Abstract: In order to solve the problems of low accuracy of water level recognition and poor migration of detection scene, a high-accuracy and robust method of water gauge reconstruction and water level recognition based on weighted directed graph is proposed. Firstly, the water gauge characters and water level lines on the ship are recognized by the object detection and image segmentation algorithm. Then, according to the structural characteristics of the ship's water gauge, the recognized water gauge characters are constructed into a weighted directed graph, and the ship's water gauge is reconstructed by finding the longest path in the weighted directed graph. Finally, the water level height of the ship is recognized according to the positional relationship between the reconstructed water gauge and the water level line. The test is carried out on the video data collected on the spot in Huanghua port. The result shows that the water level recognition method based on the weighted digraph for water gauge reconstruction can well correct the character error detection problem caused by the water gauge character detection network in the early stage, and greatly reduce the adverse impact caused by the complex background of the ship image. Within the error range of 10 mm, the water level recognition accuracy can reach 91.3%, which is significantly better than the mainstream methods in the industry.