Deep Learning-Based System for Automatic Recognition and Diagnosis of Electrical Insulator Strings

Abstract

This paper presents a complete system for automatic recognition and the diagnosis of electrical insulator strings which efficiently combines different deep learning-based components to build a versatile solution to the automation problem of the power line inspection process. To this aim, the proposed system integrates one component responsible for insulator string segmentation and two components in charge of its diagnosis. The insulator string segmentation component consists of a novel fully convolutional network (FCN) architecture, termed Up-Net, which enhances the capabilities of the state-of-the-art U-Net network by introducing new skip connections at certain levels of the architecture. Furthermore, we propose a second variant of the Up-Net network by training it within a generative adversarial network (GAN) framework. The capabilities of the proposed Up-Net variants are incremented by the application of data augmentation and transfer learning techniques, achieving accurate segmentation of the insulator string elements (i.e., discs and caps). Regarding the insulator string diagnosis, we design a convolutional neural network (CNN) which takes as input the mask generated by the insulator string segmentation component and is capable of identifying the absence of a variable number of discs. The second diagnosis component consists of a novel strategy which integrates a Siamese convolutional neural network (SCNN) designed for modeling the similarity between adjacent discs and allowing the detection of several types of disc defects using the same model. The proposed system has been extensively evaluated in several video sequences from real aerial inspections of high-voltage insulators, showing robust insulator recognition and diagnosis capabilities.

Insulator Defect Recognition Based on Global Detection and Local Segmentation

“…Pix2pix, with Condition Generative Adversarial Networks (CGAN) [37] module which was also highlighted in [38] and [22], however, has the worst recognition effect in our dataset. FCN-8s makes good segmentation for large insulators, but does not perform well for small insulators.…”

“…In [16]- [18], Faster R-CNN was adopted to find the insulator from the original images, and then the local insulator map is cut out according to the coordinates for further semantic segmentation, among them, [16] used U-Net [19] to segment the pixels of the insulator self-detonating part, [17] and [18] employed FCN-8s [20] to segment the insulator string, after that, the former relied on a priori knowledge of the insulator self-detonation characteristics to determine the location of the self-detonation, the later fed the insulator string mask into GoogLeNet [21] to classify normal and defect. the method of [22] was similar to it, but they omitted the procedure for detecting insulators, using a U-Net, which added a new skip connection to directly segment the cap and disk in the original images and sent the mask to the CNN classification network for defect identification. However, all the insulators in the results they gave were single strings, and they are not diverse in perspective.…”

“…In the result mask, the normal part and the defective part of the insulator string are shown in green and red respectively. According to the outline of the defective part, defect can be marked in the original image, compare with [22], We don't need an additional classification network to handle the mask to discriminate defect, and the defect location identified by semantic segmentation can be visually displayed.…”

Novel eye‐blink artefact detection algorithm from raw EEG signals using FCN‐based semantic segmentation method

“…In the last step, these three labelled areas were collected and labelled image areas ( P L ) were obtained by the following equation: The number of samples in the training set directly affects the performance of the network in the training of deep learning algorithms. A large number of samples provide better training for the network [33 ]. Data augmentation was applied to raw data to increase success.…”

Deep Learning for Fault Diagnostics in Bearings, Insulators, PV Panels, Power Lines, and Electric Vehicle Applications—The State-of-the-Art Approaches

“…28. [72] The surface fault of insulator IULBP NP IULBP+Rules Texture [73] Missing-cap of insulator GLCM CGT-LBP-HSV GLCM+Rules Texture [74] The surface fault of insulator GSS-GSO GrabCut Rules Shape [75] Missing-cap of insulator Up-Net+CNN Up-Net CNN Deep [76] The surface fault of insulator M-SA F-PISA Colour model Colour [77] Missing-cap of insulator SMF Colour model Morphology Fusion [78] The surface fault of insulator CGL-EGL CGL EGL Shape [79] The surface fault of insulator M-PDF OAD-BSPK AlexNet Deep [80] Missing…”