Research On Mechanism Of AI Temperature Measurement Based On Visible Image And Method Optimization

Heating is a common fault type of electrical equipment.The faults such as aging insulation,overload and virtual connection can be reflected by the temperature.In the aspect of image temperature measurement,infrared image based on thermal radiation can realize the measurement from low temperature to high temperature,but the spatial resolution is low;CCD visible image has high spatial resolution,but it can only be used for high temperature measurement.The room temperature measurement based on light reflection mechanism is only limited to laser source.There is little research on the room temperature measurement based on sunlight images.With the progress of optoelectronic devices and the development of artificial intelligence,high-pixel visible images can accurately reflect the color information of objects.Machine learning can effectively fit the mapping model from the complex nonlinear relationship between input features and output,which makes it possible to measure the temperature of objects at room temperature using visible images.In this paper,the temperature measurement mechanism of thermal modulated light reflection at room temperature based on visible image is discussed,and the metal plate temperature prediction model based on k-Nearest Neighbor machine learning algorithm is optimized.The specific achievements are as follows:1)An intelligent temperature measurement platform based on visible image was built,and the image library of common metal materials copper plate and aluminum plate with different illumination,different temperature and different image formats was established by a digital camera,with a total of 6000 images.2)Based on the mechanism of thermoreflectance,a mechanism of full wavelength and multi incidence angle thermoreflectance based on visible image is proposed.With the blackbody radiation source as the research object,firstly,according to the mechanism of thermal radiation,the experimental results show that the visible image almost does not contain the thermal radiation temperature information.Secondly,according to the mechanism of light reflection temperature measurement,it was found that the visible image contains obvious temperature information when there is reflected light.Finally,according to the mechanism of thermoreflectance,a full wavelength multi incidence angle thermoreflectance temperature measurement mechanism based on visible image is proposed,and machine learning method is introduced to fit the complex mapping relationship between image chroma features and temperature.3)In the aspect of image feature optimization for machine learning,two optimization methods are proposed.The high-level color features of the RGB-GLH are extracted,and the Fisher criterion is proposed to select features.By reducing the feature dimension,the measurement error is effectively reduced.Secondly,the brightness information of the image is combined with the high-level color features,so that the temperature prediction model can reflect the change of illumination,and improve the accuracy of the model prediction.4)In the aspect of image format,the raw image is proposed as the basic data optimization method,which further improves the prediction accuracy and makes the measurement error reach 0.7 ℃.The low temperature measurement method based on visible image and machine learning proposed in this paper has the advantages of high accuracy,simple operation and low cost.It provides a new technical route for using visible band information to detect room temperature.