Abstract: The early fault detection of anti-bird thorns on electrical towers is of great significance for reducing the occurrence of bird-damages and ensuring the safe and reliable operation of the transmission lines. The anti-bird thorns in the electrical inspection images have the features of being unnoticeable in contour and partially overlapped in distribution, which poses challenges to the research of anti-bird thorn identification and fault detection. Aiming at the characteristics of the anti-bird thorns, we propose a component identification and fault detection method based on deep convolution neural network. First, an electrical inspection image is sharpened by the sharpening filter. Then, the region of an anti-bird thorn that is processed by the sharpening, is bounded and cropped by the object detection network YOLOv3 which is trained with multi-scaling. Finally, the anti-bird thorn fault detector based on the feature extraction network Resnet152 is utilized to process the cropped area of the anti-bird thorn, realizing the fault detection. The proposed method is tested on the electrical inspection images of the validation dataset for component identification and fault detection of the anti-bird thorn with the average precision of 95.36% and 92.3% for the component identification and the fault detection respectively. The experimental results show that the proposed method can effectively realize the component identification and fault detection of the anti-bird thorns in electrical inspection images.