Temperature Soft Measurement Using Color Based On Neural Network

Temperature is a very important parameter in scientific research and engineering field,therefore the technique of high precision temperature measurement is extremely important. According to the correspondence between the color of the high-temperature object in visible light band and the temperature,a method of temperature measurement by collecting the color of the high-temperature object with a digital camera is proposed,which is based on neural network.In this thesis,the first step,gather electric stove image samples through experiment and establish standard image database;secondly pre-process the image samples:using the method of median filtering to eliminating the image noise,using the method of threshold segmentation on image samples to segment background and using average data to gain the color of image; thirdly use massive sample data to train both neural networks (BP neural network and RBF neural network),and approximate the nonlinear relationship between the information value of color(Red,Green and Blue) and the temperature of object;The last step,apply this method to the temperature measurement. The analysis resulted from massive experimental data indicated that,the method is very feasible,which express a high precision of temperature measurement and is within acceptable precision scope,especially the method is simple and low cost. Compared with BP neural network,RBF neural network is more easily designed, obviously faster training speed and the higher precision.