Research On Recognition And Location Of Insulators In Aerial Images Based On SSD

Efficient inspection of transmission line is an important prerequisite to ensure the stable operation of power system.In the traditional inspection of transmission line,manual interpretation of power component defects in aerial images is mainly used,which is time-consuming and inefficient.In order to improve the efficiency of inspection,it is of great significance to use computer to automatically identify the defects in power components.Insulator is an important insulation component in transmission line.The use of artificial intelligence technology to achieve accurate identification and positioning of insulator defects is conducive to the investigation of defective insulators,guide the maintenance personnel to make more accurate and rapid interpretation,and support the safe and reliable operation of transmission line.Firstly,this paper studies and analyzes the manual inspection mode of transmission line,the traditional target detection model,the deep learning target detection framework and the development status of insulator recognition.Secondly,aiming at the problem that it is difficult to identify and locate the insulator corrosion defect target in the aerial image of transmission line,the SSD detection model of deep learning framework is used to detect and identify the insulator corrosion defect target,and through the analysis of the detection results,the SSD target detection model is continuously improved to achieve the efficient identification and location of insulator defects using deep learning network.The main work of this paper is as follows:(1)According to the aerial image insulator pictures,this paper makes a detailed research and analysis,develops the corresponding insulator corrosion data set for the metal corrosion problems around the insulator such as the two end hanging plates of insulator string and ground wire insulator,and analyzes the insulator corrosion target thermal diagram according to the insulator corrosion problems.According to its characteristics,it improves the prediction layer structure of SSD detection framework to enhance the target detection efficiency The accuracy of the network is measured.(2)Aiming at the leakage identification problem of some small target insulator corrosion,this paper proposes the same layer feature fusion method,improves the vgg16 backbone network,reduces the target information loss caused by down sampling,improves the feature capture ability of backbone network,and makes a comparative analysis with the common cross layer feature fusion method.On the basis of SSD detection model,the network structure is further modified,and the spatial attention mechanism is added to improve the utilization of network information,and further improve the detection network recognition accuracy.In this paper,the commonly used evaluation index map in target detection is used to evaluate the results.Based on the original basic SSD model,the map of this model increases by 8.65%,and does not consume too much reasoning time.On the basis of ensuring a certain recognition speed,the recognition accuracy of the detection model is steadily improved.