The invention discloses a water gauge image water level automatic reading method based on a Mask RCNN algorithm in the technical field of water conservancy informatization water level automatic measuring and reporting, and the method comprises the following steps: (1), installing a camera and a water gauge, recording and confirming four corner point coordinates of each preset point of a dome camera after water gauge imaging, and setting the four corner point coordinates as preset point parameters; (2) acquiring water gauge images of different cameras and correcting the water gauge images in batches; (3) marking a water gauge detection frame for the corrected image, and constructing a training set; (4) obtaining the optimal anchor frame size anchors, adjusting the learning rate and the number of iterations, and starting to train, verify and test the Mask RCNN model to obtain a final model parameter file; (5) deploying a model parameter file trained in the step (4), correcting a picture collected in real time, and predicting a detection box and a segmentation mask; and (6) calculating the water level in combination with the coordinates of the preset point in the step (1); the method can adapt to water gauge image water level automatic reading in a complex environment, and has the advantages of being high in recognition accuracy, wide in application range, high in intelligent degree and the like.