Insulator Location And Fault Detection Based On Deep Learning

The working state of the insulator is of great significance to the normal operation of the transmission line,but the insulator has been working outdoors for a long time,the environment is complex,and it is prone to failures such as damage and missing pieces.It may cause a safety accident if it is not detected and replaced in time.Therefore,it is necessary to conduct safety inspection on the transmission line in time.Traditional manual inspection is a waste of labor force and is affected by personal subjective consciousness.The detection efficiency is not high and it is easy to miss the inspection.This paper uses image processing related algorithms to perform a series of processing on the insulator images collected with complex backgrounds,and then classifies the working status of insulators into two categories: intact and broken.This paper realizes the intelligent identification and detection of insulator failure.Firstly,the image samples are?preprocessed.The pre-processing can remove the noise in the image and solve the problems such as low contrast of the image and unclear details information caused by environmental lighting and other factors.Including filter denoising and image enhancement processing based on Retinex image clarity.Secondly,the SURF algorithm is used before segmenting the target insulator region.An insulator sample library was established to match and identify the image set,and the image set containing the insulator was determined.By comparison with traditional image segmentation algorithms(OSTU,K-means clustering algorithm,etc.),experiments have verified that the U-net model based on deep convolutional network used in this paper has a high accuracy rate for insulator image segmentation,and the Edge information has also been preserved to a large extent.This method solves the problem that most traditional pseudo segmentation exists in traditional segmentation.Finally,in the classification of insulator images,convolutional networks require a large number of labeled samples to train the network,otherwise overfitting problems are prone to occur.The traditional capsule network is shallow and cannot extract deep abstract feature information.This paper proposes an improved capsule network model based on the capsule network to classify and recognize insulator images.The 13-layer convolution structure composed of 3*3 small convolution kernels deepens the network depth,which is more conducive to learning abstract features,while also retaining the advantages of traditional capsule model inputs as vectors.The output is retained by the dynamic routing algorithm.The orientation and angle information can be used to classify and identify the insulator damage more accurately.The experimental comparison verifies that the recognition rate of the improved capsule network model in small training samples is higher than that of ordinary convolutional networks and traditional capsule networks.And classification accuracy and loss function convergence are better than traditional capsule network models.To some extent,it solves the problem of low manual detection efficiency and easy error detection.