Multi-scale power transmission line defect detection method based on visible light image data

Abstract

The invention discloses a multi-scale power transmission line defect detection method based on visible light image data, and the method comprises the steps: obtaining the visible light image data through a high-definition camera carried by an unmanned aerial vehicle, and obtaining a sample library of power transmission line part defects through employing a manual labeling method or a machine automatic labeling and manual correction method, and obtaining a multi-scale object detection algorithm. According to the method, the phenomenon of drifting in a transfer learning process domain is effectively prevented by using a scale normalization idea; an image pyramid technology and a sliding window cutting technology are adopted in a real-time defect detection stage, so that the contradiction that a neural network model cannot directly detect a high-resolution inspection image is effectively solved; the automation degree and the inspection efficiency of power transmission line inspection are improved, the labor intensity of inspection personnel is reduced, and the operation cost of the whole power grid system is reduced.

Description

technical field [0001] The invention belongs to the technical field of power system operation and maintenance, and in particular relates to a multi-scale transmission line defect detection method based on visible light image data. Background technique [0002] The application of electricity has penetrated into all aspects of our lives, and the power resources are mainly concentrated in inaccessible areas. To fully utilize the power resources, long-distance power transmission is required, so the transmission line plays a very important role in the power system. The transmission line has been exposed to the wild natural environment for a long time, and the specific heat of various components is aging, which seriously threatens the stable operation of the power grid system. Therefore, it is necessary for the power grid maintenance personnel to inspect the line intensively, which greatly increases the operating cost of the power system. With the rapid development of UAV and comp...

AI Technical Summary

Problems solved by technology

[0003] At present, the object detection system based on image data is based on convolutional neural network technology, whether it is a single-step detection method (SSD, YOLO) or a two-step detection method (RCNN, Fast-RCNN, Faster-RCNN) intelligent processing For low-resolution images, in the production environment, it is necessary to convert the captured high-resolution images into low-cost models to adapt to the detection model. After processing, the recognition efficiency of the model is low due to the decrease in image resolution and the amount of information is reduced. actual production requirements

In addition, the currently used pre-training models are almost all trained on the imageNet dataset. The resolution of this dataset is low, while the transmission line inspection images are high-resolution images. If the two datasets are used Domain drift is inevitable in transfer learning

Images