Abstract: Aiming at the low accuracy of water gauge identification and water level detection in complex environment, a small area guided water gauge water level detection method is proposed. Firstly, the improved YOLOv4 algorithm is used to locate the "E" character on the water gauge to segment the small area. Secondly, the improved DeepLabv3+ algorithm is used to segment the water gauge in a small area. Thirdly, Canny algorithm is used to extract the water surface edge line. Then, the water level is calculated by statistical pixel coordinates. Finally, the water level is calculated by linear interpolation. The experimental results show that the mean pixel accuracy of the improved yolov4 is 94.13%, which is higher than that of other target detection networks. The mean intersection over Union of the improved DeepLabv3+ is 82.17%, which is higher than that of other segmentation networks. The average pixel of the water level line under the guidance of a small area is 3.030, which is lower than the direct segmentation. The comparison error with the manual reading is less than 1 cm, which meets the requirements of hydrological detection specifications. Compared with traditional image detection methods, the measurement accuracy is higher.