Abstract: Since the tilted meter will cause reading error in the automatic recognition of meter image,a fast tilt correction method based on deep learning for circular pointer meter is proposed,which can realize the tilt correction and rotation correction of the meter image.The key points which is the center of the dial scale numbers are extracted by a convolutional neural network,and the least square method is used to fit the ellipse formed by the key points.The first tilt correction of the meter image is implemented by using perspective transform in combination with the ellipse transformation theory.Then the rotation angle of the meter relative to the horizontal direction is calculated according to a pair of symmetrical key points about the vertical central axis of the meter.The second correction is achieved by rotating the meter image with the geometric center of the fitting ellipse as the rotation center.The image data is collected in the real substation environment to verify the method performance.The experimental results show that this method is more robust than the traditional methods,with a correction efficiency of 100%and an average correction time of 0.45 s,which can meet the requirements of real-time correction.The average relative error of reading of the meter image identified after correction is reduced to 3.99%and the average reference error is reduced to 0.91%,which fully shows the effectiveness of the correction method.