Lin, F.; Yu, Z.; Jin, Q., and You, A., 2020. Semantic segmentation and scale recognition–based water-level monitoring algorithm. In: Hu, C. and Cai, M. (eds.), Geo-informatics and Oceanography. Journal of Coastal Research, Special Issue No. 105, pp. 185–189. Coconut Creek (Florida), ISSN 0749-0208.
Many water-level monitoring places are equipped with standard water gauges and cameras. Measuring the water level by image processing can work automatically, much better than manual observation. Water-level monitoring sites such as river channels and irrigation canals are usually in fields, which makes it difficult to recognize the water gauge. This paper imitates the human behavior of reading the scale of the water gauge and proposes a water-gauge image monitoring algorithm based on deep learning. A semantic segmentation network DeeplabV3+ is used to locate the water gauge in the surveillance image and to crop it. Then, the water-gauge image is divided into multiple regions by binaryzation and K-means clustering algorithm, and each region is recognized by convolutional neural network based on VGG-8. Finally, the water level is calculated according to the results of segmentation and recognition. In the experiment, the accuracy of the water level measured by this method achieves the minimum error of the water gauge. This algorithm has high cost performance.