Research And Realization On Transformer Oil Level Detection Method Based On Video

Oil level is one of the important monitoring parameters in power transformers.When oil level is too high or too low,it will affect the normal operation of the transformer.Therefore,it is important to monitor the oil level of the transformer in real time.As the manufacturers did not provide a dedicated device for real-time monitoring of oil level,and can't join the sensor in the substation,so the use of non-contact way to measure oil level has become the current development trend.However,the transformer works for a long time in the outdoors,and different weather will affect the measurement results,so it is very important to use the videobased method to detect the transformer oil level.Based on the video,the image characteristics of transformer under various weather conditions are studied,cloudy,foggy,rainy and high light image are measured the oil level,and the various algorithms are transplanted to the software platform to verify the availability.The main work and innovation points are summarized as follows:(1)The image preprocessing and oil level region extraction and detection principle under normal illumination are analyzed.Several image preprocessing algorithms and oil level region extraction and detection algorithms are compared and analyzed,and finally the optimal algorithm is obtained,which is color space-based method.Firstly,the RGB image is converted into the HSV color space,and then the thresholds are taken in the three channels.Finally,the oil region is extracted by the connected region,and the pixel height is obtained.(2)The method of cloudy image detection is studied.Because of the problem that the color space conversion algorithm is ineffective due to the low image brightness in the cloudy day,the image enhancement method is adopted.The principle of image enhancement is analyzed and some existing enhancement algorithms are compared.At the same time,a histogram equalization algorithm based on parameter correction is proposed.Firstly,the weights of each pixel value are calculated,then the probability distribution function is improved,and the enhanced image is obtained.Finally,the algorithm under normal illumination is used to detect the oil level detection.(3)Aiming at the problem of low image contrast under the condition of fog,the image de fog algorithm is used to dispose the image,so that the oil level can be measured more accurately.The principle and advantages and disadvantages of several existing algorithms are compared in detail,and a method suitable for real-time monitoring of transformers is found,that is,the single image fast de-fog algorithm based on dark prior algorithm.First,the transmittance of each pixel is estimated point by point through the mixed dark channel,and the rough transmission map is obtained.then according to the maximum value of dark channel,the point of the rough transmission map is corrected,so the fine transmission map and enhanced image are obtained.Finally the algorithm of oil level detection is used.(4)Aiming at the problem that the detail information of the rain image is fuzzy,the image rain removal algorithm is studied.The characteristics and principles of the rain removal method are analyzed in detail,and the comparison is made to obtain a better rain removal algorithm based on image stratification.First,the Gauss mixed model stratified on the input image,and then the image is updated with the definition of minimum error.Finally the image without the rain is obtained,and the oil level of the rain image is obtained through the image enhancement algorithm and oil level detection algorithm.(5)The night high light image of the oil level detection method is studied.The characteristics and principles of the high-light algorithm are analyzed and compared,then the improved normalized model is emphasized,and the model is applied to the highlight removal.Firstly,the image is normalized by the improved normalization model,and then the K-means clustering is used to obtain the restored image.Image enhancement and oil level detection.Finally,image enhancement and oil level detection are performed.(6)The rationality and effectiveness of the proposed algorithms are verified.The application software of the all-weather real-time monitoring transformer is designed.At the same time,the algorithm is verified in the application software.Experimental results show that the color space conversion method,the histogram equalization based on the parameter correction,the single image de-fog of the dark primary color,the stratified rain removal algorithm and the high light algorithm can effectively handle the normal light,cloudy,foggy,rain and night high light images under the non-contact measurement.At the same time,they can ensure a certain degree of robustness.It prove that the above algorithm can meet the actual needs of the scene after the algorithm transplanted to the application software,and the software can prevent the adverse effects caused by the mutation level to a certain extent,so as to provide protection for the normal operation of the transformer.