Infrared thermal imager has the characteristics of high precision and non-contact,which has been widely used in power system equipment fault detection. The equipment of detecting thermal image is the basis of automatic detection and diagnosis. For this reason,a real-time detection method based on deep learning is proposed in this paper. A deep convolution neural network is used to predict the coordinates,azimuth and category types of each equipment component. In order to improve the accuracy of prediction results,a priori knowledge of direction consistency between parts is added to the model. In order to evaluate,a large image set including various scenes is constructed. The experimental results show that the method is very robust to noise. When the over union threshold is 0.5,the average accuracy is 93.7 %.