Aiming at the problems of low efficiency of traditional manual reading of water level gauge, poor environmental adaptability and short range of water level gauge obtained by image processing based on water level gauge, a ruler free water level measurement technology based on image recognition is proposed. This technology only needs to calibrate the water level without the help of water level gauge, and input the collected water level image into U-net neural network for training, The image semantic segmentation models of water area and non water area are obtained, and the water level coordinates are obtained by image recognition. Finally, according to the mathematical model of water level calibration, the water level value is calculated by linear interpolation and coordinate fitting. The test results show that the technology has strong environmental adaptability, and the water level measurement errors in day and night are less than 1 cm. This technology can realize the real-time measurement of water level in a large range of rivers and reservoirs.