With the continuous promotion of object detection technology in electric power inspection tasks, automatic analysis of various images collected by electric power inspection has become one of the current hot research directions in electric power enterprises. Traditional target detection methods are mostly based on machine learning, and the detection accuracy in complex scenes needs to be further improved. The image-based deep learning method is widely used in power inspection target detection due to its ideal detection accuracy and environmental adaptability. Aiming at the problems of poor image quality, complex background, poor contrast, etc. collected in complex scenes of electric power inspection, an infrared image object detection method based on regional convolutional neural network fused with histogram of image orientation gradients(HOG-RCNN) was proposed. Before the image enters the RCNN, HOG feature extraction was performed on the input image for helping RCNN to select candidate regions. Algorithm experiments show that the detection effect of the method proposed is better than that of a separate RCNN network.