An insulator string self-destruction identification method based on transfer learning

Abstract

The invention discloses an insulator string self-destruction identification method based on transfer learning, and the method comprises the following steps: S1, carrying out the preprocessing of an aerially photographed insulator string image, and building a training database; S2, detecting the position of an insulator string in the image through a target detection network YOLOv3, and extracting afeature square frame; S3, carrying out secondary processing on the extracted Multi-Size insulator string feature image; S4, carrying out transfer learning on a deep residual network ResNet-50, and performing self-explosion identification on the insulator; and S5, generating an insulator string state report document. According to the invention, based on the defects in the prior art, transfer learning processing image is introduced, deep training can be completed under the condition that an insulator data set is very small, multiple advanced deep learning models are adopted, feature extractionand self-destruction recognition can be effectively carried out on an insulator string, a visual report document is generated after a recognition task is completed, and result statistics and manual checking are facilitated.

Description

technical field [0001] The invention relates to the fields of deep learning image processing and electric defect recognition, and specifically relates to a method for self-explosion recognition of insulator strings based on transfer learning. Background technique [0002] Insulators are indispensable equipment in high-voltage transmission systems. They are responsible for electrical insulation and conductor connections. Damage to insulators due to self-explosion is one of the main causes of transmission system accidents. In order to ensure the reliability of the power transmission system, regular manual inspection of insulators is an essential maintenance procedure. The traditional method is for professionals to conduct inspections under UHV conditions, and there are great risks. With the development of deep learning in the field of computer vision, some deep learning algorithms have been applied to the state recognition of insulators. [0003] Hou Chunping, Zhang Hengguang...

AI Technical Summary

Problems solved by technology

[0003] Hou Chunping, Zhang Hengguang et al. "Recognition method for self-explosion defects of transmission line insulators" (Journal of Electric Power System and Automation, 2018) applied deep learning target classification algorithms Alex Net, VGG16 and Faster R-CNN framework to train classifiers and detectors respectively , and cascaded to form a cascaded network to detect and recognize insulator targets, but this method has a huge demand for data sets, and it takes a lot of time to train a deep network from scratch

Gao Qiang, Yang Wu et al. "Insulator Fault Identification Algorithm Based on Sparse Difference Deep Belief Network" (Electrical Measurement and Instrumentation, 2018) proposed a new image classification method based on sparse difference deep belief network, referred to as the D-DBN method, and It is applied in the identification of insulator faults, but the depth model used is relatively simple

The patent document with the publication number CN106290388A discloses a method for automatic detection of insulator faults. The collected images are segmented using the image salient region detection algorithm, and the image recognition technology is used to identify the state of the insulator. However, due to the identification of many types of damage, the accuracy is low high

Images