A SVM approach for vessel fire detection based on image processing

Based on the features such as high convergence rate and global optimization of Support Vector Machine (SVM) which follows structure risk minimization principle, a method of fire detection is proposed, in which the shape of bright areas are analyzed by SVM and results are produced. After collecting images of fire and interference source under different conditions, data of shape features are extracted. Many of them are used as training set and delivered to SVM; and other data are used as testing set for pattern recognition. Fire experiments show that trained SVM with RBF kernel and SMO algorithm can recognize images with high accuracy.