Abstract: Aiming at the problem of temperature measurement error caused by distance measurement and visual angle change during UAV infrared patrol inspection of composite insulators, this paper proposes a method to improve the accuracy of infrared temperature measurement and ensure the accuracy and reliability of defect identification. Based on the built 110 kV composite insulator infrared temperature rise experimental platform, the system collects infrared image data at different shooting distances and angles, and uses the real temperature measured by the optical fiber temperature sensor as the calibration benchmark to build an infrared temperature measurement error sample set. Based on the data set, the radial basis function (RBF) neural network is used to establish the error correction model of infrared temperature measurement, and realize the nonlinear correction of the original infrared temperature measurement value. The results show that the average temperature difference of infrared thermometer after RBF model correction is reduced to 0.19℃ (the relative error is 2.07%), which is better than linear, nonlinear and exponential regression methods (the average temperature difference is 0.37~0.82℃), and has stronger generalization ability. This method effectively improves the temperature measurement accuracy of UAV infrared patrol inspections, and provides reliable technical support for the condition assessment of insulators.

Key words: composite insulator, infrared thermometry, RBF neural network, temperature correction