detect if a certain image contains fire or not, not giving any information about the localization of that fire.

Networks used:

AlexNet: code

SqueezeNet: code

Then I moved on to segmentation networks in which I used fully supervised training methods to train. The networks are able to segment the given images in areas of fire and non-fire ( binarization )

Networks used:

U-Net: code

Dataset created still to be published...

In order to solve the multi-scale problem of detection we use a Quad-Tree algorithm to dynamically slice the input images into smaller patches to help the network detect smaller regions of the phenomenon. The next code file includes the Quad-Tree implementation and the calculation for the system performance:

Global System: code

The saved parameters for both networks, for both classes, can be downloaded below:

SqueezeNet Fire : code

SqueezeNet Smoke : code

U-Net Fire: code

U-Net Smoke: code

About

Codes Used on My Thesis Work - Fire and Smoke Detection using Fully Supervised Training Methods and Search by QuadTree ( FireFront Project )

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published