Twin Support Vector Machine Based On Artificial Fish Swarm Algorithm And Its Application

How to detect the occurrence of early fires more accurately and in real time is the goal that human beings are constantly pursuing.One of the most critical problems in image fire detection is to use classifiers to distinguish and identify flames and other interferences.Classical Support Vector Machine(SVM)is the most practical method for flame image feature recognition.Although SVM has achieved good results in the field of flame recognition,SVM classification algorithm needs complicated redundant calculation when facing large-scale data,which may lead to slow classification speed and unable to meet the requirements of real scenes.Twin Support Vector Machines(TWSVM)is a better learning algorithm inspired by SVM theory.The training and learning speed of TWSVM has been greatly improved.But like SVM,the performance of TWSVM is closely related to its parameters.The difficulty of parameter selection will greatly limit the application of TWSVM in flame recognition.In this paper,the theory of TWSVM based on SVM with better performance is applied to the field of flame recognition,and the problem of parameter selection that affects the performance of TWSVM is deeply studied.The following are the main research contents:In the first part,an improved AFSA algorithm is proposed to select TWSVM parameters.The algorithm introduces Cauchy distribution in foraging behavior,realizes automatic step size adjustment in swarming and following behavior and elimination and regeneration mechanisms to obtain an improved fish swarm algorithm with better performance.Then,the improved artificial fish swarm algorithm is used to realize automatic parameter selection in TWSVM.Finally,the proposed algorithm is compared with traditional support vector machine(SVM),Grid-TWSVM,GA-TWSVM,PSO-TWSVM,FOA-TWSVM,GSO-TWSVM,AFSA-TWSVM using UCI(university of California,Irvine)standard data set.In the second part,after analyzing the characteristics of different color space models of flame images,a RGB-YCbCr mixed color space model method is proposed to segment flame images to obtain flame target regions.Then the typical low-order color moments in color features and four uncorrelated quadratic statistics in gray level co-occurrence matrix are extracted as the input feature vector of twin support vector machine.On the basis of flame characteristics,the proposed TWSVM based on improved artificial fish swarm algorithm is applied to the actual flame data set.Simulation experiments show that TWSVM is better than SVM in the field of flame recognition.The proposed improved artificial fish swarm algorithm can jump out of the local optimal solution at a faster speed and find the global optimal parameters suitable for TWSVM,thus solving the problems of difficult parameter selection and long optimization time of commonly used parameter optimization algorithms when TWSVM is applied to flame recognition.