Electric power inspection image target detection method with self-learning capability

Abstract

The invention relates to an electric power inspection image target detection method with a self-learning capability. The electric power inspection image target detection method comprises the steps: sending a picture sample containing a target object and a target object position labeling file into a Context CNN, and training a Context model; performing image fusion processing on the input background picture and the expanded target object picture based on a Context model to obtain an expanded picture sample and a target object position labeling file; training a target detection model 1 and a target detection model 2 based on the picture sample and the position labeling file which are sent into the Faster R-CNN; and outputting a detection result indicating whether the target object is contained in the collected picture sample based on the target detection model 1 and / or the target detection model 2 by the Faster R-CNN. The electric power inspection image target detection method has the beneficial effects of overcoming the condition of insufficient training data set existing in the image recognition technology applied to the power industry and automatically improving the recognition performance in the use process.

Description

【Technical field】 [0001] The invention relates to the technical field of image recognition, in particular to a power inspection image target detection method with self-learning capability. 【Background technique】 [0002] In the computer field, image target detection technology based on machine learning and deep learning is an important part of artificial intelligence. The research goal of image target detection technology is to identify the position and category of specific targets in the collected images and make meaningful judgments. [0003] In the power industry, the use of image target detection technology for foreign object intrusion detection in places such as transmission lines and substations is gradually becoming one of the important means for the realization of smart grids. However, in the application of image recognition technology in the power industry, there is a shortage of training data sets (such as pictures of garbage bags hanging on transmission lines), w...

AI Technical Summary

Problems solved by technology

However, in the application of image recognition technology in the power industry, there are insufficient training data sets (such as pictures with garbage bags hanging on transmission lines), which will affect the accuracy of the image target recognition model

Fusing the target object with the scene background is a way to expand the sample, but randomly combining the target object with the background (such as pasting a crane in the sky) will degrade the image recognition performance

Images