As a way of nondestructive testing, infrared thermography has been increasingly used in the field of building. However, it was found that temperature measured by thermographic camera will deviate from accurate values, and the experimental data in this paper show that the lower the temperature of target surface is, the larger the deviation amplitude will be. To solve this problem, two correction identification models based on RBF and BP neural network were, respectively, constructed with MATLAB. The temperature data measured by infrared thermography was used as input variable, while the data measured by thermocouple was used as output. Five types of building materials were selected as the testing targets. The results show that the identification accuracy of networks is related to the number of training samples, the more the training samples, the higher accuracy the RBF network will have. At last, one complete system which can correct the temperatures measured by infrared thermography based on RBF neural network was established.
展开▼
