Research On Insulator Fault Detection Method Based On Aerial Image

High-voltage transmission lines are exposed to the wild for a long time,and are prone to failure due to the influence of natural weather,material aging and human factors,among which insulator failure is the most common failure in transmission lines.According to statistics,insulator failure can reach more than 50% in power system failures.Manual inspection is a widely used detection method for transmission line inspection at present.However,because high-voltage transmission lines work in harsh environments such as mountains,rivers,forests and grasslands,the risk factor of manual inspection is high and a lot of manpower and material resources will be consumed.Compared with manual inspection,object detection technology isn’t affected by human and external environmental factors,and can achieve accurate and efficient detection.Due to the interference information such as illumination,shooting angle,complex background,etc.in aerial images of insulators,it’s time-consuming and laborious to use manual detection completely,and the object detection method has certain adaptability.In recent years,it has been focused by researchers and has been widely used.Insulator fault detection mainly includes two research methods: traditional object detection method based on artificial design features combined with machine learning classifier and object detection method based on deep learning model.In this paper,insulator faults are detected based on artificial features and machine learning classifier.However,artificial design features require researchers to have rich prior knowledge,and are only suitable for scenes with small background interference,specific size and specific lighting conditions,and unable to accurately identify insulator faults in complex aerial images.Therefore,aiming at the problem of low detection accuracy in complex background environment in aerial images of insulators,the research and system implementation of object detection methods based on traditional machine learning model and deep learning model are carried out.The main research contents are as follows:(1)Aiming at the target detection method of artificial design features combined with machine learning classifier,an insulator fault detection algorithm based on SURF-HSG features combined with multi-spatial information fusion is proposed.By extracting the feature information of insulator positive samples and background negative samples,67-bit SURF-HSG feature vectors are generated for training SVM classifier.The insulator ROI region is extracted by SVM classifier,and the insulator target is segmented by fusing the feature information in multiple color spaces in the insulator ROI region,and the fault is detected by image processing algorithm.(2)In order to improve the detection accuracy and recall rate of insulator faults,SSD model based on multi-scale feature fusion(PL-MFSSD)is proposed.The depth separable convolution operation is used to replace the traditional convolution operation to generate the object prediction feature layer,so that the model maintains high detection efficiency;Combining the structural information of the bottom feature layer with the semantic information of the top feature layer through feature layer channel fusion;Modify the size and width-height ratio of anchor frame according to the scale characteristics of insulators and faults;The residual module is added in each object prediction layer to alleviate the gradient disappearance of the model,and the model significantly improves the recall of fault objects.(3)In order to further improve the detection efficiency of the model,a lightweight feature fusion SSD(Mobile Netv2-MFSSD)model is proposed based on PL-MFSSD model.The lightweight network Mobile Netv2 is used instead of the feature extraction backbone network VGG16 of PL-MFSSD model,which improves the detection efficiency of the model;To improve the ability of extracting object feature information,the object prediction layer structure of SSD model is maintained in the object prediction layer of Mobile Netv2-MFSSD model.The Mobile Netv2-MFSSD model significantly improves the detection efficiency on the basis of slightly losing the object accuracy.(4)A fault detection system of insulator aerial image based on Linux platform is designed.The system includes five modules: system introduction module,system login module,single insulator image detection module,multiple insulator image continuous detection module and test result history query module,which has certain practicability and can basically meet the whole operation process of insulator fault detection.