Fire Monitoring and Assessment Platform (FireMAP): Object Identification in Images using Spectral and Spatial Cluster Analysis

Faculty Mentor Information

Dale Hamilton Barry Myers

Presentation Date

7-2016

Abstract

Wildland fires are destructive to properties and dangerous to people in close proximity, with the cost of some large fires exceeding $1 billion. The goal of the Fire Monitoring Assessment Platform (FireMAP) is to provide fire managers with the tools and knowledge for acquiring, analyzing, and managing hyper-resolution imagery to map burn severity in a faster, safer, and more affordable manner than is currently possible.

When imagery is taken it needs to be analysed for various kinds of information, but to start looking for details in the image and extrapolate data is a lot of work if it’s a raw image due to the huge dataset. By using clustering methods to group pixels spectrally and spatially we can detect objects in the image. This process also helps to normalize the image, reducing salt and pepper effects by either removing the data or absorbing it into nearby groups that it still represents. In identifying these objects we get the size, shape, type, and color, as well as additional object attributes, of groups of data for machine learning classification.

Comments

Poster #Th38

This document is currently not available here.

Share

COinS
 

Fire Monitoring and Assessment Platform (FireMAP): Object Identification in Images using Spectral and Spatial Cluster Analysis

Wildland fires are destructive to properties and dangerous to people in close proximity, with the cost of some large fires exceeding $1 billion. The goal of the Fire Monitoring Assessment Platform (FireMAP) is to provide fire managers with the tools and knowledge for acquiring, analyzing, and managing hyper-resolution imagery to map burn severity in a faster, safer, and more affordable manner than is currently possible.

When imagery is taken it needs to be analysed for various kinds of information, but to start looking for details in the image and extrapolate data is a lot of work if it’s a raw image due to the huge dataset. By using clustering methods to group pixels spectrally and spatially we can detect objects in the image. This process also helps to normalize the image, reducing salt and pepper effects by either removing the data or absorbing it into nearby groups that it still represents. In identifying these objects we get the size, shape, type, and color, as well as additional object attributes, of groups of data for machine learning classification.