Recognition method and system for power widgets in UAV inspection images based on faster R-CNN

Abstract

The invention discloses a method and system for identifying power components of UAV inspection images based on Faster R-CNN; it includes the following steps: pre-training the ZFnet model, extracting the feature map of UAV inspection images; initializing The obtained RPN area proposal network model is trained, and the area extraction network is obtained, and the area extraction network is used to generate a candidate area frame on the feature map of the image, and the features in the candidate area frame are extracted to extract the position feature and deep feature of the target; The location features, deep features and feature maps of the target are trained on the initialized Faster R-CNN detection network to obtain a power widget detection model; step (4): use the power widget detection model to perform actual power widget recognition and detection . Beneficial effects of the present invention: Using Faster R-CNN to identify and locate various types of electric power components can achieve a recognition speed of nearly 80ms per sheet and an accuracy rate of 92.7%.

Description

technical field [0001] The invention relates to a Faster R-CNN-based method and system for identifying power widgets in unmanned aerial vehicle inspection images. Background technique [0002] In recent years, with the gradual popularization of UAV (Unmanned Aerial Vehicle, UAV) applications, power line inspection drones have received extensive attention from major power grid companies and have been demonstrated and promoted. UAV line inspection has the characteristics of low risk, low cost and flexible operation in field operations; at the same time, it also brings an increase in the workload of line inspection operations in the industry, making massive data require a lot of manual interpretation to obtain the final inspection Report. [0003] At present, power component recognition still stays at the traditional shallow feature-based recognition level, through finely designed shallow features, such as SIFT (Scale-invariant feature transform), edge detector, HOG (Histogram...

AI Technical Summary

Problems solved by technology

[0005] Chinese invention patent (application number: 201510907472.X, patent name: a method for identifying small parts of transmission lines), although this method can realize the identification and positioning of spacers and anti-vibration hammers according to the relationship between wires and small parts, but for The recognition efficiency and recognition effect in complex backgrounds are not good, and cannot meet the needs of the scene

Images