Home>Results
Advanced Search
Indexed by:
Abstract:
Fire is one of the most common hazards in the process industry. Timely and accurate fire detection is essential. The camera-based technics for fire detection are one of the promising technologies. However, its uncertainty of fire monitoring quality, such as noise artifacts within digital images caused by the inherent interference of hot environments, is always a key defect hindering the further application of this technology. Taking a simple fire scenario of the cable fire as an example, the noise reduction model (SA-DCGAN, Spatial Attention-Deep Convolution Generative Adversarial Network) is discussed for three kinds of typical fire image noise (white, black and mottled). Compared with traditional noise reduction algorithm, the model has greater advantages in restoring flame profile and texture. Through the verification process of applying this method in promoting fire detection based on image recognition, the effectiveness of the theoretical model is confirmed in improving the detection accuracy. It shows that the "True Detection" is increased by 375% and the "Missed Detection" and "False Detection" are decreased by 54% and 587%, respectively. These results show that the proposed theoretical model is of great significance for improving camera-based fire detection performance in thermal environments, which makes possible to further promote the intelligent fire protection in the process industry.
Keyword:
Reprint 's Address:
Version:
Source :
JOURNAL OF THERMAL ANALYSIS AND CALORIMETRY
ISSN: 1388-6150
Year: 2022
Issue: 3
Volume: 148
Page: 1191-1199
4 . 4
JCR@2022
3 . 0 0 0
JCR@2023
ESI Discipline: CHEMISTRY;
ESI HC Threshold:74
JCR Journal Grade:1
CAS Journal Grade:3
Affiliated Colleges: