A New Approach to Vision-Based Fire Detection Using Statistical Features and Bayes Classifier

Duong, Ha Dai; Tinh, Dao Thanh

doi:10.1007/978-3-642-34859-4_33

Cited by 5 publications

(1 citation statement)

References 4 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The final flame detection was realized by the change of suspicious region in the two continuous images. Duong et al [6] proposed a method based on the changes of the statistical features in the flame region between different frames. The final result was obtained by Bayes classifier and defined as fire-alarm rate for each frame.…”

Section: Introductionmentioning

confidence: 99%

Combustion State Recognition of Flame Images Using Radial Chebyshev Moment Invariants Coupled with an IFA-WSVM Model

et al. 2018

View full text Add to dashboard Cite

Accurate combustion state recognition of flame images not only plays an important role in social security, but also contributes to increasing thermal efficiency and product quality. To improve the accuracy of feature extraction and achieve the combustion state recognition, a novel method based on radial Chebyshev moment invariants (RCMIs) and an improved firefly algorithm-wavelet support vector machine (IFA-WSVM) model is proposed. Firstly, the potential flame pixels and the potential flame contour are obtained in the pre-processing phase. Then, the rotation, translation and scaling (RTS) invariants of radial Chebyshev moments are derived. Combing the region and contour moments, the RCMIs of pre-processed and edge images are calculated to construct multi-feature vectors. To enhance the recognition performance, an IFA-WSVM model is built, where the IFA is applied to search the best parameters of WSVM. Then, the IFA-WSVM model is used to recognize the combustion state. Finally, the result for case studies show that the proposed method is superior to methods based on HMIs and ZMIs, achieving the highest rate of 99.07% in real time. The IFA algorithm also outperforms other benchmark algorithms. Even for the images transformed by RTS and small size of training sets, the proposed method continues to exhibit the best performance.

show abstract

Section: Introductionmentioning

confidence: 99%