Method and system for extracting target image from power transmission line inspection image

Abstract

The invention discloses a method and system for extracting a target image from a power transmission line inspection image, and belongs to the technical field of power transmission line operation and maintenance. The method comprises the following steps: making an initial sample set by using a B / S architecture, and storing the initial sample set; generating a sample set; and determining main information of the target image, retrieving the sample set through the main information, and outputting a target image file. Label information can be queried and combined as required according to an application environment, a training sample set required by a mainstream deep learning detection network can be output as required, positive and negative samples of a certain type of equipment can also be output at the same time, an adversarial network sample set is constructed, and the training efficiency and recognition effect of a deep learning model are improved. The accurate and effective diagnosis of common faults of the power equipment in the inspection image is realized.

Description

technical field [0001] The present invention relates to the technical field of transmission line operation and maintenance, and more specifically, to a method and system for extracting target images from transmission line inspection images. Background technique [0002] Different types of power inspection terminals such as drones, helicopters, and visual surveillance cameras have obtained a large amount of inspection image data. At present, inspection personnel need to use manual review to identify defects in power components in inspection images, which has disadvantages such as heavy workload, low inspection efficiency, and misjudgment or misjudgment. [0003] In order to improve the inspection efficiency of inspectors, a power equipment defect recognition technology based on deep learning has emerged. However, its recognition accuracy depends on the scale of the labeled sample data set, the quality of labeling, and the granularity of label classification. By clarifying th...

AI Technical Summary

Problems solved by technology

In addition, for inspection drones, there are still unstructured information such as shooting lighting conditions, shooting distance, orientation, and camera focal length that have not been effectively recorded.

Obviously, the existing labeling method with a single label cannot meet the actual labeling requirements of inspection image power equipment with a hierarchical structure, and is also not conducive to the storage, management, retrieval of subsequent labeling data and the production of deep learning model training sets.

Images