Abstract:

Feature extraction is one of the most important stages of processing aerial patrol images of transmission lines.The recognition accuracy of fault or target by the traditional methods is not high,single,time-consuming,and easy to be affected by the factors such as background,shape and material,so it is difficult to be applied in practice.In order to solve the defects of the above traditional methods,the deep learning is introduced,a region based full convolution network is adopted.The network is trained with labeled data,and the improved methods such as online difficult sample mining,sample optimization and soft non maximum suppression are used for optimization.The experimental results show that the proposed aspects are faster and more accurate in locating targets.The detection accuracy applied in transmission line patrol inspection is high,so as to meet the needs of transmission line intelligent patrol inspection.

Key words: deep learning, image recognition, region based full convolution network, target detection, transmission line