Research On Insulator Anomaly Detection Method Based On Digital Image Processing And Deep Learnin

In recent years,in order to meet the electricity demand of various industries on the fast track of China’s development,it is inevitable to carry out long-distance power transmission,so that the scale of power transmission lines has gradually increased.As an important component in the power transmission system,insulators have the functions of support and insulation.The insulator is in the wild environment,affected by the sun,rain,wind and snow,and in a high-voltage environment.As the service life increases,burst damage often occurs.If the damaged insulator is not replaced in time,it will affect the safety of line operation.Since the insulator is in a high-voltage and high-altitude environment,close-contact detection is difficult and requires a power outage.Therefore,it is of great engineering significance to use the insulator image information collected by the ground non-contact method to detect insulator defects.Therefore,the main contents of the study are as follows:Firstly,the study status of automatic inspection,image processing and image recognition at home and abroad is investigated,and the shortcomings of existing research are summarized.Three aspects of detection location and defect identification are studied.In the image preprocessing stage,the data augmentation method is used to expand the training data set for the problem of insufficient training data set.Aiming at the problem of complex and diverse external environment interference,the image enhancement algorithm and denoising algorithm are studied,and the algorithm of wavelet threshold image denoising is mainly studied,and the adaptive wavelet threshold denoising algorithm is improved.Several classical image enhancement algorithms and denoising algorithms are compared and studied,and the multi-scale Retinex with color recovery,adaptive histogram equalization and improved adaptive wavelet threshold image denoising algorithm are used to predict the insulator image.Processing operations,thereby improving the quality of the image,ensuring the accuracy and validity of the underlying data,and improving the analysis and understanding of the image by the detection model.In the target detection and positioning stage,in view of the problem that the target insulator has a small proportion and the background proportion is large and complex,which will interfere with the segmentation of the target insulator,the Faster R-CNN,SSD and YOLOv3 network models are used to conduct a comparative study of the target insulator positioning.It is concluded that the YOLOv3 insulator positioning network has high accuracy and positioning speed,so the algorithm based on YOLOv3 is used as the insulator positioning network model.Then,the target insulator is trimmed,and then the abnormality detection of the target insulator is performed,the calculation amount is reduced,and the precision and speed of abnormality detection of the insulator are improved.In the defect identification stage,the Alexnet model is used to identify the segmented target insulators,so as to detect abnormal insulators.Experiments show that the cascade network model of YOLOv3 insulator positioning network + clipping layer + Alexnet insulator abnormality detection network is used to detect the abnormality of insulators,identify the defective insulators,and locate the position of the defect,the detection speed is fast,and the average The average accuracy reaches89.81%,which is 8.24% higher than the single-stage detection method.Based on the cascade detection network,an insulator anomaly detection App is designed to realize human-computer interaction.The adopted method can quickly and effectively identify abnormal insulators,effectively eliminate hidden dangers of abnormal transmission line insulators,and has certain engineering application value.