Crowdsourced Label Truth Inference Algorithm for Power Line Component

The technical foundation of intelligent inspection of power lines is the detection of components in aerial images. With the rapid development of deep learning, new object detection technologies are constantly emerging. However, accurate annotation of data sets is equally crucial for improving detection performance. This study proposes a truth inference algorithm for transmission line components to improve the accuracy and reliability of annotated data. Propose Multi component label removal algorithm and Large box label removal algorithm for both normal sized and small-sized components, respectively. The experimental results show that our method has achieved significant results in improving the quality and accuracy of annotated data. This study is of great significance for improving the accuracy of annotation on power line datasets.