Building a brain: how convolution neural networks can predict sprinkler activations

J. de Vries; Stanislav Kostka

doi:10.1117/12.2310130

14 May 2018 Building a brain: how convolution neural networks can predict sprinkler activations

J. de Vries, Stanislav Kostka

Proceedings Volume 10661, Thermosense: Thermal Infrared Applications XL; 106610E (2018) https://doi.org/10.1117/12.2310130
Event: SPIE Commercial + Scientific Sensing and Imaging, 2018, Orlando, FL, United States

Abstract

A convolutional neural network (CNN) was developed to recognize sprinkler activation based on long-wave infrared (LWIR) images, creating a nonintrusive, real-time model for detecting sprinkler activation. Training data were taken from ten large-scale fire tests with storage heights ranging between 4.6 m and 13.7 m and ceiling heights ranging between 6.1 m and 15.2 m. A sample of 25,000 LWIR images was extracted from the fire tests, split 70/30 between training/testing data. To prevent overfitting, the images were randomly reversed and cropped. The time required to train the model was reduced by 96% through GPU computing. The overall accuracy of the model was 99.7% for both pendent and upright sprinklers. The methodology described in this study can be generalized and applied to other image classification problems.

Conference Presentation

Citation Download Citation

J. de Vries and Stanislav Kostka "Building a brain: how convolution neural networks can predict sprinkler activations", Proc. SPIE 10661, Thermosense: Thermal Infrared Applications XL, 106610E (14 May 2018); https://doi.org/10.1117/12.2310130

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available