Research And Implementation Of Multi-spectral Imaging Device For Oil Leakage Of Power

Transformer oil in power oil-filled equipment has many important functions.Oil leakage will affect the normal operation of power equipment and even damage the equipment.For a long time,power inspectors have been troubled by problems such as complicated oil leakage inspection procedures and operational hazards during the production process of substations.They need equipment to assist in the detection of oil leakage and evaluation of leakage situation.Electric power inspectors put forward the requirements of portable,realtime,strong applicability,high accuracy and intuitive results for oil leakage detection equipment.However,the existing equipment and methods cannot take these into consideration.Therefore,there is an urgent need for a device that meets these requirements.In this thesis,some progresses have been made in design and implementation on oil leakage detection equipment,including:1)A fast calculation method for the polarization image of the oil spill scene is proposed,which speed-up the calculation of linear polarization decomposition of a single polarization image by more than 4 times comparing with the least squares decomposition.From the characteristics of linear polarization of transformer oil,the method simplifies the description formula of the light reflected by the transformer oil,and then calculates the linear polarization component of the image by the method of matrix calculation.Based on the rapid decomposition method,a method is further implemented for detecting oil leakage based on polarized images and visible light images.It uses classic image processing algorithms to process linear polarization images and ultraviolet(UV)images,so that oil leakage can be effectively detected without being affected by natural light conditions.The accuracy rate is higher than 95%,and the detection rate is higher than 85%.2)A multi-view UV-visible image fusion method for oil leakage scenes is proposed,which fuses the UV-visible images collected by sensors at different locations in the actual scene,and eliminates the parallax between these images.It can retain the scene detailed information in the visible light image,and highlight the oil leakage information in the UV image.This method uses a multi-scale cascaded feature extraction network to extract image features.A graph neural network-based fusion mechanism for feature fusion is used,and finally reconstructs the features to obtain the fusion result.This method has achieved better performance in terms of the subjective effect of eliminating parallax,retaining effective information,and various objective evaluation indicators of image fusion,which is better than state-of-the-art algorithms in this fields.3)A multi-spectral imaging device for oil leakage in power equipment is designed and implemented.In order to meet the requirements of portability,an embedded development board is used as the computing unit of the imager to integrate the image fusion,area estimation,and 3D reconstruction modules of the oil leakage area.However,the computing resources of the development board are very limited.Under these conditions,in order to meet the requirements of real-time,strong applicability and high accuracy,the fast decomposition method of linear polarization images of the oil leakage scene is used in the module to reduce the calculation time,and the oil leakage detection is used based on the polarization image and the ultraviolet image.In order to visualize the detection results,a multi-spectral image fusion module,an oil leakage area estimation module,and a display and interaction module are designed and implemented.In the end,the device can meet all the requirements put forward by electric power inspectors,and passed the acceptance of scientific and technological achievements of Yunnan Electric Power Research Institute.