Research On Radiation Temperature Measuring Technology Based On SVM

Thermometry is very important in many fields such as industrial production andscientific studies. The traditional thermometry method uses thermal sensitive resisters tomeasure the temperature by contacting the object, which is low price and easy to operate.However, it is hard to measure high temperature or the moving object. The sensors can changetemperature distribution of the detected object and easy to be corroded and wasted. Due tothese such restrictions on the application of conventional thermometry, the research ofnon-contact thermometry method is significant especially. Support vector machine (SVM) is anew type of machine learning algorithms that is based on statistical theory, which has beenused in intelligent modeling and system identification widely since its excellent nonlinearfitting and generalization ability. This paper attempts to apply SVM theory to the infraredradiation temperature measurement system, which provides a new feasible thermometrymethod for high temperature and complex industrial environment.(1) The purpose of the radiation temperature measurement and its development wasillustrated at the beginning of the papers. Then some common problems among currentradiation temperature measurements were pointed out, and a number of importantcorresponding methods were listed. Because the colorimetric temperature measurement is themost important way of radiation temperature measurement, we discuss the details about itsprinciples, advantages and disadvantages, and prove the feasibility of the artificial intelligencein radiation temperature measurement.(2) Continuous and sufficient research about the basic theory of SVM and the specificalgorithm in regression estimation model was done for further discussion. After studying thetemperature measurement problems in complex industry environment, this paper proposesthat the nonlinear function relationship between the color and temperature of measured objectcan be used to establish the regression model.(3) By an exhaustive study of support vector regression (SVR) theory and its structureparameters selection, this paper presents an improved SVR algorithm which is based on the SOR (Successive over relaxation) method in the classification problem. The improved modelneeds the less support vectors (SVs) and it has high prediction accuracy, generalization abilityand learns faster. Next, the genetic algorithm was used to tune the parameters according to theimpact of the parameters on the performance of SVM model. Finally, a series of experimentsand simulations have been done and the results proved the feasibility of three-channelradiation temperature measurement which based on SVM.