Research On Infrared Detection Of Power Equipment Based On Computer Vision

Power equipment is the basic component of the power grid.High-frequency inspection or real-time monitoring of power equipment by using infrared detection technology will be helpful for the preventive maintenance of power equipment in order to improve the stability and reliability of large power grids.However,the infrared detection of the existing intelligent inspection equipment has the disadvantage of poor imaging quality in special inspection environment;meanwhile,the judgment methods such as artificially setting temperature thresholds,human eye observations,image processing,and other judgment methods are also faced with the problem of low intelligence,which is difficult to meet the demand of future intelligent inspection equipment for mass infrared data processing.Therefore,aiming at the problems existing in infrared inspection of electric power equipment,the research on infrared inspection of electric equipment based on computer vision is carried out in this thesis.Firstly,the basic principle and device of infrared imaging are introduced.The main fault types and mechanisms of infrared detection of power equipment are analyzed,and the specific fault phenomena and causes of infrared detection for 13 types of power equipment are discussed in detail,the visual detection algorithm based on deep convolution network is discussed also.Then,in view of the complex environment which the power equipment infrared detection process may encounter,simulate multi-noise,low-brightness and low-contrast imaging in complex environments,and propose an integrated filtering scheme combining traditional filter and fast-guided filter for image denoising,and propose an improved MSRCP method for image enhancement.Secondly,the data set of abnormal area detection,abnormal area and equipment detection for infrared detection of power equipment is established.The infrared detection network of Faster R-CNN power equipment based on TensorFlow deep learning framework is built.The abnormal region,abnormal region and equipment detection model of power equipment infrared detection are trained.Through model verification and sample effect test,the results show that the method has a high detection accuracy,but the detection speed is low,can not real-time detection.Finally,for the infrared image detection method based on Faster R-CNN,the detection speed is low which is difficult to detect in real time and difficult to apply the application on the actual intelligent inspection equipment.The SSD detection network based on the lightweight network MobileNets is presented.The abnormal area,abnormal area and equipment detection dual model of equipment infrared detection are trained.Through model verification and sample effect test,the results show that the proposed method has good detection accuracy and can achieve real-time detection effect.In this thesis,the infrared image denoising enhancement method has good image enhancement effect,and the Faster R-CNN detection method has higher detection accuracy but lower detection speed,while the SSD detection method has better detection accuracy and can detect in real time,which will lay a foundation for realizing the "Intellisense" of infrared detection of existing power equipment.