The invention discloses a water gauge reading method based on
deep learning. The method comprises the following steps: acquiring the height information of a water gauge and the target height information of an E character; obtaining a target image, and recognizing the position of a water gauge and the position of an
angular point to correct the target image; carrying out instance segmentation on an E-character target in the corrected image to obtain a recognition frame coordinate and a segmentation result of the E-character target; sorting the E-character targets from top to bottom according to the coordinates of the recognition frame, respectively calculating
y coordinate value giant change times of horizontal projections of the E-character targets in the vertical direction according to segmentation results, and judging whether the E-character targets are complete or not according to the giant change times; if the updated coordinates of the identification frame exceed the position of the water gauge, the E-
word target is judged to be an incomplete E-
word target connected with the water surface, and the effective number of the E-
word target is calculated; and calculating the current
water level according to the sum of the copies of all the E-shaped targets, the height of the water gauge and the height of the E-shaped targets. According to the invention, human errors can be avoided, the influence of environmental factors on
water level measurement can be overcome, and the accuracy and reliability of
water level detection are greatly improved.