Design And Implementation Of Intelligent Inspection System For Secondary Equipment Of Substation

The construction and popularization of the national smart grid is inseparable from the safety and reliability of the substation system.There are the situation of missed inspection and false inspection in the early manual inspection and the remote message supervision inspection in the form of human-computer interaction.Therefore,engineering personnel have developed a series of intelligent inspection systems for secondary device with robot patrol mode as the core technology.However,this method of robot patrol cannot obtain images of multiple devices at the same time,it is hard to provide accurate detection results in combination with multi-screen cabinet information,which is not conducive to the full coverage of safe operation and maintenance of substations.Based on the above problems,this paper designs and implements a multi-camera-based secondary device inspection system for substations,completes the function of regularly collecting images,locating and recognizing device-unit,and finally giving alarm details.In the image acquisition module,a parallel collection unit is designed to acquire the secondary screen cabinet images at the same time,which provides an accurate detection method for the subsequent multi-screen cabinet judgment.In the image location module,in order to meet the safety requirements of maintenance,the camera needs to be lifted and lowered.The algorithm needs to adaptively adjust the effective area,determine whether the image is offset,and provide high-quality location results for the collected images at different location with slight changes.In order to avoid the problem of image alignment failure caused by slight disturbance caused by translation,rotation,illumination change,and similar textures,the image alignment algorithm based on SIFT feature points is improved,and a priori coordinates represented by the white text box anchor are introduced.The search range of SIFT feature points is constrained by geometric space to improve the alignment accuracy.In the image recognition module,the data categories are subdivided to reduce the interference of different data types on the final recognition accuracy;for a series of problems such as the low prediction accuracy of the continuous digital model and the inconsistent number of digits,this paper adopts the scheme of segmentation at first and then predicting a single digit,which effectively improving the accuracy of digital recognition.According to the long-term online operation of the system and the inspection of the system accuracy,the current location accuracy of indicator lights,pressure plates and digits has reached more than 99%,and the status recognition accuracy has reached more than 98%.