Temperature Recognition for Normal Temperature Metal Based on The Statistical Features of Visible Image and K-nearest Neighbor Algorithm

The temperature status monitoring of power equipment is very important to ensure the safe operation of the power grid, and the fault temperature is generally in the normal temperature range. Under normal temperature conditions, the magnitude of the thermal radiation intensity in the visible waveband is small and changes slightly with temperature. Therefore, infrared detection is generally used to detect temperature. In this paper, the chromaticity information of visible images of the copper plate and aluminum plate caused by normal temperature changes is studied. First, establish image libraries of copper plate and aluminum plate at 10°C-100°C, extract the chromaticity gray values of the R, G, and B components of the images at different temperatures, and calculate the frequencies of each gray level to obtain the gray frequency distribution of each component. Then, the change law of the gray frequency distribution curve with temperature is analyzed qualitatively, and the statistical features of the gray frequency distribution of each component are calculated. Some of features are selected by the Fisher criterion. Finally, the k-Nearest Neighbor (KNN) algorithm is used for temperature recognition whose input features is a combination of the selected features. The results show that the average test error of the KNN temperature prediction model is within 1°C, which achieves a good prediction effect, and the dimension of the input feature can influence the prediction effect. The above results provide a new technical route for detecting normal temperature using visible image information.