Twin support vector machine based on improved artificial fish swarm algorithm with application to flame recognition

Abstract

In this paper, a twin support vector machine (TWSVM) based on improved artificial fish swarm algorithm (IAFSA) for fire flame recognition is proposed in view of the large computation burden and slow classification speed of the traditional support vector machine (SVM). Twin support vector machine is a machine learning algorithm developing from standard support vector machine. However, twin support vector machine cannot deal with the parameter selection problem well. The difficulty of parameter selection may greatly restrict the application of TWSVM in flame recognition problem. So a novel artificial fish swarm algorithm (AFSA) is used to solve the parameter selection problem of TWSVM. In order to make up for the drawbacks of the basic AFSA, the chaotic transformation is first applied to initialize the position of artificial fish swarm since it may be non-uniformly initialized in the basic artificial fish swarm algorithm. Then, the Cauchy mutation is used to make the fish swarm jump out of the local optimal solution after continuously expanding the visual scope of the artificial fish during the foraging procedure. An adaptively step-size adjusting method is then developed to optimize the moving steps of the swarming and following behaviors in order to accelerate the convergence speed of the developed algorithm. Last, to further improve the efficiency and accuracy of the algorithm, an elimination and regeneration mechanism based on adaptive t-distribution mutation is utilized to update the artificial fish swarm at each iterative procedure. Experimental results show that the TWSVM algorithm based on improved artificial fish swarm algorithm is a more effective method to identify the flame and greatly improves the accuracy and real-time performance of the flame recognition compared with PSO-TWSVM, Grid-TWSVM, GA-TWSVM, FOA-TWSVM, GSO-TWSVM, AFSA-TWSVM and the traditional SVM.